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.1003707 | B Cells Enhance Antigen-Specific CD4 T Cell Priming and Prevent Bacteria Dissemination following Chlamydia muridarum Genital Tract Infection | B cells can contribute to acquired immunity against intracellular bacteria, but do not usually participate in primary clearance. Here, we examined the endogenous CD4 T cell response to genital infection with Chlamydia muridarum using MHC class-II tetramers. Chlamydia-specific CD4 T cells expanded rapidly and persisted as a stable memory pool for several months after infection. While most lymph node Chlamydia-specific CD4 T cells expressed T-bet, a small percentage co-expressed Foxp3, and RORγt-expressing T cells were enriched within the reproductive tract. Local Chlamydia-specific CD4 T cell priming was markedly reduced in mice lacking B cells, and bacteria were able to disseminate to the peritoneal cavity, initiating a cellular infiltrate and ascites. However, bacterial dissemination also coincided with elevated systemic Chlamydia-specific CD4 T cell responses and resolution of primary infection. Together, these data reveal heterogeneity in pathogen-specific CD4 T cell responses within the genital tract and an unexpected requirement for B cells in regulating local T cell activation and bacterial dissemination during genital infection.
| Sexually transmitted infections caused by Chlamydia are increasing every year in the US and an effective vaccine is urgently required. Unfortunately, we currently only have a rudimentary understanding of the natural host immune response to Chlamydia infection, especially in the context of the female genital tract. Here, we have developed new reagents that allow direct visualization of the host T cell responses to vaginal Chlamydia infection using a mouse model of infection. These new tools reveal an unexpectedly complex CD4 T cell response to infection and a surprising role for B cells in preventing the spread of bacteria to multiple host tissues. This greater understanding of the host response to infection may eventually allow the construction of an effective vaccine.
| Chlamydia trachomatis is an obligate intracellular pathogen that causes the most prevalent bacterial sexual transmitted infection worldwide [1]. In the US, Chlamydia is now the most common notifiable disease reported to the US Centers for Disease Control (CDC). The 1.4 million cases of Chlamydia infection reported in 2011 represent an 8% increase over the previous year and is the largest number of annual infections ever reported to the CDC for any condition [2]. The introduction of a Chlamydia screening and control program in the mid-1990s has not prevented annual increases in infection, although a portion of this increase is due to improved disease surveillance [3]. Overall, the CDC reports a median 8.3% Chlamydia positivity test among women aged 15–24, making this one of the most prevalent bacterial infections in the US.
Most Chlamydia infections are initially asymptomatic and therefore unlikely to be treated. However, 5–15% of females with untreated infection will eventually develop serious pelvic inflammatory disease (PID) as a consequence. Furthermore, 1 in 6 women who develop PID will become infertile, and many others will develop chronic pelvic inflammation and pain, or suffer from ectopic pregnancy [4]–[6]. The combination of an extraordinarily high number of infections, the asymptomatic nature of initial disease, and the potential for serious reproductive pathology in young women, means that Chlamydia is now recognized as a growing health care problem in the US. The current consensus among scientists and clinicians is that an effective Chlamydia vaccine is urgently needed [7].
The development of an effective Chlamydia vaccine would likely alleviate the burden of Chlamydia on the public health care system. However, the rational design of a Chlamydia vaccine would be aided by improved understanding of the cellular immune response to infection of the female reproductive tract. As Chlamydia is an obligate intracellular pathogen, IFN-γ production by CD4 Th1 cells is essential for protective immunity to primary and secondary infection [8]–[13]. Unfortunately, we have at present only a rudimentary understanding of the development of protective Th1 responses in the context of the female upper reproductive tract and the extent of T helper heterogeneity is unclear. One of the major roadblocks to improving this situation is the lack of antigen-specific reagents that would allow detailed investigation of Chlamydia-specific CD4 T cell responses using a relevant genital challenge model.
Previous studies have demonstrated that antibody production by B cells can assist protective immunity during secondary Chlamydia infection [14]–[16]. In contrast, B cells are thought to be dispensable for resolving primary Chlamydia infection, and B cell-deficient and wild type mice shed similar numbers of Chlamydia, as measured by vaginal swabs [17], [18]. However, another study using the respiratory route of infection demonstrated that intranasal infection with Chlamydia requires B cells for efficient CD4 T cell activation [19]. Therefore, the issue of whether B cells contribute to initial CD4 T cell priming during vaginal infection requires additional analysis.
In this study, we generated MHC class-II tetramers to visualize the endogenous CD4 T cell response to systemic and genital tract Chlamydia infection. We show that, unlike intravenous infection, reproductive tract infection is associated with a short delay in the clonal expansion of Chlamydia-specific CD4 T cells in the local draining lymph node. While almost all expanded CD4 T cells expressed the Th1 marker, T-bet, we detected an expanded pool of Chlamydia-specific Tregs that co-expressed T-bet and Foxp3, and a population of Chlamydia-specific Th17 cells that were specifically enriched in the reproductive tract. In addition, we noted a surprising requirement for B cells in Chlamydia-specific CD4 cell priming within local draining lymph nodes. Loss of local priming in the absence of B cells coincided with bacterial dissemination to the peritoneal cavity inducing inflammatory infiltrate and ascites. Together, these data demonstrate heterogeneity in Chlamydia-specific T helper responses and an unexpected role for B cells in local CD4 T cell priming and bacterial containment within the upper reproductive tract.
In order to develop an overview of the adaptive response to Chlamydia infection, we initially examined the kinetics of bacterial growth and Chlamydia-specific CD4 T cell expansion during systemic infection with Chlamydia. When C57BL/6 female mice were infected intravenously (i.v.) with 1×105 inclusion-forming units (IFUs) of C. muridarum, the bacterial burden in the spleen peaked around day 4 post-infection and decreased quickly thereafter (Fig. 1A). At day 20, no infectious Chlamydia was detected in the spleen (Fig. 1A). Consistent with previous findings [20], a small number of Chlamydia were found in the lung during the first week of systemic infection, but no bacteria were detected in kidney or heart at any time point (data not shown).
Numerous C. muridarum MHC class-II epitopes have been uncovered by Immunoproteomic analysis of infected APCs [21]. We used an ELISPOT assay to monitor the frequency of CD4 T cells responding to multiple C. muridarum epitopes after systemic infection. A population of IFN-γ-secreting CD4 T cells responding to RplF51–59, Aasf24–32, and PmpG-1303–311 was detected as early as 4 days after infection (Fig. 1B and C). Expansion of IFN-γ-secreting CD4 T cells peaked around day 4–7, and was followed by a slow contraction of the population over the next 90 days, before a plateau was reached that lasted for at least 352 days (Fig. 1B and 1D). Thus, peak expansion of IFN-γ-secreting CD4 cells closely mirrored peak bacterial burdens in vivo, and stable Chlamydia-specific CD4 T cell memory frequencies were maintained in the absence of active Chlamydia infection.
Previous studies have demonstrated that pMHC class-II tetramers can be used in conjunction with tetramer enrichment, to visualize low frequency endogenous antigen-specific CD4 T cells in infected and immunized mice [22], [23]. We constructed three distinct pMHC class-II tetramers, containing I-Ab with a Chlamydia-specific epitope (RplF51–59, Aasf24–32, or PmpG-1303–311) bound to the MHC class-II β chain. Uninfected C57BL/6 mice contained low frequency antigen-specific CD4 T cell population specific for each Chlamydia epitope (Fig. 2A). However, in mice immunized subcutaneously with peptide/CFA, or infected intravenously with C. muridarum, an expanded population of CD44hi CD4 T cells was detectable 7 days post immunization or infection (Fig. 2A). Tetramer staining was specific, as infection with Salmonella Typhimurium did not induce expansion of tetramer-specific CD4 T cells (Fig. 2A). Furthermore, no CD8 T cells were detected that bound to Chlamydia tetramers (Fig. 2A) Together, these results demonstrate that all three tetramers, RplF51–59:I-Ab, Aasf24–32:I-Ab, and PmpG-1303–311:I-Ab can be used to detect endogenous C. muridarum-CD4 T cells in vivo.
We next used the PmpG-1303–311:I-Ab tetramer to visualize clonal expansion of antigen-specific CD4 T cells during intravaginal (i.vag.) infection. We focused on PmpG-1 because CD4 T cell clonal expansion against RplF, Aasf and PmpG-1 is similar (Fig. 1B, 1C and 1D) yet PmpG-1 is a promising Chlamydia vaccine candidate [24]. Consistent with previous findings, C. muridarum bacterial loads measured by vaginal swab peaked around day 4 post-infection and steadily decreased until clearance around day 35 (Fig. 2B). To visualize the primary site of endogenous T cell priming to Chlamydia infection, we examined PmpG-1-specific T cell activation in multiple secondary lymphoid tissues. One week after infection, PmpG-1-specific CD4 T cells expanded in iliac lymph nodes and spleen, but were barely detectable in other non-draining lymph nodes (Fig. 2C and 2D). At day 14, endogenous PmpG-1-specific CD4 T cell expansion peaked in all secondary lymphoid tissues, and was followed by a notable contraction phase (Fig. 2D). The kinetics of the CD4 T cell response to local Chlamydia genital tract infection was therefore markedly delayed in comparison to systemic infection with the same pathogen (Fig. 1).
To examine CD4 T helper differentiation, the expression of lineage-specific transcription factors was examined in expanded CD44hi PmpG-1303–311:I-Ab+ CD4 T cells. Following either systemic or intravaginal infection, almost all PmpG-1-specific CD4 T cells expressed T-bet while no GATA3 expression was detected (Fig. 3A). This is consistent with previous reports that Th1 CD4 T cells are the dominant helper subset following Chlamydia infection [10], [11]. However, a distinct population of PmpG-1-specific CD4 T cells that co-expressed Foxp3 and T-bet was also detected after i.v. and i.vag. infection (Fig. 3B), suggesting that induced Chlamydia-specific Treg cells are also contained within the expanded CD4 pool. We also utilized RORγt-GFP reporter mice and combined staining with all three Chlamydia tetramers to examine the potential development of Chlamydia-specific Th17 cells after vaginal infection. While GFP-positive cells were undetectable among expanded tetramer-positive cells in the spleen or draining lymph nodes of infected mice, approximately 7% of CD4+CD44hitetramer+ T cells in infected non-lymphoid tissues expressed GFP (Fig. 3C). Furthermore, stimulation of lymphocytes purified from the genital tract confirmed the presence of T cells producing both IL-17A and IFN-γ (Fig. 3D). Thus, Chlamydia infection of the reproductive tract induces a heterogenous T helper response that comprises expanded T-bet+ Th1 cells, T-bet+Foxp3+ Tregs, and Th17 cells that are enriched in infected non-lymphoid tissues.
Next, we examined bacterial shedding after vaginal infection of WT and μMT mice with Chlamydia. Consistent with previous reports [17], [18], bacterial shedding was unaffected by the absence of B cells (Fig. 4A). However, more detailed analysis of the local draining lymph nodes of μMT mice suggested significantly reduced activation of CD4 T cells (Fig. 4B). Indeed, using the PmpG-1303–311:I-Ab tetramer, we detected much lower clonal expansion of Chlamydia-specific CD4 T cells in the local draining lymph node of infected μMT mice compared to WT mice (60±12 in μMT mice vs 166±36 in WT mice, p<0.01; Fig. 4C). This reduced local response was also accompanied by dissemination of Chlamydia to the spleen and peritoneal cavity, where ascites was noted 14 days post infection (Fig. 5A and 5B). Analysis of ascites fluid from μMT mice revealed a large proportion of macrophages (F4/80+), monocytes (Gr-1+) and T lymphocytes (Fig. 5C). In addition, PmpG-1-specific CD4 T cells were abundant in ascites, demonstrating that much of the lymphocyte infiltrate into the peritoneal cavity is likely to be Chlamydia-specific (Fig. 5C). Thus, the absence of B cells reduces local CD4 T cell priming and allows bacterial dissemination.
In contrast to the local response, polyclonal CD4 T cells in the spleen of μMT mice displayed evidence of increased activation (Fig. 4B). Consistent with increased systemic activation, expansion of PmpG-1-specific CD4 T cells was markedly increased in the spleen of μMT mice (33600±6044 in μMT mice vs 5494±1164 in WT mice, p<0.001; Fig. 4C). Furthermore, Chlamydia-specific CD4 T cells in μMT mice expressed higher levels of T-bet and produced more IFN-γ than CD4 T cells from WT mice (Fig. 4D and E). A greater percentage of multifunctional CD4 T cells producing IFN-γ and TNF-α was also detected (Fig. 4F). Consistent with an enhanced effector response, the percentage of Chlamydia-specific CD4 T cells expressing CCR7 was considerably lower in μMT mice (Fig. 4D). These data suggest a compensatory systemic T cell response is induced to clear the disseminated bacteria that accompany B cell deficiency.
In vivo visualization studies of pathogen-specific CD4 T cell responses have typically involved adoptive transfer of monoclonal TCR transgenic T cells and have rarely focused on sexually transmitted infections [25], [26]. Here, we describe the generation of peptide-MHC class II tetramers that allow direct visualization of endogenous, polyclonal antigen-specific CD4 T cell responses to Chlamydia infection. Using Chlamydia tetramers and ELISPOT assays, we detected differences in the tempo of the initial CD4 T cell responses to systemic or vaginal challenge with Chlamydia. Unlike other mucosal tissues, the female genital mucosa lacks defined lymphoid structures [7], which may explain the delayed clonal expansion of CD4 T cells after Chlamydia genital tract infection. Alternatively, this delay may represent an unappreciated virulence mechanism of Chlamydia to delay early T cell priming, as has been noted during respiratory infection with MTB [27]. Thus, delayed CD4 T cell priming in the draining lymph node may be due to limited access to bacterial antigen by tissue dendritic cells or impeded migration to the local lymph nodes. Further studies will be required to examine this issue directly.
Our studies show that Chlamydia-specific memory CD4 T cells persist for at least one year after infection, which may suggest the presence of persistent antigen and/or persistent Chlamydia without culturable Chlamydia organisms [28], after the clearance of infectious bacteria. Interestingly, a recent study by Johnson et al suggested that the PmpG-1303–311 epitope can be detected on splenic APC for at least 6 months after the clearance of primary genital tract infection [29]. Our data, together with others, also support the potential role of PmpG-1 specific CD4 T cells in Chlamydia protective immunity [24], [29].
Our data confirm directly the previous notion that T-bet expressing Th1 cells are the predominant CD4 effector lineage among Chlamydia-specific CD4 T cells. However, we also identify small populations with the phenotypic marker of Tregs in the lymph node and spleen following both systemic and mucosal infection, as well as the enrichment of Chlamydia-specific Th17 cells at the genital mucosa itself. These data suggest that Chlamydia-specific Tregs are expanded to regulate the large Th1 response that develops during infection [30]. The enrichment of Th17 cells at mucosal surfaces has also been detected in other infection models [23], [31]–[33]. For example, flagellin-specific Th17 cells are predominantly enriched in the gut mucosal sites after Salmonella oral infection [23]. Further experimentation is required to determine whether these pathogen-specific Th17 cells contribute to Chlamydia clearance and/or the induction of upper genital tract pathology.
The presence of large numbers of B cells in the lower and upper genital tract during Chlamydia infection, as shown by both immunohistochemical staining and flow cytometry [34] (and data not shown), led us to speculate that B cells play an important role during primary Chlamydia infection. Indeed, our data show that in the absence of B cells, there is a marked reduction in antigen-specific CD4 T cell priming within the draining iliac lymph nodes. Given this effect of early CD4 T cell expansion it is possible that B cells participate directly in antigen presentation during the early stages of primary infection. Although this has not previously been observed during genital tract infection, a similar finding was reported after C. muridarum lung infection [19]. A recent study has suggested that macrophage deficiency could also account for protective defects of μMT mice against viral infections [35]. However, our observations that ascites did not occur at early time point (7 days post infection, data not shown) suggested that the phenotype we observed is mediated by altered adaptive immune mechanisms.
Although mild Chlamydia dissemination to other mucosal sites has been reported previously in C57BL/6 mice [36], a large number of disseminated Chlamydia has only previously been found in IFN-γ-deficient and SCID mice and largely attributed to deficient Th1 development [8]. B cell deficient mice therefore provide an unexpected additional model where Chlamydia also disseminate to non-mucosa tissues. Notably, our data provides a clear example where the marked pathology of Chlamydia infection does not always correlate with the inability of host to clear the infection. A likely explanation for highly efficient bacteria clearance in μMT mice is the robust systemic CD4 T cell response that may compensate for the loss of initial CD4 T cell priming within the local draining lymph nodes of the genital tract. The fact that Chlamydia genital tract infection can lead to ascites in the absence of B cells is also clinically relevant: Chlamydia infection induces ascites in patients with salpingitis and peritonitis [37]–[40], although it is unclear whether Chlamydia dissemination in mouse models reflects the same mechanism that leads to symptoms in human such as PID and Fitz-Hugh-Curtis syndrome. Further pathological studies are needed to understand differences of mouse and human infections.
Overall, our data demonstrate unappreciated heterogeneity of the CD4 T cell response to genital tract infection in a model where Th1 cells are essential for protective immunity. In addition, our data uncover a surprising involvement of B cells in local expansion of effector Chlamydia-specific T cells in the genital tract and prevention of bacterial dissemination. Greater understanding of the mechanism of Chlamydia dissemination from the genital tract and the T and B cell responses that restrain this spread of bacteria may reduce the risk of sequelae after Chlamydia genital infection in infected women.
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 University of California Davis is accredited by the Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC). All animal experiments were approved by University of California Davis Institutional Animal Care and Use Committee (IACUC) (Protocol number 16612).
C57BL/6 (B6) mice were purchased from the National Cancer Institute (Frederick, MD) and The Jackson Laboratory (Bar Harbor, ME). μMT mice were purchased from The Jackson Laboratory (Bar Harbor, ME). RORγt-GFP reporter mice were obtained from Dr. Marc Jenkins (University of Minnesota). Mice used for experiments were 6–12 weeks old, unless otherwise noted. All mice were maintained in accordance with University of California Davis Research Animal Resource guidelines.
Chlamydia muridarum strain Nigg II was purchased from ATCC (Manassas, VA). The organism was cultured in HeLa 229 cells in Dulbecco's Modified Eagle Medium (DMEM) (Life Technologies, Grand Island, NY) supplemented with 10% fetal bovine serum (FBS). Elementary bodies (EBs) were purified by discontinuous density gradient centrifugation as previously described and stored at −80 degree [41]. The number of IFUs of purified EBs was determined by infection of HeLa 229 cells and enumeration of inclusions that were stained with anti-Chlamydia MOMP antibody. Heat-killed EBs (HKEBs) were prepared by heating at 56°C for 30 min. A fresh aliquot was thawed and used for every infection experiment. Salmonella enterica serovar Typhimurium strains BRD509 (AroA−D−) were kindly provided by Dr. D. Xu (University of Glasgow, Glasgow, U.K).
For systemic infection, mice were injected intravenously in the lateral tail vein with 1×105 C. muridarum. To enumerate the bacteria burden in tissues, the spleen, liver and kidney, were crushed in 5 mL SPG buffer and tissue homogenate was placed in a tube with glass beads to disrupt cells. After shaking for 5 min, and centrifugation at 500 g for 10 minutes, supernatants were collected and serial dilutions were plated on HeLa 229 cells. For intravaginal infections, estrus was synchronized by subcutaneous injection of 2.5 mg medroxyprogesterone acetate (Greenstone, NJ) 7 days before infection. 1×105 C. muridarum in 5 µL SPG buffer were then deposited into vaginal vaults. To enumerate bacteria, vaginal swabs were collected, shaken with glass beads, and serial dilutions were plated on HeLa 229 cells.
Spleen and lymph nodes (axillary, brachial, inguinal and mesenteric) were harvested and a single-cell suspension prepared. After RBC lysis, CD4+ T cells were enriched using LS MACS columns and anti-CD4 magnetic beads (Miltenyi Biotec, Auburn, CA). Enriched CD4+ T cells were incubated with irradiated APCs from naive mice in the presence of 10 µM Chlamydia peptide (RplF51–59, FabG157–165, Aasf24–32 or PmpG-1303–311) in 96-well ELISPOT plates (Millipore, Billerica, MA) that had been pre-coated with purified anti-IFN-γ (BD Biosciences, San Diego, CA). The RplF51–59, FabG157–165, Aasf24–32 and PmpG-1303–311 epitopes used for stimulation have been described previously [21], [42]. After 20 h of incubation at 37°C, cells were washed and cytokine spots developed using biotinylated anti-IFN-γ (BD Biosciences), AKP Streptavidin (BD Biosciences), and 1-Step NBT/BCIP substrate (Thermo Scientific, Waltham, MA). Cytokine spots were counted using an ImmunoSpot S5 Core Analyzer (C.T.L., Shaker Heights, OH), and the total number of IFN-γ-producing CD4+ T cells per spleen was calculated.
The methodology for construction of pMHCII tetramers has been described in detail [22]. Briefly, biotinylated I-Ab monomers containing a covalently linked C. muridarum peptide (RplF51–59, Aasf24–32 or PmpG-1303–311) were expressed by S2 Drosophila insect cell lines and cultured in a Wave Bioreactor (GE Healthcare Biosciences, Pittsburgh, PA). After purification, I-Ab monomers were tetramerized by co-incubated with fluorochrome-conjugated streptavidin at a pre-determined optimal ratio at room temperature for 30 min [22]. To test for tetramer specificity, C57BL/6 mice were immunized with each of the three peptides and CFA (Sigma-Aldrich, St. Louis, MO). Seven days post-immunization, draining lymph nodes were isolated and tetramer positive cells enriched using the methodology described below.
Spleen and LNs were harvested from naïve or infected mice. Single cell suspensions were prepared in FACS buffer (PBS with 2% FCS) and stained with tetramers in Fc block (culture supernatant from the 24G2 hybridoma, 2% mouse serum, 2% rat serum, and 0.01% sodium azide) for 1 h at room temperature in the dark. Cells were then washed and tetramer positive cells enriched via LS MACS columns and anti-fluorochrome magnetic beads (Miltenyi Biotec, Auburn, CA). Bound and unbound fractions were stained with a panel of monoclonal antibodies (listed below) and analyzed on a FACSCanto or an LSRFortessa flow cytometer (BD Biosciences, San Jose, CA). To stain for intracellular transcription factors and cytokines, cells were left untreated or stimulated with phorbol 12-myristate 13-acetate (PMA, 50 ng/ml), ionomycin (200 ng/ml) in the presence of brefeldin A (10 µg/ml) for 4 h at 37°C. After surface staining, cells were fixed, permeablized and stained using the Foxp3 staining Kit (eBioscience, San Diego, CA). Antibodies for staining included FITC-CD11b, CD11c, F4/80, B220, TNFα; PerCP-eFlour710-CD4; APC-CCR7; eFlour660-T-bet; Alexa700-CD44; eFlour450-CD3, Foxp3, IFN-γ (eBioscience, San Diego, CA); FITC-IL-17A and APC-Cy7-CD8 (BD Biosciences, San Diego, CA). Flow data were analyzed using FlowJo software (Tree Star, Ashland, OR) and endogenous, tetramer-specific CD4 T cells were identified using a previously published gating strategy [22].
All data sets were analyzed by unpaired Student's t-test using Prism (GraphPad Software, La Jolla, CA). A p value<0.05 were considered statistically significant.
|
10.1371/journal.ppat.1002566 | Kaposi's Sarcoma Herpesvirus Upregulates Aurora A Expression to Promote p53 Phosphorylation and Ubiquitylation | Aberrant expression of Aurora A kinase has been frequently implicated in many cancers and contributes to chromosome instability and phosphorylation-mediated ubiquitylation and degradation of p53 for tumorigenesis. Previous studies showed that p53 is degraded by Kaposi's sarcoma herpesvirus (KSHV) encoded latency-associated nuclear antigen (LANA) through its SOCS-box (suppressor of cytokine signaling, LANASOCS) motif-mediated recruitment of the EC5S ubiquitin complex. Here we demonstrate that Aurora A transcriptional expression is upregulated by LANA and markedly elevated in both Kaposi's sarcoma tissue and human primary cells infected with KSHV. Moreover, reintroduction of Aurora A dramatically enhances the binding affinity of p53 with LANA and LANASOCS-mediated ubiquitylation of p53 which requires phosphorylation on Ser215 and Ser315. Small hairpin RNA or a dominant negative mutant of Aurora A kinase efficiently disrupts LANA-induced p53 ubiquitylation and degradation, and leads to induction of p53 transcriptional and apoptotic activities. These studies provide new insights into the mechanisms by which LANA can upregulate expression of a cellular oncogene and simultaneously destabilize the activities of the p53 tumor suppressor in KSHV-associated human cancers.
| Aurora kinases are evolutionally conserved serine/theronine kinases that regulate cell mitotic progression in eukaryotic cells. Aurora kinase A, B and C were identified in mammalian cells. Among them, Aurora A was first known to regulate genomic instability and tumorigenesis, and is frequently amplified in multiple human cancers. Aurora-kinase inhibition has been shown to effectively block cell growth and induce death of cancer cells. Kaposi's sarcoma-associated herpesvirus (KSHV) encoded latency-associated nuclear antigen (LANA) is essential for KSHV-induced transformation of primary human B-lymphocytes and endothelial cells. In this study, we discovered that LANA remarkably enhances Aurora A production, and that elevated Aurora A acts as a negative regulator to induce phosphorylation and LANA-mediated ubiquitylation of p53. Importantly, inhibition of Aurora A production leads to cell death of KSHV-positive B lymphoma cells. This study clearly demonstrates that Aurora A is targeted by an oncogenic virus for inhibition of p53 function, and is a potential target for viral associated cancer therapy.
| Kaposi's sarcoma-associated herpesvirus (KSHV), also named human herpesvirus 8, is a member of the gamma-herpesviruses and is associated with Kaposi's sarcoma (KS), multicentric Castleman's disease (MCD) and primary effusion lymphoma (PEL) [1]–[4]. Studies have shown that PELs are dependent on KSHV for survival, as loss of the KSHV genome results in cell death [5]. These findings demonstrate that KSHV infection can reprogram cellular gene function and thereby mediate viral oncogenesis. KSHV is predominantly latent in most cells in KSHV-associated lesions and during latency only a few viral genes are expressed. The latency associated nuclear antigen (LANA) encoded by open reading frame (ORF) 73, is one of the critical KSHV encoded latent antigens, and is expressed in viral infected tumor cells of KSHV-associated malignancies [6], [7]. LANA plays a multifunctional role contributing to viral persistence and tumorigenesis through targeting DNA replication, chromosome tethering, anti-apoptosis, cell cycle regulatory, and gene regulatory functions [8]–[13]. At the gene transcription level, LANA exerts broad repressive or activation effects by interacting with a number of transcriptional factors including mSin3A, CBP, RING3, GSK-3β and p53 for its transcription repression activities [8], [14]–[16], and E2F, Sp1, RBP-Jκ, ATF4, CBP, Id-1, and Ets to drive transcriptional activation [17]–[22].
Aurora A, a centrosome-associated Serine/Threonine oncogenic kinase, was first identified as a human homologue of the Aurora/Ipl1p kinase family [23]. The human Aurora A gene is located at chromosomal region 20q13.2 and contains a 1209-bp open reading frame that encodes 403 amino acids with a molecular weight of 46 kDa [24]. The promoter of Aurora A contains three putative binding sites for transcription factors: E2F, Sp1 and Ets [25]. Aurora A localizes around centrosomes during interphase and prophase, on the microtubules near spindle poles in metaphase and the polar microtubles during anaphase & telophase [26]. Aurora A participates in multiple functions associated with mitotic events, including centrosome maturation and separation, bipolar spindle assembly, chromosome alignment and cytokinesis [27]. Enhanced expression of Aurora A can lead to centrosome amplification and aneuploidy as a results of incomplete cytokinesis, which results in either cell death or survival through malignant transformation in a p53-dependent manner [28], [29]. Aberrant expression of Aurora A has been reported in a wide variety of tumor types and in most human cancer cell lines [24], [28], [30]–[32]. A number of substrates of Aurora kinase A have been identified, such as TPX2 [33], Eg5 [34], CDC25B [35], p53 [36] and BRCA-1 [37].
Like the aberrant expression of Aurora A, loss of p53 function also induces similar phenotypes of centrosome amplification and aneuploidy in cells [38], [39]. It is well known that wild type p53 is able to induce growth arrest or apoptosis when cells are exposed to stress, and p53 is frequently mutated or deleted in human cancers [38], [39]. As a substrate of Aurora A kinase, p53 can be phosphorylated on both Ser215 and Ser315 and this leads to destabilization and inhibition of p53 via the Mdm2-mediated ubiquitylation and proteasomal degradation pathway [36], [40]. In KSHV latently infected cells, our previous studies have shown that p53 can be degraded by the cellular EC5S ubiquitin complex-mediated pathway targeted by the SOCS motif of LANA [41]. In this study, we further show that the protein levels of Aurora A kinase is upregulated by LANA, and that elevated Aurora A induces phosphorylation of p53 which enhances the interaction of LANA with p53, and promotes LANA-mediated p53 ubiquitylation and degradation, and hence inhibition of p53 transcriptional and apoptotic activities.
Aurora A has been shown to aberrantly accumulate and inhibit p53 function in most cancer cells. To analyze whether LANA-mediated inhibition of p53 in KSHV latently infected cells is associated with Aurora A, we first tested the protein levels of Aurora A in KSHV positive KS tumor samples and KSHV negative normal tissues by immunohistochemistry assays. The results showed that Aurora A expression is highly expressed in KS patient tissue and not in normal tissue (Figure 1A). The observation from de novo infection of human primary cells with KSHV in vitro shows that although there was an overall increase of Aurora A mRNA transcript and protein levels up to 7-days post-infection, the peak level of Aurora A was at 2-days post-infection (Figure 1B and C). This strongly indicates that elevated expression of Aurora A was indeed associated with KSHV infection, and that the early rapid enhancement of Aurora A expression could be important for KSHV to establish long term latent infection. To elucidate the role of LANA on Aurora A expression, LANA in KSHV-positive BC3 cell line was transiently knocked down by introduction of small interference RNA specifically against LANA without interrupting other latent transcripts (supplementary Figure S1). The levels were monitored by western blot analysis. The results showed that the protein levels of Aurora A were greatly decreased once LANA expression was reduced (Figure 2A, left panel). In addition, a significant reduction in Aurora A transcripts was observed after LANA was inhibited (Figure 2A, right panel), supporting the hypothesis that LANA can modulate the transcriptional activity of Aurora A promoter and so regulate Aurora A expression. This is further confirmed by the fact that LANA induces a dose-dependent increase in Aurora expression in HEK293 cells with transient transfection (Figure 2B). To determine which phase of cell cycle expressing Aurora A is affected by LANA, the levels of Aurora A at different cell cycle phases in the presence or absence of LANA were determined by flow cytometric analysis. The results showed that LANA dramatically enhanced Aurora A expression at G1 and S phases but mildly so at G2/M phase (Figure 2C and D). This indicated that LANA can play a role in maintaining higher level of Aurora A at different phases of cell cycle.
Given the modulation of Aurora A expression by LANA, we constructed a luciferase reporter driven by Aurora A promoter (position −527 to +387) (supplementary Figure S2A), and performed reporter assays in the presence or absence of LANA in both HEK293 and DG75 cells. The results showed that LANA dramatically enhanced the transcription level of the Aurora A promoter when compared to vector control (supplementary Figure S2B), which was further corroborated by a dose-dependent induction (supplementary Figure S2C).
A transcription factor analysis and previous studies identified three cis elements E2F, Sp1 and Ets within the Aurora A promoter [42]. To define which cis element is critical for LANA to activate Aurora A transcription, a series of mutants of the Aurora A promoter driving the luciferase reporter gene were generated and subjected to reporter assays in HEK 293 cells in the presence or absence of the LANA. As shown in Figure 3A, in the presence of LANA, the Aurora A promoter with specific deletions of E2F or Ets transcription factor-related locus (ΔE2F or ΔEts) resulted in a mild difference in stimulation of Aurora A promoter activity when compared to the wild type Aurora A promoter (E2F+Sp1+Ets). However, the Aurora A promoter with deletion of both Sp1 and Ets elements (ΔSp1+Ets) or Sp1 alone (ΔSp1) led to a remarkable decrease in the promoter activity (Figure 3A). These results strongly suggested that the Sp1 responsive element within the Aurora A promoter is a major cis element modulated by LANA. To further confirm that the Sp1-binding site within the Aurora A promoter is a major target for LANA, we performed chromatin-immunoprecipitation assays with Sp1 or E2F specific antibodies using BC3 cells with or without LANA knockdown. Consistently, the results showed that Sp1 had a much higher affinity than E2F when compared to the non-specific IgG control in binding to Aurora A promoter (Figure 3B). Inhibition of LANA expression dramatically reduced the association of Sp1 to its cis-acting element (Figure 3B). This suggests that LANA can enhance Aurora A transcript levels through targeting of the Sp1 cis-element within the promoter.
To further determine if Aurora A is important for LANA to inhibit p53 function, we performed reporter assays in the p53-null Saos-2 cells by using a luciferase reporter driven by 13 consensus p53-binding sites. The results from reporter assays showed that both LANA and Aurora A individually reduced the transcriptional activity of p53 (Figure 4A, lane 3 and 4), and co-expression of Aurora A dramatically enhanced LANA-mediated inhibitory function of p53 expression (Figure 4A, lane 5). To determine if Aurora A-associated Mdm2 plays a role in cooperation of Aurora A with LANA as it relates to inhibition of p53, we also performed similar reporter assays in both Mdm2+/+ and Mdm2−/− cells. The results further showed that in the absence or presence of Mdm2, Aurora A maintained its ability to collaborate with LANA in regards to inhibition of p53 transactivation (Figure 4B and C, lane 1 and 2). However, compared to wild type Aurora A, the kinase inactive mutant (KR) of Aurora A can greatly reverse LANA-mediated inhibition of p53 transactivation (Figure 4B and C, compare lane 2 with 3). This indicates that the KR mutant of Aurora A can act as a dominant negative molecule and that the kinase activity of Aurora A is required for LANA -induced repression of p53 transcriptional activity. Strikingly, in all three cell lines (p53−/−, p53−/−Mdm2−/− and p53+/+Mdm2+/+), we observed that co-expression of wild type Aurora A but not its KR mutant dramatically reduced the protein levels of p53 inhibited by LANA (Figure 4A, B and C, lower panels). Therefore, the kinase activity of Aurora A contributes to LANA-mediated degradation of p53 and so a reduction in p53 transcriptional activity.
Our previous work has shown that LANA can recruit the EC5S ubiquitin complex to degrade p53 [41]. To determine if Aurora A enhances LANA-mediated p53 proteolytic degradation, we assessed the stability of p53 in HEK293 cells transfected with LANA plus or minus Aurora A by exposure to cycloheximide for different time points. The results consistently showed lower levels of p53 in cells transfected with Aurora A than in cells transfected with a control vector (supplementary Figure S3). Treatment with cycloheximide for 210 min decreased p53 levels by almost 75% in Aurora A transfected cells compared to about 40% in the control cells (supplementary Figure S3, lane 4 and 8). Phosphorylation of p53 by Aurora A results in greater binding affinity with Mdm2 [40]. To determine if active Aurora A kinase enhanced the interaction of p53 with LANA, we monitored the association of p53 with LANA in the presence of wild type Aurora A or its inactive mutant KR by coimmunoprecipitation assays. Substantially more LANA protein was pulled down by p53 in the presence of wild type Aurora A than in the presence of the mutant control in both Saos-2 and MEF cell lines (Figure 5A).
To further determine if Aurora A-induced greater binding of p53 to LANA will result in LANASOCS-mediated ubiquitylation of p53, we performed p53 ubiquitylation assays in MEF cells expressing different combinations of wild type LANA, mutants of SOCS-motif deleted mutant LANA (ΔSOCS), wild type Aurora A and its mutant KR. The results showed that there is substantially higher amount of ubiquitylated p53 in cells expressing wild type LANA and Aurora A than in cells expressing only wild type LANA or Aurora A (Figure 5B, compare lane 4 with 2 and 3). Consistently, mutation of the SOCS motif of LANA, the inactive form of Aurora A or both led to a dramatic reduction of ubiquitylated p53 (Figure 5B, compare lane 4 with 5, 6, and 7). This indicates that LANASOCS-mediated ubiquitylation of p53 through the EC5S ubiquitin complex requires the kinase activity of Aurora A.
Previous studies reported that Aurora A is able to phosphorylate p53 at Ser215 and Ser315 [36], [40]. To determine if LANA-mediated p53 ubiquitylation is dependent on Aurora A-induced phosphorylation of p53 on Ser215 or Ser315, we performed similar ubiquitylation assays by using wild-type, S215A or S315A variants of p53 with wild-type Aurora A and LANA in MEF cells. The results showed that compared with wild type p53, less ubiquitylated p53 appeared in cells expressing S215A, S315A or the S215A/S315A mutant of p53, and S315A had a greater impact than S215A (Figure 6A, lane 2 with 3, 4, and 5), suggesting that both Ser215 and Ser315 are phosphorylated by Aurora A and in turn facilitates LANA-induced ubiquitylation of p53. To further confirm that LANA does induce p53 phosphorylation through Aurora A, we coexpressed p53-FLAG with either LANA or empty vector in the presence of shAurora A or non-specific shLuc control in Saos-2 cells, followed by immunoblotting analysis with antibodies against the Ser315-phosphorylated p53. The results showed that LANA dramatically enhanced the level of phosphorylated p53 on Ser315 which dependent on Aurora A expression (Figure 6B, compare lane 2 with 1 and 3). Furthermore, Aurora A-mediated phosphorylation of p53 is required for LANA induced p53 ubiquitylation and degradation.
It has been shown earlier that the increased expression of LANA can inhibit p53-induced apoptosis in p53-null Saos-2 cells [8]. To determine if Aurora A cooperation with LANA could affect p53-mediated apoptosis, Saos-2 cells transiently transfected with constructs expressing wild-type p53, plus either LANA, Aurora A or both LANA (WT) and Aurora A (WT or KR) in different combination, were subjected to cell cycle profile analysis. The results showed that the percentage of p53-induced apoptosis as determined by the subG1 population was greatly decreased with coexpression of LANA and Aurora A (Figure 7A, compare lane 2 with 3, 4 and 5, supplementary Figure S4). However, the combination of LANA with Aurora A mutant KR or LANA ΔSOCS with wild type Aurora A, resulted in a dramatic reversal of the levels of p53 mediated apoptosis (Figure 7A, compare lane 2 with 6 and 7, supplementary Figure S4). To further confirm that LANA does require Aurora A to inhibit p53-mediated apoptosis, we performed colony formation assays by coexpressing p53 with either wild type LANA or its deleted mutant (ΔSOCS), or empty vector in the presence or absence of Aurora A knockdown. Consistent with the expression level of p53 at 2 day posttransfection, the results of 3-weeks culture showed that p53 expression alone dramatically blocks colony formation, and wild type LANA but not its ΔSOCS mutant markedly reversed the inhibitory activities of p53 on colony formation (Figure 7B, compare lane 2 with 3 and 4). In contrast, there was no significant difference between LANA and p53 coexpression compared to p53 alone, once the expression of Aurora A was knocked down (Figure 7B, compare lane 6 with 7 and 8). Unexpectedly, we did observe that Aurora A knockdown alone remarkably reduced colony formation even without p53 coexpression (Figure 7B, compare lane 2 with 1). These suggest that Aurora A is important for LANA to inhibit p53-induced apoptosis by combining their phosphorylation and ubiquitylation activities.
To further investigate the effect of Aurora A on the growth and survival of KSHV-infected B cells, the Aurora A expression in BC3 cells was knocked down by lentivirus-mediated shRNA against Aurora A. Lack of Aurora A expression rescued p53 expression about 3 fold in BC3 cells and greatly induced PARP1 cleavage, compared with BC3 cells transfected with control shRNA (Figure 8A). We also found that exposure of BC3 cells to low sera resulted in a marked increase of more than 4N DNA content in the cells, as well as the number of apoptotic cells due to inhibition of Aurora A expression (Figure 8B and C). Subsequently, the results of immunofluorescence analysis revealed that cells with shRNA against Aurora A exhibited a dramatic alteration in cellular and nuclear morphology. The cells were enlarged and exhibited an increased presence of multiple and fragmented nuclei (Figure 8D). Cells with multiple nuclei (>6N) could be readily seen. Furthermore, the fragmented nuclei were apparent in cells transduced with shAurora A. The proportion of multinucleated cells or apoptotic cells was progressively increased once Aurora A was knocked down (Figure 8D and E), indicating that inhibition of Aurora A in KSHV-infected cells was sufficient to induce cell apoptosis. In addition, the result of Aurora A knockdown led to dramatic increase of p53 accumulation and subG1 population in 293 cells with KSHV infection but not mock 293 cells (Figure 8F), further supporting the notion that Aurora A is targeted by KSHV for inhibition of p53-mediated apoptotic function.
We and other groups have previously shown that LANA downregulates p53-associated pathways in the KSHV latently infected B lymphoma cells [8], [41]. In this study, we aimed to better understand the mechanism underlying LANA-mediated suppression of p53, a key protein already known to link with chromosomal instability and apoptosis. With this in mind, we performed apoptotic and chromosomal stability gene microarray analysis by using LANA stable expressing cell line and identified the Aurora A kinase, a mitotic checkpoint protein as one of the genes upregulated due to LANA expression. Further studies showed that expression of Aurora A upregulated by LANA is mainly through enhancing the binding capacity of transcriptional factor Sp1 to the Aurora A promoter. The elevated levels of Aurora A subsequently resulted in phosphorylation of p53 at Ser215 and Ser315 thus facilitating LANA-mediated ubiquitylation and destabilization of p53 (Figure 9). Therefore, LANA has a dual function in terms of regulating p53 by: 1) Enhancing Aurora A kinase expression to phosphorylate p53; and 2) Recruiting the EC5S ubiquitin complex to induce ubiquitylation of phosphorylated p53. A previous report showed that increased expression of Aurora A as a result of genetic mutations increased the growth and survival of HTLV-1-infected T cells [43]. Our data provide the first evidence showing that a viral protein can directly target the oncogenic Aurora A kinase for inhibition of p53 by enhancing the transcriptional activity of the Aurora A promoter.
The Aurora A gene is located at chromosome 20q13.2, a region frequently amplified and over-expressed in a variety of human tumors and cancer derived cell lines [27], [29], [44]. However, we did not observe a consistent difference in levels of Aurora A expression in KSHV-positive cell lines (BC3, BCBL1, JSC1 and BC1) and in KSHV negative cell lines (BJAB, Ramos, Loukes and DG75). To support our hypothesis that KSHV latent infection is directly associated with the elevated production of Aurora A, the levels of Aurora A transcripts in human primary cells with and without GFP-tagged KSHV infection were analyzed. Although the increased pattern of Aurora A (similar pattern for Aurora B, data not shown) transcripts wasn't consistently correlated with LANA expression within 7-days posttransfection, the base levels of Aurora A transcripts in KSHV-infected cells remained higher than that in KSHV-uninfected cells. One potential explanation for higher level of Aurora A mRNA transcripts by 2-days primary infection may be a benefit to rapidly establish KSHV latent infection. However, the fact that Aurora A transcript drops following persistent LANA expression after 2-days primary infection could be another benefit to the dynamical modulation of Aurora A by LANA and/or other factors during cell cycle progression. In agreement with the critical role of LANA on Aurora A expression, the evidence have showed that increased transient expression of LANA can upregulate Aurora A expression in a dose-dependent manner, and that knock-down of LANA in KSHV-positive cell lines decreased both the transcript and protein levels of Aurora A. In addition, immunohistochemistry assays further revealed that expression of high levels of Aurora A in cells from KS tissue is correlated with the presence of LANA in KSHV-positive cells.
There are at least 20 phosphorylation sites reported for human p53 [45]. Most of the amino terminal specific phosphorylation sites prevents Mdm2-mediated ubiquitylation and thus stabilizes p53 [45]. In contrast, phosphorylation of p53 at its carboxyl terminus often promotes p53 degradation [45]. For example, phosphorylation of Ser362/366 by NF-κB induces p53 degradation [46]. In regards to the effect of Aurora A-induced phosphorylation of p53, Aurora A is able to induce Mdm2-mediated p53 degradation by phosphorylating Ser315 [40], and abrogate the DNA binding and transactivation activity of p53 by phosphorylating Ser215 not Ser315 [36]. Although this is somehow controversial, these reports implicate that p53 is a physiological substrate of Aurora A and that Aurora A may exert its regulatory functions on p53 through phosphorylation of Ser215 and Ser315 in different cell types. Our data showed that mutation of p53 at Ser315 dramatically reduced LANA-mediated ubiquitylation of p53 in the presence of Aurora A. However, mutation at Ser215 also reversed the inhibition of p53 ubiquitylation induced by LANA. Therefore, the virus has broad control of the kinase activity of Aurora A in its inhibition of p53 which overcomes the issue of cell type specificity. In addition, identification of Aurora A kinase for Ser315 phosphorylation does not rule out possible involvement of other cell cyclin-dependent kinases, as precedents for multiple distinct kinases targeting the same phosphorylation site of p53 (ATM and ATR for Ser15) have been reported [47], [48].
To determine the role of LANA in destabilization of p53 phosphorylated by Aurora A kinase, we compared the effect of wild-type LANA and mutant LANA (ΔSOCS) lacking the ability to ubiquitinate p53 and showed that the ubiquitination activity of LANA is critical for destabilization of p53 after induction of Aurora A expression. This suggests that Aurora A phosphorylates p53, thus enhancing LANA-mediated degradation of p53. Recent reports show that KSHV targets Mdm2 to deregulate the p53 tumor suppressor pathway [49], [50]. Interestingly, our reporter assay found that the protein level and transactivation activity of p53 were dramatically reduced by LANA together with Aurora A in an Mdm2-independent manner. Phosphorylation of p53 by Aurora A results in greater binding affinity to Mdm2 and acceleration of its degradation [40]. However, the possibility exists that Mdm2 contributes to ubiquitylation and degradation of Aurora A-mediated phosphorylation of p53 in LANA-expressing cells. Moreover, wild type Aurora A but not its kinase deficient mutant enhanced interaction of p53 with LANA, and suggests that Aurora A-induced phosphorylated p53 potentially has a higher affinity for LANA than the native p53. However, whether Aurora A is directly involved in a complex which contains p53 and LANA to increase their ability to interact remains to be further investigated.
Aurora is a subfamily of serine/threonine protein kinases that plays critical roles in centrosome cycling, spindle assemble, chromosome segregation and cell division [51], [52]. The observation of growth arrest of cells at the G2-M phase after silencing of Aurora A kinase in KSHV positive PEL cells, suggests that degradation of p53 phosphorylated at Ser215 and Ser315, has physiological relevance in allowing progression of normal cell proliferation cycle. In regard to our observation that Aurora A knockdown alone also induces less colony formation in the p53-null cells, the function of Aurora A on p53 is likely to be the major but not unique signaling pathway for blocking apoptosis. Our data also indicated that coexpression of LANA and Aurora A was able to inhibit p53-mediated apoptosis in the p53-null cell line, Saos-2. Based on our ubiquitylation assays and protein stability assays, Aurora A is important for LANA to suppress p53-dependent apoptosis by phosphorylating p53, followed by its ubiquitination and degradation. This function of LANA may explain in part why KSHV latent infection causes oncogenic transformation in mammalian cells [53].
In summary, Aurora A is a candidate target for inhibition of tumor growth in a broad range of cancers including pancreatic and leukemia cells in which it is amplified or induced. Our studies demonstrates that transcription activation of Aurora A by the viral protein LANA leads to induced expression of this protein in KSHV infected cells. Down-regulation of Aurora A by RNA interference induces cell-cycle arrest, aberrant chromosomal segregation and apoptosis (about 4-fold higher) in PEL cells, suggesting that Aurora A could be a promising therapeutic target in PEL cells.
De-indentified Human peripheral blood mononuclear cells (PBMCs) were obtained from the University of Pennsylvania CFAR Immunology Core. The Core maintains an IRB approved protocol in which Declaration of Helsinki protocols were followed and each donor gave written, informed consent.
DNA constructs expressing LANA full length and SOCS-box mutant in the pA3M vector were described previously [22], [41]. The p53 reporter plasmid containing 13 copies of p53-binding sites at upstream of the luciferase gene was provided by Wafik S. EI-Deiry (Milton Hersley Medical School, Hersley) [54]. pcDNA4(TO/myc-His)-Aurora A wild type and kinase inactive K/R mutant were provided by Erich A. Nigg (Max-Planck Institute of Biochemistry, Martinsried, Germany). HA-p53 WT, S215A, S315A and S215A/S315A mutant were a kind gift from Jin Q. Cheng (University of South Florida, Tampa, Florida). The Aurora A promoter-driven luciferase plasmids pGL3-pAurora A (E2F+Sp1+Ets) were generated by PCR amplicon inserted into pGL3 vector with KpnI and BglII digestion. The deleted (ΔE2F, ΔEts, ΔSp1 or ΔSp1+Ets) mutants of Aurora A promoter are derived from pGL3-pAurora A. The monoclonal antibody anti-myc (9E10) and HA (12CA5) were prepared from hybridoma cultures. Goat ployclonal antibody against Aurora A (Ark-1, N-20) was purchased from Santa Cruz Biotechnology, Inc. (Santa Cruz, CA). Mouse monoclonal antibody against FLAG epitope (M2) was purchased from Sigma-Aldrich Corp. (St. Louis, MO). Rabbit antibody against phosphorylated p53 (Ser315) was purchased from Cell Signaling.
The B lymphoma cell lines BC3 (KSHV positive) and DG75 (KSHV negative) were cultured in RPMI 1640 medium supplemented with 7% fetal bovine serum, 2 mM L-glutamine, and penicillin-streptomycin (5 U/ml and 5 µg/ml, respectively). HEK 293, MEF, and Saos-2 cells were cultured in Dulbecco's modified Eagle's medium (DMEM) supplemented with 5% fetal bovine serum, 2 mM L-glutamine, and penicillin-streptomycin (5 U/ml and 5 µg/ml, respectively). All cell lines were grown at 37°C in a humidified environment supplemented with 5% CO2.
Slides mounted with sections of paraffin-embedded, archival, deidentified KS tissue specimens were a generous gift from Michael Feldman (Department of Pathology and Laboratory Medicine, Hospital of the University of Pennsylvania, Philadelphia, PA). Slides were deparaynized in xylene and rehydrated through graded alcohols (70, 80, and 100% alcohol; 5 min each). Endogenous peroxidase activity was blocked with 3% hydrogen peroxide for 10 min. Following antigen retrieval in 10 mM sodium citrate buffer (pH 6.0), samples were blocked with 10% normal rabbit/goat serum prior to incubation with primary antibodies overnight at 4°C. Secondary antibody (biotinylated anti-rabbit/goat IgG) and streptavidin–peroxidase conjugate (S-P kit, DAKO) were added according to the manufacturer's instructions. Color reaction was developed using diaminobenzidine chromogen solution (Liquid DAB, DAKO).
The 293/Bac36 (GFP-KSHV) cells were subjected to induction by TPA and Sodium butyrate for KSHV virion production. After induction, the supernatant of culture medium was collected and filtered through 0.45 µm filter, and viral particles were spun down at 25,000 rpm for 2 h, at 4°C. The concentrated virus was collected and used for infection or viral DNA quantitation. For primary infection, 10×106 PBMC cells were incubated with virus suspension in 1 ml of RPMI 1640 (10% FBS) medium in the presence of 5 µg/ml Polybrene (Sigma, Marlborough, MA) and incubated for 4 hrs at 37°C. Cells were centrifuged for 5 min at 1500 rpm, the supernatant discarded, pelleted cells were resuspended in fresh RPMI 1640 (10% FBS) medium in 6 well plates and cultured at 5% CO2, 37°C humidified incubator. The positive infection was checked by detection of LANA expression.
Total RNA from cells was extracted using Trizol reagent and cDNA was made with a Superscript II reverse transcription kit (Invitrogen, Inc., Carlsbad, CA). The primers for real-time PCR were as followings: for Aurora A: 5′-GGAGAGCTTAAAATTGCAGATTTTG-3′ and 5′-GGCAAACACATACCAAGAGA CCT-3′; for LANA: 5′-CATACGAACTCCAGGTCTGTG-3′ and 5′-GGTGGAAGAGCC CATAATCT-3′; for vCyclinD: 5′-TAATAGAGGCGGGCAATGAG-3′ and 5′-ACTCCTTT TCCCGCCAAGAAC-3′; for vFLIP: 5′-GTGTAAGAATGTCTGTGGTGTGC-3′ and 5′- GCGGAATGTCTGTTTCGTGC-3′; and for GAPDH: 5′-CTCCTCTGACTTCAACAGC G-3′ and 5′-GC CAAATTCGTTGTCATACCAG-3′. The cDNA was amplified by using 10 µl of Master Mix from the DyNAmo SYBR green quantitative real-time PCR kit (MJ Research, Inc.), 1 µM of each primer, and 2 µl of the cDNA product in a 20-µl total volume. Thirty cycles of 1 min at 94°C, 30 s at 55°C, and 40 s at 72°C were followed by 10 min at 72°C in an MJ Research Opticon II thermocycler (MJ Research, Inc., Waltham, MA). A melting curve analysis was performed to verify the specificities of the amplified products. The values for the relative levels of change were calculated by the “delta delta threshold cycle” (ΔΔCT) method and each sample were tested in triplicates.
Small hairpin RNAs (shRNA) complementary to the C-terminal (GCTAGGCCACAACACATCT) fragment of LANA as described previously [41], or N-terminal (GTCTTGTGTCCTTCAAATT) fragment of Aurora A was individually cloned into pGIPz vector according to the manufacture's instructions (Open Biosystem, Inc, Huntsville, AL) to generate shLANA and shAurora A constructs. pGIPz vector with luciferase target sequence (shLuc) was used as a control. Fifteen million BC3 cells were transfected by electroporation (220v, 975 µF, 0.4-cm-gap cuvette) with 20 µg of shLANA (or shLuc). BC3 knockdown stable cells were selected and maintained in 4 µg/ml puromycin.
Transfected cells were harvested, washed with ice-cold PBS, and lysed in ice-cold RIPA buffer. Cell debris was removed by centrifugation at 21,000× g (10 min and 4°C), and the supernatant was transferred to a fresh microcentrifuge tube. Lysates were then precleared by end-over-end rotation with normal mouse serum and a 1∶1 mixture of Protein A and Protein G-conjugated Sepharose beads (1 hr, 4°C). Beads were spun out, and the supernatant was transferred to a fresh microcentrifuge tube. The protein of interest was captured by rotating the remaining lysate with 1 µg of appropriate antibody overnight at 4°C. Immune complexes were captured with 30 µl of a 1∶1 mixture of Protein A and Protein G Sepharose beads, pelleted, and washed five times with ice-cold RIPA buffer. For immunoblotting assays, input lysates and immunoprecipitated (IP) complexes were boiled in Laemmli buffer, fractionated by SDS-PAGE, and transferred to a 0.45-µm nitrocellulose membrane. The membrane was then probed with appropriate antibodies followed by incubation with appropriate infrared-tagged secondary antibodies and finally was scanned with an Odyssey Infrared scanner (Li-Cor Biosciences, Lincoln, NE). Densitometric analysis was performed with the Odyssey scanning software.
Reporter assay was essentially performed as described previously [55]. Briefly, the cells were transiently transfected with the combined plasmids as indicated. The differences in the amounts of total DNA were normalized with vector control to reach the same amount of total transfected DNA. At 24 h post-transfection, cells were harvested and lysed in reporter lysis buffer (Promega Inc., Madison, WI). A 40 µl aliquot of the lysate was transferred to a 96-well plate. Luciferase activity was measured using an LMaxII384 luminometer (Molecular Devices, Sunnyvale, CA) by automatically injecting 25 µl of luciferase substrate into each well and integrating the luminescence for 20 s post-injection. The results represent experiments performed in duplicate.
The chromatin immunoprecipitation (ChIP) experiments were done essentially as previously described with some modifications. BC3 cells (10×106) with LANA or control knockdown were cross-linked with 1.1% (v/v) formaldehyde, 100 mM NaCl, 0.5 mM EGTA, and 50 mM Tris-HCl (pH 8.0) in growth medium at 37°C for 10 min, then at 4°C for 50 min. Formaldehyde was quenched by adding 0.05 vol 2.5 M glycine. Fixed cells were washed with PBS, incubated for 15 min in 15 ml of 10 mM Tris-HCl (pH 8.0), 10 mM EDTA, 0.5 mM EGTA, and 0.25% (v/v) Triton X-100, followed by 15 min in 15 ml of 10 mM Tris-HCl (pH 8.0), 1 mM EDTA, 0.5 mM EGTA, and 200 mM NaCl, and finally sonicated in 1 ml of 10 mM Tris-HCl (pH 8.0), 1 mM EDTA, 0.5 mM EGTA, 1% (w/v) SDS plus 1 mM PMSF, 1 µg/ml aprotonin, leupettin, and pestatin to an average fragment size of 300–500 bp. 20% of solubilized chromatin extracts were saved for input followed with cross-link reverse step, and the remaining were clarified by centrifugation at 12,000 g, and diluted to 6 OD260 U/ml in IP buffer [140 mM NaCl, 1% (w/v) Triton X-100, 0.1% (w/v) sodium deoxycholate, 1 mM PMSF, 100 µg/ml salmon sperm DNA, and 100 µg/ml BSA]; preincubated for 1 h at 4°C with 10 µl/ml 50% (v/v) Protein A-agarose (Invitrogen Life Technologies, Camarillo, CA) with normal mouse/rabbit sera; reconstituted in PBS, and washed several times in IP buffer. Aliquots (600 µl) were incubated with 20 µg of each specific antibody for overnight at 4°C. Immune complexes were separated into bound and unbound complexes with protein A-agarose and cross-links were reversed by treatment at 65°C overnight. After treatment with RNase A and proteinase K, samples were extracted once with phenol/chloroform, and the DNA was precipitated with 2 volumes of ethanol. Precipitated DNA was pelleted, washed once with 70% ethanol, dried, and resuspended in 100 µl of water. The DNA was analyzed by quantitative PCR with Aurora promoter primers (sense: 5′- TTCGATCGACCAGCTGGTCC GGTTCT -3′, anti-sense: 5′- TTCTCGAGCACTTGCTCCCTAAGAAC -3′).
MEF cells (15×106) were transfected by electroporation with appropriate combination of plasmids expressing HA-Ub (5 µg), p53-FLAG (or HA-p53) (8 µg), Aurora A-myc (8 µg) and LANA-myc (8 µg). Cells were incubated for 36 hr and pretreated for an additional 6 hr with 20 µM MG132 (Biomol International) before harvesting. The extent of ubiquitylation of p53 proteins was determined by western blot analysis using the FLAG-specific antibody (M2) or p53 (Do-1) antibody.
Ten million of Saos-2 cells were typically transfected using electroporation with different combinations of expression plasmids as shown in the text. At 48 hr posttransfection, 1×104 transfected cells were cultured in the selection medium (DMEM supplemented with 2 µg/ml puromycin). After 14 days, cells were fixed on the plates with formaldehyde and stained with 0.1% crystal violet. The amount of the colonies in each dish was scanned by Li-Cor Odyseesy and counted.
Cells were harvested and washed with ice-cold PBS and fixed in 70% cold ethanol overnight at 4°C. The fixed cells were then stained with PBS containing 40 µg/ml of propidium iodide (Sigma, St Louis, MO), 200 µg/ml of RNase A (Sigma) and 0.05% Triton X-100 for 1 h at 4°C in dark. Different cell cycle phases cells were characterized by using a FACSCalibur (BD Biosciences, San Joe, CA) and the results were analyzed using the FlowJo software (Tree star, Ashland, OR). For cytometric assessment of Aurora A expression profile at G1, S and G2 phase, 10×106 transfected cells were fixed by 3% paraformaldelhyde for 20 min at room temperature, and washed with PBS and subsequently blocked in 1% BSA with 0.1% Triton X-100, followed by incubation with 1 µg goat anti-Aurora A antibody (Ark-1, N-20) or normal IgG serum for overnight at at 4°C. Cells were washed three times with blocking buffer and exposed to secondary chicken anti-goat antibody (1∶500 dilution) conjugated with Alexa Fluor 488 for 1 hr at room temperature, followed by three washes with PBS, and staining with propidium iodide and RNase A for cell cycle.
|
10.1371/journal.ppat.1004890 | TRAF1 Coordinates Polyubiquitin Signaling to Enhance Epstein-Barr Virus LMP1-Mediated Growth and Survival Pathway Activation | The Epstein-Barr virus (EBV) encoded oncoprotein Latent Membrane Protein 1 (LMP1) signals through two C-terminal tail domains to drive cell growth, survival and transformation. The LMP1 membrane-proximal TES1/CTAR1 domain recruits TRAFs to activate MAP kinase, non-canonical and canonical NF-kB pathways, and is critical for EBV-mediated B-cell transformation. TRAF1 is amongst the most highly TES1-induced target genes and is abundantly expressed in EBV-associated lymphoproliferative disorders. We found that TRAF1 expression enhanced LMP1 TES1 domain-mediated activation of the p38, JNK, ERK and canonical NF-kB pathways, but not non-canonical NF-kB pathway activity. To gain insights into how TRAF1 amplifies LMP1 TES1 MAP kinase and canonical NF-kB pathways, we performed proteomic analysis of TRAF1 complexes immuno-purified from cells uninduced or induced for LMP1 TES1 signaling. Unexpectedly, we found that LMP1 TES1 domain signaling induced an association between TRAF1 and the linear ubiquitin chain assembly complex (LUBAC), and stimulated linear (M1)-linked polyubiquitin chain attachment to TRAF1 complexes. LMP1 or TRAF1 complexes isolated from EBV-transformed lymphoblastoid B cell lines (LCLs) were highly modified by M1-linked polyubiqutin chains. The M1-ubiquitin binding proteins IKK-gamma/NEMO, A20 and ABIN1 each associate with TRAF1 in cells that express LMP1. TRAF2, but not the cIAP1 or cIAP2 ubiquitin ligases, plays a key role in LUBAC recruitment and M1-chain attachment to TRAF1 complexes, implicating the TRAF1:TRAF2 heterotrimer in LMP1 TES1-dependent LUBAC activation. Depletion of either TRAF1, or the LUBAC ubiquitin E3 ligase subunit HOIP, markedly impaired LCL growth. Likewise, LMP1 or TRAF1 complexes purified from LCLs were decorated by lysine 63 (K63)-linked polyubiqutin chains. LMP1 TES1 signaling induced K63-polyubiquitin chain attachment to TRAF1 complexes, and TRAF2 was identified as K63-Ub chain target. Co-localization of M1- and K63-linked polyubiquitin chains on LMP1 complexes may facilitate downstream canonical NF-kB pathway activation. Our results highlight LUBAC as a novel potential therapeutic target in EBV-associated lymphoproliferative disorders.
| The linear ubiquitin assembly complex (LUBAC) plays crucial roles in immune receptor-mediated NF-kB and MAP kinase pathway activation. Comparatively little is known about the extent to which microbial pathogens use LUBAC to activate downstream pathways. We demonstrate that TRAF1 enhances EBV oncoprotein LMP1 TES1/CTAR1 domain mediated MAP kinase and canonical NF-kB activation. LMP1 TES1 signaling induces association between TRAF1 and LUBAC, and triggers M1-polyubiquitin chain attachment to TRAF1 complexes. TRAF1 and LMP1 complexes are decorated by M1-polyubiquitin chains in LCL extracts. TRAF2 plays a key role in LMP1-induced LUBAC recruitment and M1-chain attachment to TRAF1 complexes. TRAF1 and LMP1 complexes are modified by lysine 63-linked polyubiquitin chains in LCL extracts, and TRAF2 is a target of LMP1-induced K63-ubiquitin chain attachment. Thus, the TRAF1:TRAF2 heterotrimer may coordinate ubiquitin signaling downstream of TES1. Depletion of TRAF1 or the LUBAC subunit HOIP impairs LCL growth and survival. Thus, although TRAF1 is the only TRAF without a RING finger ubiquitin ligase domain, TRAF1 nonetheless has important roles in ubiqutin-mediated signal transduction downstream of LMP1. Our work suggests that LUBAC is important for EBV-driven B-cell proliferation, and suggests that LUBAC may be a novel therapeutic target in EBV-associated lymphoproliferative disorders.
| Epstein-Barr virus (EBV) is an oncogenic gamma-herpesvirus that is the causative agent of infectious mononucleosis. While EBV infection generally results in subclinical lifelong infection for most individuals, EBV is nonetheless associated with multiple human malignancies [1,2,3,4,5]. These include Hodgkin lymphoma, post-transplant lymphoproliferative disease (PTLD), and HIV-associated lymphomas. In these malignancies, the principal EBV oncoprotein, Latent Membrane Protein 1 (LMP1), is often expressed. LMP1 constitutively activates growth and survival pathways by mimicking CD40 signaling [6,7,8]. CD40 is a member of the tumor necrosis factor receptor (TNFR) family and serves as a key B-cell costimulatory molecule [9,10,11]. LMP1 expression transforms rodent fibroblasts and murine B-cells, and is necessary for EBV-mediated conversion of human B lymphocytes into immortalized lymphoblastoid cell lines (LCLs) [12,13,14,15,16,17].
LMP1 is comprised of a 24-residue N-terminal cytoplasmic tail, 6 transmembrane domains (TM), and a 200 residue C-terminal cytoplasmic tail. Deletion of the LMP1 N-terminus abrogates EBV-mediated B-cell transformation and alters LMP1 localization [18]. However, specific roles of the LMP1 N-terminus remain to be defined at the molecular level. The LMP1 TM domains drive assembly of LMP1 signalosome oligomers, which constitutively signal in a ligand independent manner from C-terminal tail domains [19,20,21,22,23]. The membrane proximal Transformation Effector Site (TES1)/C-terminal Activation Domain (CTAR1) spans residues 186–231. The TES1 P204QQAT210 motif binds directly to the TRAF domain of TNF receptor associated factor 2 (TRAF2), and likely also to conserved residues in the TRAFs 1, 3, and 5 domains [24]. The LMP1 PQQAT motif is necessary for TES1/CTAR1-medaited MAP kinase, canonical and non-canonical NF-kB pathway activation [24,25,26,27,28,29,30]. LMP1 TES1 activates additional pathways, including PI3K [31]. The LMP1 TES2/CTAR2 domain spans residues 351–386 and uses TRAF6 to further activate canonical NF-kB, MAP kinase, and IRF7 pathways [32,33,34]. The LMP1 CTAR3 domain, located between residues 231 and 350, associates with UBC9 and contributes to LMP1-mediated cellular migration [35]. The composition of TRAF complexes in LMP1-expressing cells has yet to be fully defined, and important components have recently been described [36,37,38,39].
LMP1 TES1 is critical for primary B lymphocyte growth transformation, since recombinant EBV that lacks LMP1 residues 185–211 does not initiate LCL outgrowth in tissue culture [25,40]. By contrast, the N-terminal 231 LMP1 residues support EBV-mediated B-cell outgrowth for up to five weeks in culture [41], and long-term on fibroblast feeder layers [40]. Interestingly, while LMP1 regulates the expression of a wide-array of host cell genes [42,43,44,45], a subset are uniquely induced by TES1 signaling. Notably, TRAF1 is amongst the earliest and most highly up-regulated LMP1 B-cell targets [29,46]. TRAF1 is abundantly expressed in EBV-infected immunoblasts in patients with infectious mononucleosis [47], and is also highly expressed in EBV-associated PTLD and Hodgkin lymphoma, where TRAF1 serves as an important biomarker [47,48,49]. TRAF1 expression is higher in EBV-positive Hodgkin lymphoma than in EBV-negative tumor samples [49]. Some nasopharyngeal carcinomas express LMP1 and TRAF1 [50]. In LCLs, most TRAF1 is associated with LMP1, either as TRAF1 homotrimers, or as TRAF1:TRAF2 heterotrimers [29]. LMP1 and TRAF1 co-localize in acquired immunodeficiency syndrome (AIDS)-associated lymphoma, PTLD, and Hodgkin lymphoma samples [51]. Despite these intriguing associations, little is known about the extent to which TRAF1 plays a pathogenic role in EBV-associated malignancy.
LMP1 TES1-mediated activation of the JNK/AP-1 axis is critically dependent on both TRAF1 and TRAF2 [52]. Although TRAF1 is the only TRAF family member that is not equipped with an N-terminal RING finger domain, TRAF1 is nonetheless the only TRAF that co-activates TES1-mediated NF-kB and JNK pathway activation [29,52]. Likewise, TRAF1 enhances signaling from the TNF receptor family member 4-1BB, and is important for CD8+ T-cell responses during chronic viral infection [53,54]. The mechanism by which TRAF1 enhances LMP1 signaling remains incompletely understood.
The role of TRAF proteins in NF-kB and MAP kinase pathway activation is perhaps best understood in the context of TNF receptor signaling. Upon TNF stimulation, TNF receptor 1 binds TRADD, which in turn recruits TRAF2, cIAP1 and cIAP2, and subsequently the linear ubiquitin assembly complex (LUBAC) [55]. LUBAC is comprised of two ring-in-between-ring E3 ligases subunits, HOIP and HOIL-1L, and the scaffold protein SHARPIN. LUBAC catalyzes a peptide bond between the N-terminal methionine alpha-amino group of one ubiquitin molecule and the C-terminal glycine of another ubiquitin molecule [56]. Linear (Met1 or M1) linked poly-Ub (pUb) chains stabilize the TNFR1 complex and enable recruitment of downstream activators [57,58]. LUBAC is also recruited to CD40 in a TRAF2-dependent manner, and SHARPIN deficiency impairs CD40-mediated NF-kB activation [55,59,60]. TNFR1 induces linear ubiquitination of RIP1 and IKK-gamma, and CD40 signaling also induces linear ubiquitination of IKK-gamma [55]. Interestingly, the IKK-gamma UBAN domain strongly associates with M1 chains, and thereby recruits the IKK-alpha and IKK-beta kinases to activated receptors to activate canonical NF-kB [61,62,63]. The extent to which M1-linked pUb chains participate in LMP1 signaling remains unknown.
To gain insight into the molecular mechanism by which TRAF1 functions downstream of LMP1, we performed proteomic analysis of immuno-purified TRAF1 complexes from cells uninduced or induced for LMP1 1–231 expression. We identified LUBAC components as high-confidence TRAF1 interactors in cells that express LMP1 residues 1–231, and found that TRAF1 and LMP1 complexes are highly M1-pUb linked in LCLs. LMP1 and TRAF1 complexes immuno-purified from LCL extracts were each K63-pUb chain modified.
Human embryonic kidney (HEK)-293 cells were obtained from Elliott Kieff (Brigham and Women’s Hospital, Boston, MA). 293 cells with inducible LMP1 1–231 expression were constructed, using a tightly regulated Tet-on inducible system that was previously described [43]. Briefly, the inducible system for LMP1 1–231 expression consisted of three parts: (i) an untagged LMP1 1–231 cDNA with stop codon after residue 231, cloned into the tetracycline-regulated pTRE-tight vector (Clontech); (ii) a tetracycline suppressor (tTS) that binds Tet operator sites in the absence of tetracycline and silences expression; (iii) a reverse tetracycline transactivator fused to the 4-hydroxy tamoxifen (4HT) ligand-binding domain (rTTA M2). LMP1 expression was induced by addition of doxycycline (1ug/ml) and 4HT (100 nM). A clone that had undetectable LMP1 1–231 expression at baseline, and inducible LMP 1–231 expression upon doxycycline and 4HT addition, was selected. Conditional LMP1 1–231 cell lines with stable N-terminal FLAG epitope-tagged TRAF1, TRAF2, TRAF3, or green fluorescence protein (GFP) expression were established by murine stem cell leukemia virus (MSCV) transduction and puromycin selection, as previously described [64–65]. GM12878 cells were provided by Elliott Kieff. MSCV transduction was also used to derive GM12878 LCLs with stably expressed N-terminally epitope-tagged TRAF1, GFP, SHARPIN, IKK-gamma or IKK-epsilon. LCLs that express N-terminal FLAG-tagged LMP1 at physiological levels, in place of untagged LMP1, were previously described [66], and were generously provided by Elliott Kieff. Briefly, a FLAG-tagged LMP1 cDNA was recombined into the P3HR-1 genome LMP1 locus by second site homologous recombination. 293 cell lines were cultured in DMEM with 10% tetracycline-free fetal calf serum (FCS), LCLs and EBV-negative BL2 Burkitt lymphoma B-cells (kindly provided by Elliott Kieff) were cultured in RPMI with 10% FCS. All cell lines were grown in a humidified ThermoFisher incubator at 37 C, with 5% CO2.
The following vectors were used in transient transfection assays: empty pSG5, pSG5 vectors with N-terminally FLAG-tagged GFP, TRAF1, TRAF2, or TRAF3 cDNAs; pSG5 expression vectors with untagged or N-terminally epitope-tagged full length LMP1 (wildtype), LMP1 residues 1–231, C-terminally HA-tagged LMP1 1–231; the TES2 null mutant 384YYD386 ->384ID385, the TES1 null alanine point mutant LMP1 204PQQAT208-> 204AQAAA208, or the TES1/TES2 null double mutant (LMP1 DM) containing both of these mutations [25,67]; a PGK-puro vector with untagged TRAF1 cDNA. N-terminally GST-tagged TRAF1 was cloned into a modified Gateway-compatible pSG5 expression vector by Gateway cloning.
For western blot analysis, antibodies against the following were used: Cell Signaling Technologies TRAF3 (#4729), cIAP1 (#7065), cIAP2 (# 3130), phospho-JNK (#9251), phospho-p38 (#9211), total JNK (#9252), total p38 (#9212), phospho-ERK (# 4377), total ERK (# 9102), RelA-phosphoserine 536 (3033); total RelA (8242) total Ub (3936), ABIN1 (4664), A20 (cat #4625); Bethyl Laboratories SHARPIN (#A303-559A), TRAF2 (#A303-460A), HOIP (A303-560A); Sigma Aldrich tubulin (# T5168) and FLAG M2; Santa Cruz TRAF1 (#sc1831) and NEMO (sc-8330); Covance HA.11; EMD Millipore p100/p52 (05–361) and anti-K63 05–1308; Chemicon/Millipore GAPDH (MAB375); Genentech anti-M1 Ub 1F11/3F5/Y102L. Detection of endogenous phospho-TAK1 was performed on inducible 1–231 LMP1 293 TRAF1 cells uninduced or induced for LMP1 expression overnight, and treated with 50 nM Calyculin A (Cell Signaling #9902) for 5 minutes prior to harvest. Cell Signaling antibodies against phospho-TAK1 (#4508) and total TAK1 (#4505) were used. LMP1 monoclonal antibodies OT22CN against the LMP1 N-terminus, or S12 against the LMP1 C-terminus (recognizes an epitope between TES1 and TES2), were used. Cell Signaling HRP-tagged secondary antibodies, or with Rockland TrueBlot HRP-tagged secondary antibodies were used for western blot.
All immune-purified and whole cell lysate samples were boiled for 5 minutes in Laemmli SDS-loading buffer with a final concentration of 1% SDS and 5% beta-mercaptoethanol, at 95°C for 5 min. SDS/PAGE was performed on Bio-Rad precast gels. 4–20% gradient gels were used for experiments with anti-ubiquitin western blots. Proteins were transferred to nitrocellulose filters for 1 hour at 100V using Bio-rad a power pack and minigel transfer apparatus. Blots were blocked with 5% non-fat dry milk for 30 minutes, then probed overnight with primary antibody at 4 degrees C, washed in TBST for 5 minutes x 4 cycles, incubated with secondary antibody for 1 hour, washed in TBST for 5 minutes x 4 cycles, developed with Western Lightening ECL developer, and imaged on a Carestream Molecular Imaging workstation. Where indicated, western blot band intensities were measured using Carestream software, using background subtracted net values.
M1-pUb and K63-pUb linked chains were purified under denaturing conditions, according to the manufacturer’s instructions. Briefly, for M1-pUb purification, cells were lysed in buffer containing 8M urea and 20mM Tris (pH 7.4), 135 mM NaCl, 1% Triton-X100, 10% glycerol, 1mM EDTA, and 1.5mM MgCl2, supplemented with Roche complete EDTA-free protease inhibitor tablet, 1 mM PMSF, 4mM 1 10 o-phenanthroline, sodium pyrophosphate, 10 mM-glycerophosphate, 2 mM sodium pyrophosphate, 1% aprotinin, and 2 mM N-ethylmaleimide, at RT for 10min. Insoluble debris was pelleted by 13,000 RPM microfuge, and urea concentration was then reduced to 7M by addition of lysis buffer. 2ug of M1-Ub antibody was added, and samples were rotated at RT overnight. Precipitate was pelleted by microcentrifuge, and then 20ul of Protein A sepharose beads (Invitrogen 101042) were added, and rotated for 2 hours at RT. Beads were washed 5X with 7M urea lysis buffer, with inhibitors. Western blots were preformed according to Genentech instructions, using wet transfer at 30V for 2 hours to nitrocellulose membranes. Primary antibody was added to blots for 1 hr at RT.
NF-kB activity was measured by a GFP reporter assay, as previously described [38]. Briefly, conditional LMP1 1–231 and LMP1-231 TRAF1 293 cells with a stably integrated NF-kB GFP reporter were used. LMP1 1–231 expression was induced for 20 hours by the addition of doxycycline and 4HT. NF-kB GFP reporter values were measured on a FACScalibur flow cytometer (BD Biosciences), and analyzed by Cell Quest software (BD Biosciences).
The SMAC mimetic TL-32711 was obtained from Active Biochem (#A-1901), and used according to the manufacturer’s instructions at a concentration of 20 uM.
293 inducible LMP1 cells were treated with Dharmacon/Thermo Fisher siRNAs for 72 hours prior to LMP1 induction, as previously described [38]. Briefly, LMP1 1–231 conditional 293 TRAF1 cells were reverse transfected with Dharmafect I lipid in 12-well plates. 72 hours later, LMP1 1–231 expression was induced by addition of 4HT and doxycycline, where indicated, for 16 hours. The non-targeting siRNA control (Catalogue # D-001810-10-20), and siGenome siRNAs against TRAF2, TRAF3, RNF31/HOIP, RBCK1/HOIL-1L and SHARPIN were used at a final concentration of 50 nM per siRNA pool. For GM12878 shRNA analysis, LCLs were transduced with VSV-G pseudotyped lentiviral vectors from the Broad Institute of Harvard and MIT RNAi consortium on day 0 and 1. Anti-GFP shRNA was used as a control. On day 2, LCLs were selected with puromycin (3 ug/ml), and analyzed at the indicated timepoints. Knockdowns were validated by western blot and qPCR analysis. All shRNA sequences are available upon request.
GM12878 LCLs with stable S. pyogenes Cas9 expression were established by infection by lentiviral transduction and blasticidin selection, using pLentiCas9-Blast (Addgene plasmid # 52962). We verified that Cas9 was highly active in the selected LCL pool by transduction with a lentivirus that encodes GFP, and a sgRNA against GFP [68]. The PXPR-011 plasmid was kindly provided by John Doench, Broad Institute, and encodes GFP, as well as an sgRNA against GFP. PXPR-011 is therefore a convenient way to monitor Cas9 activity in cell lines. GM12878 cells transduced with PXPR-011 based lentivirus and selected with puromycin initially expressed GPF, but were then found to lose GFP expression in >85% of transduced cells (the residual 15% of cells that continue to express GFP despite sgRNA against GFP may be cells where the non-homologous end-joining pathway correctly repaired the Cas9-induced DNA double strand break) [68]. By contrast, nearly 100% of Cas9 negative GM12878 cells were GFP positive after transduction with the same lentivirus and puromycin selection. CRISPR single guide RNAs (sgRNA) targeting human RNF31 (which encodes HOIP) were designed using the online program CRISPRdirect (http://crispr.dbcls.jp/)[69], and the oligo GCCCTCAGCGGCCTCGGTAC was Synthesized by Life Technologies, cloned into the lentiGuide-Puro vector (Addgene plasmid # 52963), according to the protocol from the Zhang laboratory website (http://genome-engineering.org/)[70], and sequence verified. Lentiviruses encoding the HOIP sgRNA were constructed and used to transduce GM12878 Cas9+ cells. Transduced cells were selected by purmoycin. HOIP depletion efficiency was validated by western blot.
293 cell lines were transiently reverse transfected as previously described [38], using Effectene lipid (Qiagen). For most experiments, cells were transfected for 18 hours.
293 cells were transiently transfected with the indicated plasmids (pSG5 LMP1, pSG5 FLAG-TRAF1, and/or the UBAN-GFP sensor [71,72]). UBAN-GFP is a fusion between the conserved linear Ubiquitin Binding domain of ABIN1 and NEMO/IKK-gamma and GFP. The UBAN-GFP biosensor has been validated to be highly specific for M1-pUb chains in vitro and in vivo [71,72]. 20 hours after transfection, cells were fixed, permeabilized, and stained where indicated with antibodies against LMP1 or TRAF1 (Santa Cruz, rabbit polyclonal). Secondary antibodies used were Alexa-561-conjugated anti-mouse and Alexa-633-anti-rabbit (both from Life Technologies). Cells were analyzed by confocal microscopy, and images were processed with Fiji (http://wiki.imagej.net/Fiji).
N-terminally GST-tagged TRAF1 expression vectors were constructed using Gateway cloning, and used for purification of recombinant TRAF1 from unstimulated HEK-293 cells. In vitro ubiquitin assays were performed as previously described [73]. Briefly, in vitro ubiquitination assays were performed according to the manufacturer’s protocol (Boston Biochem). Ubiquitin (5 μg), the E1 enzyme (200 ng), UBE2L3 (300 ng) (Boston Biochem), the indicated LUBAC components (0.8 μg) and TRAF1-GST(2μg) were co-incubated with 2 mM ATP (Sigma) at 37°C 2 hours, in ubiquitin assay buffer (20 mM Tris-HCl pH7.5, 5 mM MgCl2, 2 mM DTT). 1x stop solution (Boston Biochem) was added to terminate the reaction. Following GST pull-down, beads were washed four times, and then boiled in Laemmli SDS-loading buffer with 5% beta-mercaptoethanol at 95°C for 5 min. The samples were subsequently analyzed by SDS-PAGE followed by Western blotting using a PVDF membrane.
FLAG affinity purification and liquid chromatography-mass spectrometry analysis were performed, as previously described[74]. Seven 15 cm^2 dishes (approximately 100 million cells) of 293 TRAF1 or GFP control cells, either uninduced or induced for LMP1 1–231 expression for 16 hours, were washed twice with PBS and then lysed on ice for 30 minutes in lysis buffer with protease inhibitors (Roche EDTA Free Complete, Cat #11836145001, 1% aprotinin (Sigma Cat #A6279), 1 mM PMSF (Sigma), and 4 mM 1,10 o-phenanthroline (Sigma), and the phosphatase inhibitors 10 mM beta-glycerophosphate and 2 mM sodium pyrophosphate (Sigma) [64]. 30 uL of packed anti-FLAG beads (Sigma Cat #A2220) were added into the lysates and were rotated at 4 degrees C for 4 hours, then washed 5 times in lysis buffer with protease and phosphatase inhibitors, with Eppendorf tube change prior to the last wash, and eluted with 50 ul of 3X-FLAG peptide (0.5 mg/ml, Sigma #F4799) at room temperature for 30 minutes, three times sequentially. Samples were analyzed by WB to confirm absence of antibody heavy/light chain contamination, and then run into a 10% pre-cast mini-gel (Bio-Rad) for 1 cm, cut into two equal slices, and sent for liquid chromatography mass spectrometry (LC/MS-MS) analysis at the Harvard Taplin Biological Mass Spectrometry Facility (Harvard Medical School). Gel slices were processed by the Taplin Proteomics facility staff. Briefly, gel slices were subjected to a modified in-gel trypsin digestion procedure. Gel pieces were washed and then dehydrated with acetonitrile for 10 min. Following acetonitrile removal, slices were speed-vac dried, rehydrated with a 50 mM ammonium bicarbonate solution containing 12.5 ng/μl modified sequencing-grade trypsin (Promega, Madison, WI) at 4°C. After 45 min., the excess trypsin solution was removed and replaced with 50 mM ammonium bicarbonate solution to just cover the gel pieces. Peptides were then extracted by removing the ammonium bicarbonate solution, followed by one wash with a solution containing 50% acetonitrile and 1% formic acid. Extracts were speed-vac dried for ~1 hr and stored at 4°C until analysis. On the day of analysis, samples were reconstituted in 5–10 μl of HPLC solvent A (2.5% acetonitrile, 0.1% formic acid), subjected to nano-scale reverse-phase HPLC using a capillary column (5 μm C18 spherical silica beads packed into a fused silica capillary (100 μm inner diameter x ~12 cm length) with a flame-drawn tip. After equilibrating the column, each sample was loaded via a Famos auto sampler (LC Packings) onto the column. A gradient was formed and peptides were eluted with increasing concentrations of solvent B (97.5% acetonitrile, 0.1% formic acid). Upon elution, peptides were subjected to electrospray ionization and analyzed by a LTQ Velos ion-trap mass spectrometer (ThermoFisher, San Jose, CA). Peptides were detected, isolated, and fragmented to produce a tandem mass spectrum of specific fragment ions for each peptide. Dynamic exclusion was enabled such that ions were excluded from reanalysis for 30 s. Peptide sequences (and hence protein identity) were determined by matching protein databases, using Sequest (ThermoFisher). The human IPI database (Ver. 3.6) was used for searching. Precursor mass tolerance was set to +/- 2.0 Da and MS/MS tolerance was set to 1.0 Da. A reversed-sequence database was used to set the peptide false discovery rate at 1%. Filtering was performed using the Sequest primary score, Xcorr and delta-Corr. Spectral matches were further manually examined.
To assign statistical significance and to identify high-confidence TRAF1 interacting proteins, our data were compared with a publicly-available database of 30 negative control FLAG-tagged baits, purified from HEK-293 stable cell lines under similar conditions (http://www.crapome.org) [75]. Control dataset IDs and bait peptide counts are also provided for comparison in S1 Table. The SAINT algorithm (http://sourceforge.net/projects/saint-apms) was used to evaluate the MS data [76,77]. SAINT is designed for AP-MS analysis and has been validated in analysis of several protein interactomes [76,78,79]. The default SAINT options were low Mode = 1, min Fold = 0, norm = 0. SAINT probabilities computed independently for each biological replicate were averaged (AvgP) and reported as the final SAINT score. Fold change was calculated for each prey protein as the ratio of average spectral counts from replicate bait purifications over the average spectral counts across all negative controls (total peptide spectral counts were summed for each protein). A background factor of 0.1 was added to the average spectral counts of negative controls to prevent division by zero. The highest number of spectral counts for each protein were selected to establish the negative control database for SAINT analysis. Selection of the threshold for SAINT scores was based on receiver operating curve analysis performed using publicly available protein interaction data and the FLAG AP-MS data set as a list of true positive interactions. A SAINT score of AvgP ≥ 0.80 was considered a high-confidence interacting protein, with an estimated FDR of ≤1%.
Real-time reverse transcription-PCR (qPCR). qPCR was performed on a Bio-Rad CFX Connect Real-time system, using the Power SYBR green RNA-to-CT 1-step kit (Applied Biosystems), for 40 cycles. Fold changes were determined using the CT method and normalized by 18S rRNA expression levels. The RBCK1 primers 5’-TGCAAGACCCCAGATTGCA-3’, AND 5’-ACAGGGCAGGTGAACTCATTG-3’ were used.
96 hours after initial shRNA transduction (and 48 hours after puromycin selection), LCLs were plated at a density of 300,000 cells/ml in 96 well plates in 100 uL of RPMI/FCS, in triplicate. Cells were fed 100 uL of RPMI at 48 hours and 96 hours thereafter. Relative live cell numbers were then accurately quantitated by the CellTiter-Glo luminescent cell viability assay (Promega). All values obtained were within the linear range of the instrument. For shRNA growth curves, a LMaxII instrument was used (Molecular Devices). For CRISPR/cas9 growth curve analysis, SpectraMax L (Molecular Devices) was used, as the LMaxII was no longer available.
All bar-graphs and growth curves were produced using GraphPad software.
To characterize signaling by the LMP1 TES1 domain, 293 cells with conditional LMP1 1–231 expression were derived (described in detail in the Methods) [38,43]. We then established conditional LMP1 1–231 293 cell lines with stable expression of N-terminally FLAG- tagged GFP or TRAF1. FLAG-TRAF1 was expressed at LCL physiological levels (S1 Fig). The conditional 293 LMP1 cell pair provided an isogenic background with which to compare the effects of TRAF1 on LMP1 TES1 domain-mediated pathway activation. Consistent with prior LMP1 studies [29,52], TRAF1 co-expression markedly boosted LMP1 TES1-mediated JNK pathway activation (Fig 1A and 1B). The ratio of phospho-JNK to total JNK in whole cell extracts increased from 2.1-fold in 293 cells to 7.1-fold in 293 TRAF1 cells 16 hours after LMP1 1–231 induction, as judged by western blot analyses from three independent experiments. We likewise found that TRAF1 co-expression significantly increased LMP1 1-231-mediated p38 and ERK phosphorylation (Fig 1A and 1B). Our results suggest that TRAF1 enhances LMP1 TES1 domain-mediated activation at a level upstream of the three MAP kinases.
Consistent with a prior report, we found that TRAF1 co-expression also significantly enhanced LMP1 TES1-mediated NF-kB activation (Fig 1C) [29]. We next examined whether TRAF1 co-expression affected LMP1 1-231-mediated canonical and/or non-canonical NF-kB pathway activation. TRAF1 co-expression significantly up-regulated LMP1 TES1-mediated canonical NF-kB, as judged by RelA serine 536 phosphorylation, a commonly used marker of canonical NF-kB activity (Fig 1A and 1B). To examine the effect of TRAF1 co-expression on LMP1 1-231-mediated non-canonical NF-kB activation, we analyzed the p100:p52 ratio in conditional 293 and 293 TRAF1 cells. Non-canonical NF-kB activity triggers proteasomal processing of p100 to p52. Interestingly, TRAF1 co-expression did not significantly enhance LMP1 1-231-mediated non-canonical NF-kB pathway activity, as judged by the ratio of p100:p52, which remained 1.5 in both conditions (Fig 1D). Since the kinase TAK1 plays key roles in LMP1 MAP kinase and canonical NF-kB pathways, we next tested whether TRAF1 enhanced LMP1 1-231-mediated TAK1 activation. Indeed, TRAF1 markedly enhanced LMP1 1–231 induction of TAK1 activation loop serine 187 phosphorylation (S2 Fig). Collectively, our results suggest that TRAF1 enhanced LMP1 1-231-mediated MAP kinase and canonical NF-kB pathway activation at or above the level of TAK1 activation.
To gain insights into TRAF1 effects on LMP1 TES1 signaling, we used affinity purification and mass spectrometry analysis (AP-MS) of TRAF1 complexes as a discovery tool for subsequent analysis. FLAG-TRAF1 complexes were immune-purified from conditional 293 cells that were uninduced or induced for LMP1 1–231 expression for 16 hours [64]. Complexes were eluted from agarose beads by co-incubation with FLAG peptide (see Methods section for details). As a negative control, FLAG-GFP was immuno-purified from conditional LMP1 1–231 cells induced for 16 hours. Independent FLAG-purifications were analyzed by liquid chromatography-mass spectrometry/mass spectrometry (LC-MS/MS) proteomic analysis for each condition. Data resulting from AP-MS analysis are presented in S1 Table.
To identify high-confidence interactions in our TRAF1 datasets, we further compared our datasets with thirty publically-available 293 cell FLAG AP/MS control datasets (http://www.crapome.org)[75]. This analysis provides additional statistical power to remove common 293 cell contaminants, which are frequently high abundance proteins including heat-shock, cytoskeletal, histones, ribonucleoproteins, and ribosomal proteins. We used the well-established ‘Significance Analysis of Interactome' (SAINT) computational algorithm to assign confidence scores to our TRAF1 datasets [76,77,80]. SAINT uses quantitative AP-MS data to derive the probability of a bona fide protein-protein interaction. At a FDR < 1% cutoff (SAINT score ≥0.8), we identified 19 high-confidence TRAF1 interactors in extracts from LMP1 1–231 expressing 293 cells. Three additional proteins had a SAINT score of 0.79 and were also considered high-confidence interactors. 23 high-confidence TRAF1 interactors were identified from uninduced 293 cells purifications (S3 Fig and S1 Table). Well-characterized TRAF1 interactors were enriched in both TRAF1 datasets, including TRAF2, cIAP1, cIAP2, TBK1, and TANK. Other TRAF1 high confidence interactors were identified in either the LMP1 1–231 uninduced or induced condition. Interestingly, all seven of the LMP1 1-231-induced TRAF1 high-confidence protein interactions have established roles in ubiquitin biology, including ubiquitin itself. Notably, LMP1 1–231 expression induced association between TRAF1 and the LUBAC catalytic subunits HOIP (SAINT score 0 without induction, 0.99 with induction) and HOIL-1L (SAINT score 0 without induction, 0.79 with induction). By comparison, HOIP or HOIL-1L peptides were not retrieved in our FLAG-GFP samples or in any of the 30 control 293 cell FLAG runs (S1 Table). LMP1 1–231 expression also induced TRAF1 association with the ubiquitin editor protein A20 (SAINT score 0 without induction, 0.96 with induction). A20 contains deubiquitinase, ubiquitin ligase, and ubiquitin-binding zinc finger domains, the latter of which bind to both M1- and K63-pUb chains. The K63-polyubiquitin binding protein SQSTM1/P62 was also identified as a high-confidence LMP1-induced TRAF1 interactor (uninduced SAINT score 0, induced score 0.79). The K63- and M1-pUb sensor ABIN1 nearly reached significance (SAINT score without induction, 0.49 with induction).
The LMP1 1-231-induced association between TRAF1 and LUBAC was further validated by IP/western blot analysis, using both conditional 293 cells and GM12878 LCLs with stably FLAG-TRAF1 expression at physiological levels (S1 Fig). First, FLAG-TRAF1 complexes were immuno-purified from conditional 293 cells, either undincued or induced for LMP1 1–231 expression for 16 hours (Fig 2A). LMP1 induction increased the level of HOIP and SHARIPN in FLAG-TRAF1 complexes. By contrast, we did not observe significant co-purification of LUBAC components with FLAG-TRAF3 complexes immuno-purified from conditional LMP1 1–231 293 cells with stable FLAG-TRAF3 expression. Of note, twice as many FLAG-TRAF3 cells were used in this experiment, to achieve similar levels of immuno-purified FLAG-TRAF1 and FLAG-TRAF3. We were unable to find a suitable antibody for analysis of endogenous HOIL-1L. Also of note, LMP1 1–231 induction caused TRAF1 and TRAF3 steady state levels to decrease, perhaps as a result of increased turnover. To validate that TRAF1 associates with LUBAC in LCL extracts, we tested whether FLAG-TRAF1 purified from GM12878 cells also retrieved LUBAC components. Both HOIP and SHARPIN co-immunoprecipitated with FLAG-TRAF1, but not with a FLAG-GFP control (Fig 2B). Likewise, HA-SHARPIN complexes immuno-purified from GM12878 cells with stable HA-SHARPIN expression co-immunoprecipitated TRAF1 and LMP1 (S4 Fig). Taken together, our data suggest that TRAF1 and LUBAC are present together in protein-protein complexes in LMP1+ cells.
Most TRAF1 in LCLs is associated with LMP1 [29]. We therefore investigated whether LMP1 complexes are modified by M1-linked pUb chains in LCL extracts. FLAG-LMP1 was immuno-purified from LCLs established from recombinant EBV, in which FLAG-tagged LMP1 is expressed at physiological levels from the EBV genome [18]. As a negative control with physiologic LMP1 expression from the EBV genome, we used GM12878 LCLs (where LMP1 is untagged). FLAG immuno-purified material was subjected to western blot analysis, using a M1-linked pUb (M1-pUb) chain specific monoclonal antibody [81]. M1-pUb chains were readily detected in FLAG-LMP1 purified from FLAG-LMP1 LCL extracts, but not from GM12878 extracts, suggesting that either LMP1, or an LMP1 signalosome protein, was modified by LUBAC-catalyzed M1-pUb chains (Fig 3A).
We next investigated whether TRAF1 complexes purified from GM12878 LCLs were likewise decorated with M1-pUb chains. As a negative control, we used GM12878 that express N-terminally FLAG-tagged GFP at similar levels. As shown in Fig 3B, FLAG-immuno-purified TRAF1 complexes, but not FLAG-GFP complexes, were highly modified by M1-pUb chains. Thus, LMP1 and TRAF1 are each present in complexes that are highly modified by M1-pUb chains.
To determine whether LMP1 induces M1-pUb chain attachment to TRAF1 complexes, we used 293 cell transient transfection assays. 293 cells were co-transfected for 24 hours with N-terminally FLAG-tagged GFP, TRAF1 or TRAF2 and either empty vector or untagged wildtype LMP1. We used the transfection system rather than the conditional system for this analysis to achieve similar FLAG-tagged protein expression levels across all three baits. FLAG immuno-purified complexes were analyzed by western blot for M1-pUb chains. Interestingly, FLAG-TRAF1 complexes, retrieved from LMP1 co-transfected cells, were highly modified by M1-pUb chains (Fig 3C). By contrast, background levels of M1-pUb chains were observed in FLAG-GFP and FLAG-TRAF2 pulldowns, and in FLAG-TRAF1 pulldown from cells without LMP1. This result suggests that LMP1 stimulates M1-pUb chain attachment to TRAF1, or a TRAF1-associated protein, and that in the absence of TRAF1, LMP1 does not induce M1-pUb attachment to TRAF2 complexes.
To next determine whether HOIP is important for M1-pUB chain attachment to LMP1 complexes in LCLs, we depleted HOIP from FLAG-LMP1 LCLs, using two independent shRNAs. Each anti-HOIP shRNA, but not the anti-GFP shRNA control, markedly reduced M1-pUb chain decoration of FLAG-LMP1 complexes, suggesting that HOIP plays a key non-redundant role in M1-chain attachment to LCL TRAF1 complexes (Fig 3D). Finally, we investigated whether LMP1 1–231 expression stimulates LUBAC activity. Whole cell lysates from 293 TRAF1 conditional cells that were uninduced, or induced for LMP1 1–231 for 16 hours, were analyzed for M1-pUb chain content by western blot (Fig 3E). Extracts from induced cells had abundant immune-reactive material. Of note, similar M1-pUb chain formation in 293 cells was previously demonstrated by HOIP/HOIL-1L transient transfection [81].
To further establish that LMP1 and TRAF1 complexes are decorated by M1-pUb chains in vivo, we next tested whether LMP1 and TRAF1 co-localize with a M1-pUb chain biosensor. The biosensor is comprised of a fusion protein between GFP and the Ubiquitin Binding of ABIN1 and NEMO/IKK-gamma (UBAN) domain. The UBAN-GFP biosensor selectively visualizes the localization of M1-pUb chains in mammalian cells activated by multiple independent stimuli [71,72]. 293 cells were transiently co-transfected with UBAN-GFP, TRAF1 and or/ LMP1. The characteristic 293 cell LMP1 punctate staining pattern was observed by confocal microscopy analysis of LMP1-transfected cells. By contrast, TRAF1 and UBAN-GFP exhibited diffuse cystosolic staining patterns in the absence of LMP1 co-expression (Fig 4A). Interestingly, LMP1 co-expression with TRAF1 and UBAN-GFP altered the TRAF1 and UBAN-GFP patterns, and induced marked co-localized of all three into punctate foci (Figs 4A and S5). We did not observe similar punctate foci of UBAN-GFP or co-localization with LMP1 in the absence of TRAF1 co-expression. Of note, TRAF1 and the UBAN-GFP sensor colocalized to a lesser extent, even in the absence of LMP1. Taken together with the proteomic and biochemical data presented above, these results are further suggest that in cells with LMP1 and TRAF1 co-expression, LMP1 and TRAF1 are present in complexes modified by M1-pUb chains.
We next tested whether TRAF1 and IKK-gamma also associate in GM12878 LCL extracts. Using GM12878 cells that stably express FLAG-tagged TRAF1 or FLAG-GFP as a control, we found that endogenous IKK-gamma co-immunoprecipitated with FLAG-TRAF1, but not with the FLAG-GFP negative control (S6 Fig). Likewise, using stable GM12878 cell lines, we found that endogenous TRAF1 reciprocally co-immunoprecipitated with HA-tagged IKK-gamma, but not with the HA-GFP negative control (S6 Fig). Of note, IKK-gamma expression in LCLs is significantly higher than in 293 cells, perhaps explaining why the 293 cell TRAF1 AP/MS analysis did not also identify IKK-gamma as a high confidence TRAF1 interactor, given the limit of detection of the assay. Collectively, these results are consistent with a model in which M1-pUb-linked pUb chains attached to TRAF1 complexes recruit the IKK-gamma.
We further validated the AP/MS result that LMP1 1–231 expression induced association between TRAF1 and the M1- and K63-pUb sensors A20 and ABIN1. A20 and ABIN1 are feedback regulators that each contain M1- and K63-pUb binding domains A20 down-modulates LMP1 TES1 signaling [82]. We found that immuno-purified FLAG-TRAF1 co-immunoprecipitated A20 and ABIN1 in 293 cells induced for LMP1 1–231 expression for 16 hours. By contrast, FLAG-TRAF2 complexes, purified from conditional LMP1 1–231 cells that stably express FLAG-TRAF2, did not co-immunoprecipitate A20 or ABIN1, even 16 hours after LMP1 1–231 induction (Fig 4B). Consistent with the higher A20 SAINT score in our TRAF1 AP/MS analysis, A20 association with TRAF1 complexes appeared to be more robust than that of ABIN1 by western blot analysis. Also of note, LMP1 expression increased A20 expression levels, as has previously been reported. Finally, endogenous A20 co-immunoprecipitated with FLAG-TRAF1, immuno-purified from GM12878 extracts (Fig 4C).
We previously found that HOIP and HOIL-1L are important for LMP1 TES2-mediated canonical NF-kB pathway activation in a genome-wide siRNA screen [38], suggesting that LMP1 TES2 may also activate LUBAC activity. To determine whether signaling by LMP1 TES2 also stimulates addition of M1-pUb chains to TRAF1 complexes, 293 cells were co-transfected with either wildtype (WT) LMP1, or with LMP1 mutants deficient for TES1, TES2 or TES1/TES2 signaling and with FLAG-tagged TRAF1. M1-pUb chains were detectable on FLAG-TRAF1 complexes immuno-purified from cells that co-expressed wildtype LMP1, or the LMP1 384ID385 mutant, which is null for TES2 signaling. By contrast, FLAG-TRAF1 complexes purified from cells that co-expressed either a LMP1 TES1 null alanine point mutant (LMP1 204PQQAT208-> 204AQAAA208) deficient for TRAF recruitment or a LMP1 double mutant (DM) deficient for TES1 and TES2 signaling, were not modified by M1-pUb chains (Fig 5A). These results suggest that association between TRAF1 and the LMP1 TES1 domain are required for M1-pUb chain attachment to TRAF1 complexes, and that LMP1 TES2-mediated NF-kB activation does not stimulate LUBAC to modify TRAF1.
We next determined whether LMP1 TES1 signaling was important for M1-pUb chain attachment to LMP1 complexes. 293 cells were co-transfected with FLAG-tagged WT, 1–231, or DM LMP1 vectors and with untagged TRAF1. FLAG-LMP1 immuno-purified complexes were analyzed by western blot for M1-pUb chain attachment (Fig 5B). M1-pUb chains decorated WT and LMP1 1–231 complexes, but not DM LMP1 complexes. These results again suggest that LMP1 complexes are modified by M1-pUb chains in cells with TRAF1 expression, and that LMP1 TES1 signaling is important for M1-pUb chain attachment to LMP1 complexes.
To identify TRAF1 domains important for LMP1-induced M1-pUb attachment, 293 cell transient transfection assays were performed with FLAG-tagged TRAF1 constructs, co-transfected with untagged WT LMP1. 24 hours after transfection, FLAG immuno-purified complexes were analyzed by western blot. Interestingly, FLAG-TRAF1 183–416 complexes, but not FLAG-TRAF1 264–416 complexes expressed at a similar level, were modified by M1-pUb chains (Fig 5C). TRAF1 183–264 residues form a coiled-coil (CC) domain, which is required for the formation of TRAF homo- and hetero-trimers. Notably, FLAG-TRAF 264–416 did not associate with LMP1, suggesting that TRAF1 trimerization and/or physical association with LMP1 are important for incorporation into complexes that contain M1-pUb chains. This result is consistent with the prior observation that TRAF trimers, rather than monomers, associate tightly with activated CD40 receptors.
Loss-of-function approaches were used to test the importance of LUBAC subunits, TRAF2, TRAF3 and cIAP1/2 in M1-pUb chain attachment to TRAF1 complexes, since each were identified as high-confidence TRAF1 interactors. First, we used an siRNA approach to investigate the role of the three LUBAC components. 72 hours after 293 TRAF1 cell siRNA transfection, LMP1 1–231 expression was induced for 16 hours. The M1-pUb chain content of FLAG- TRAF1 immuno-purified complexes was analyzed by western blot. Knockdown efficiency was measured by western blot, using whole cell lysates (Figs 6A and S7). We were unable to identify commercially available antibodies that recognized endogenous HOIL-1L in our 293 cells, and instead used quantitative PCR analysis to validate HOIL-1L mRNA depletion in a parallel experiment (S8 Fig). Interestingly, we found that depletion of HOIP, HOIL-1L, or SHARPIN each impaired M1-pUb chain attachment to TRAF1 complexes, suggesting that all three LUBAC components play important and non-redundant roles, at least in 293 TRAF1 cells (Fig 6A). TRAF2 depletion likewise reduced M1-pUb chain abundance in purified FLAG-TRAF1 complexes, and also diminished association between TRAF1 and the LUBAC components HOIP and SHARPIN (Figs 6A and S7). By contrast, TRAF3 depletion did not impair M1-pUb chain attachment to purified TRAF1 complexes (S7 Fig). Taken together with our prior observation that TRAF2 complexes are not modified by M1-pUb in cells that lack TRAF1 expression, our results suggest that a TRAF1:TRAF2 heterotrimer, rather than a TRAF1 homotrimer, may be the functional unit that associates with LUBAC. Of note, HOIL-1L knockdown increased TRAF2 steady state levels (Fig 6A), while HOIP and SHARPIN knockdown also increased TRAF2 levels to a lesser extent. To our knowledge, LUBAC has not previously been implicated in control of TRAF2 steady-state levels, though it remains possible that this effect is specific to cells that express LMP1. Depletion of HOIP and HOIL-1L from 293 TRAF1 cells impaired LMP1 1-231-mediated p38, JNK and canonical NF-kB activation, consistent with a role for M1-pUB chains at or above the level of TAK1 kinase activation (S9 Fig).
Given the key roles that TRAF2 and cIAP ligases play in TNFR1-mediated LUBAC recruitment and activation, and since the TRAF1:TRAF2 heterotrimer more efficiently recruits cIAP ligases than TRAF2 homotrimers [83], we next tested the importance of cIAP1 and cIAP2 in M1-pUb chain attachment to TRAF1 complexes. cIAP1 and cIAP2 perform largely redundant functions, potentially complicating siRNA loss-of-function approaches that would require their compound knockdown. We therefore used a cell-permeable SMAC mimetic peptide to deplete cIAP1 and cIAP2. SMAC mimetics induce cIAP1/2 auto-ubiquitination and rapid proteasomal degradation [84]. Interestingly, 293 TRAF1 cell treatment with SMAC mimetic prior to and throughout the 16 hours of LMP1 1–231 expression did not impair subsequent attachment of M1-pUb chains to LMP1/TRAF1 complexes, despite efficiently inducing cIAP1/2 depletion from 293 cell whole cell lysates (Fig 6B). This result suggests that LMP1-induced LUBAC association with TRAF1:TRAF2 complexes differs from TNF-alpha induced LUBAC recruitment to TRAF2:cIAP complexes, which require cIAP-catalyzed pUb chain formation.
The LMP1 N-terminus can be modified by ubiquitin conjugates in transient overexpression assays [85]. We therefore tested whether LMP1 itself might be modified by M1-pUb chain attachment in GM12878 LCLs. EBV-negative BL2 Burkitt lymphoma cells were used as a B-cell negative control. To disrupt protein-protein complexes, GM12878 and BL2 cells were lysed under highly denaturing conditions, using buffer containing 8M urea. M1-pUb chains were then immune-purified, using a M1-pUb-specific monoclonal antibody under denaturing conditions in a buffer that contained 7M urea (Methods) [81]. Immuno-purified M1-pUb material was analyzed by western blot, using an LMP1-specific monoclonal antibody (LMP1 is not epitope tagged in GM12878). Surprisingly, high molecular weight LMP1 conjugates were evident, consistent with the possibility that LMP1 is a direct target of M1-pUb chain attachment (Fig 7A). Since WT LMP1 has only a single lysine residue, K330, we hypothesized that it could be the M1-pUb attachment site. We therefore tested whether the lysine-less and untagged LMP1 1–231 molecule was modified by M1-pUb chains in conditional 293 TRAF1 cells. M1-pUb chains were immuno-purified under denaturing conditions from uninduced 293 TRAF1 cells, and from 293 TRAF1 cells induced for LMP1 1–231 expression for 16 hours. M1-pUb immuno-purified material was analyzed by western blot, using an anti-LMP1 monoclonal antibody. Surprisingly, high molecular weight LMP1 conjugates were again identified in extracts from LMP1 1–231+ cells (Fig 7B). Notably, high molecular weight LMP1 1–231 species were not observed in M1-pUb pulldowns from 293 TRAF1 cells treated with HOIP siRNAs 72 hours prior to LMP1 1–231 induction, and then western blotted either for M1-pUb or total Ub (Fig 7C). These results suggest that LMP1 may be a direct LUBAC target, and support the specificity of the anti-M1-pUb monoclonal antibody. Further studies are required to identify the M1-pUb attachment site in the lysine-less LMP1 1–231 molecule. Of note, all SDS/PAGE samples were boiled in loading buffer with fresh 5% beta-mercaptoethanol, which disrupts cysteine-ubiquitin linkages [86]. Possible LMP1 M1-pUb attachment sites include the LMP1 N-terminus itself or a non-lysine LMP1 residue.
We next used ubiquitination assays to determine whether LUBAC ubiquitinates recombinant TRAF1 in vitro. Recombinant N-terminally GST-tagged TRAF1 was added to reaction buffer containing ubiquitin, ATP, the ubiquitin E1 enzyme, as well as the indicated combinations of the ubiquitin E2 enzyme UBE2L3 and LUBAC components (S10 Fig and Methods). Reactions were stopped after two hours, boiled in Laemmli sample buffer, and were analyzed by western blot for M1-pUb chain linkages. Interestingly, high-molecular weight, immune-reactive species were abundantly present in lanes 7 and 8, from reactions that contained E1, E2, HOIP, HOIL-1L and TRAF1. The lane 8 reaction also contained SHARPIN. Since SHARPIN was found to be important for M1-pUb attachment to TRAF1 complexes in 293 cells, perhaps protein concentrations used in the in vitro Ub assay circumvent the need for SHARPIN’s regulatory role.
Aberrant HOIP activity has recently been implicated in the pathogenesis of the activated B cell-like (ABC) subtype of diffuse large B-cell lymphoma (DLBCL) [87,88]. LUBAC inhibition was synthetically lethal to ABC DLBCL, but not the germinal center lymphoma subtype, which have lower NF-kB activity [88]. Given these and our results, we tested the effect of HOIP knockdown on GM12878 LCL growth and survival. Interestingly, by comparison with a non-targeting shGFP control, HOIP depletion by five independent shRNAs significantly impaired LCL growth in biological triplicate assays (Fig 8A). While four anti-HOIP shRNAs yielded very similar effects, a fifth anti-HOIP shRNA (shRNA #3 on the growth curve) had a statistically significant effect in the same direction, but a more modest growth phenotype. This attenuated phenotype may reflect partial rescue by an off-target shRNA effect, or partial rescue by an alternatively spliced HOIP transcript that lacks the shRNA targeting sequence, and which results in a truncated protein not recognized by our anti-HOIP antibody. To further validate the overall shRNA result, we used CRISPR/Cas9 mutagenesis in GM12878 cells that stably express Cas9 to deplete HOIP (Methods). An anti-HOIP exon 1 small guide RNA (sgRNA) knocked down HOIP expression and caused a statistically significant decrease in LCL growth, by comparison with a control anti-GFP sgRNA (Fig 8B). CRISPR/Cas9 edited cells with residual HOIP expression, for example as a result of mono-allelic HOIP disruption, may account for residual LCL growth observed in this experiment. Anti-HOIP sgRNA expression triggered marked induction of caspase 3 and 7 activity, and to a lesser extent, caspase 8 activity (S11 Fig). Western blot analysis of GM12878 whole cell lysates obtained 6 days after transduction with sgRNA-expressing lentiviruses also demonstrated cleaved caspases 3, 7, 9 and cleaved PARP. Overall, these results suggest that HOIP depletion predominantly triggers the intrinsic apoptosis pathway. While TRAF1-independent HOIP roles may also have contributed to this phenotype, it nonetheless suggests an important role for M1-pUb chains in LCL growth and survival, and highlights LUBAC as a potential therapeutic target in EBV-associated lymphoproliferative disorders. We next tested the effect of TRAF1 knockdown in GM12878 LCLs, and found that five independent TRAF1 shRNAs each significantly impaired LCL growth relative to the shRNA control (Fig 8C). We note that all five TRAF1 shRNAs had similar effects on LCL growth, despite variation in the extent of TRAF1 knockdown evident on whole cell extract western blot four days after lentivirus transduction. This result raises the possibility that GM12878 are quite sensitive to TRAF1 depletion, and that even partial TRAF1 depletion impaired cell proliferation. Alternatively, off-target shRNA effects may also have contributed to shRNA effects on LCL growth, in particular for shRNA #3, which depleted TRAF1 to the least extent. TRAF1 levels may also have become more similar across shRNA conditions at subsequent timepoints. Nonetheless, the ability of all five TRAF1 shRNAs to impair LCL proliferation argues against off-target effects being solely responsible for the observed phenotypes. Collectively, our results support an important role for TES1 and TRAF1-dependent M1-pUb chains in the LCL immortalized growth phenotype.
K63-linked pUb chains have important roles in canonical NF-kB and MAP kinase activation pathways. Notably, our proteomic analysis suggested that TRAF1 associated with multiple E3 Ub ligases that catalyze K63-linked pUb chains, including TRAF2, TRAF3, cIAP1, and cIAP2 (S1 Table). Likewise, A20, ABIN1 and SQSTM1, each of which have domains that bind to K63-pUb chains, associated with TRAF1 in LMP1 1–231 induced cells. Given also the important role that K63-linked pUb chains play in LUBAC recruitment to TNFR1, we examined whether LMP1 or TRAF1 complexes were decorated by K63-pUb chains. First, we used 293 TRAF1 cells to test whether LMP1 1–231 expression induces attachment of K63-pUb-linked chains to TRAF1 complexes. FLAG-TRAF1 complexes were immuno-purified from 293 TRAF1 cells uninduced or induced for LMP1 1–231 expression for 16 hours, and subjected to western blot analysis with a K63-pUb chain specific antibody [89]. While TRAF1 was not associated with K63-pUb chains in unstimulated 293 cells, TRAF1 complexes were highly modified by K63-pUb chains in extracts from cells that co-express LMP1 1–231 (S12 Fig).
The LMP1 TES2 domain highly activates TRAF6 K63-pUb ligase activity, but does not associate with TRAF1. To test whether LMP1 TES2 signaling nonetheless stimulates K63-pUb attachment to TRAF1 complexes, we co-transfected 293 cells with FLAG-TRAF1 and either WT LMP1, or LMP1 mutants deficient for TES1 and/or TES2 signaling. FLAG-TRAF1 complexes purified from cells that co-expressed wildtype LMP1, or the TES2 null LMP1 mutant, were modified by K63-pUb chains. By contrast, the LMP1 TES1 domain 204AQAAA208 triple point mutant did not stimulate attachment of k63-pUb chains to TRAF1 complexes (Fig 9A). This result is consistent with a model in which TRAF1 recruitment to LMP1 TES1 is important for subsequent K63-pUb attachment to TRAF1 complexes.
We examined whether LMP1 complexes are also modified by K63-pUb chains. Indeed, FLAG-LMP1 complexes, immuno-purified from FLAG-LMP1 LCLs, were modified by K63-pUb chains (Fig 9B). To investigate whether LMP1 1–231 signaling is important for K63-pUb chain attachment to LMP1 complexes, 293 cells were co-transfected with FLAG-tagged LMP1 WT, 1–231, or DM vectors and untagged TRAF1. 24 hours after transfection, FLAG-LMP1 immuno-purified complexes were analyzed by western blot for K63-pUb chain attachment (Fig 9C). K63-pUb chains decorated WT and 1–231 LMP1 complexes, but not FLAG-LMP1 DM complexes, suggesting that TES1 signaling is important for K63-pUb chain attachment to LMP1 complexes. To identify target(s) of LMP1 TES1-induced K63-pUb chains, we analyzed the K63-pUb chain status of proteins known to associate with the LMP1 TES1 domain. 293 cells were co-transfected with untagged TRAF1, HA-tagged WT LMP1, and FLAG-tagged TRAF1, TRAF2, TRAF3 or GFP negative control, as indicated.
To determine whether LMP1 might itself be modified by K63-linked pUb chains, 293 cells were also co-transfected with untagged TRAF1 and FLAG-LMP1. To denature protein-protein complexes, 1% SDS was added to 293 cell lysates, and samples were boiled for 5 minutes. The SDS concentration was then reduced to 0.1% by addition of NP40 lysis buffer, and FLAG-tagged proteins were immune-purified. Interestingly, western blot analysis demonstrated K63-pUb chain modification of FLAG-TRAF2 (lane 3), whereas signals in other lanes were similar to the FLAG GFP negative control (Fig 10). To determine whether LMP1 signaling induced K63-pUb chain attachment on TRAF2, we also analyzed 293 cells co-transfected with TRAF2 and untagged TRAF1, but no LMP1 (lane 6). Interestingly, FLAG-TRAF2 complexes had only background levels of K63-pUb chains when not co-expressed with LMP1, despite similar TRAF2 expression levels. Collectively, our results suggest that LMP1 and TRAF1 stimulate K63-pUb attachment to TRAF2, likely in the context of a TRAF1:TRAF2 heterotrimer.
Abundant TRAF1 expression is a hallmark of multiple EBV-associated human malignancies, including Hodgkin lymphoma and post-transplant lymphoproliferative disorder [47,48,49]. TRAF1 is one of the most highly LMP1-induced genes [42,46,67], and is up-regulated early in the course of EBV-mediated primary B-cell transformation, with close correlation to LMP1 expression [90]. TRAF1 promotes Hodgkin disease Reed-Sternberg cell survival [91]. However, the molecular mechanisms that underlie TRAF1 function downstream of LMP1, or downstream of immune receptors more generally, have remained incompletely understood. In particular, how TRAF1 enhances LMP1 TES1 domain-mediated activation of the JNK [92] and NF-kB pathways [29] have remained uncharacterized.
To gain insight into TRAF1 function, we took a proteomic approach, and found that LMP1 1–231 expression induced association between TRAF1 and LUBAC components. Indeed, TRAF1 complexes purified from GM12878 LCLs contained LUBAC components and were modified by M1-pUb chains. Likewise, we found that LMP1 complexes immuno-purified from LCLs were highly decorated by M1-pUb chains, and LMP1 1–231 expression in 293 TRAF1 cells stimulated LUBAC activity, as judged by the appearance of high molecular weight M1-pUb conjugates by western blot analysis in whole cell extracts. Since TRAF1 associates with the LMP1 TES1 PQQAT motif, likely through interactions with conserved TRAF domain residues or as a heterotrimer with TRAF2 [24], and since most TRAF1 is associated with LMP1 in LCLs [29], our results suggest that a complex containing LMP1, TRAF1 and TRAF2 may be the target of M1-pUb chains. Indeed, TRAF2 depletion impaired the LMP1-induced association between TRAF1 and LUBAC, and reduced M1-pUb chain attachment to TRAF1 complexes. M1-linked-pUb chains play essential roles in NF-kB and JNK activation by TNFR1, CD40, IL1R1, NOD2, and TLR signalosomes [56,93,94,95,96,97,98], though to our knowledge, have not previously been implicated in a pathway downstream of a viral oncoprotein.
TRAF1:TRAF2 heterotrimers may also be a target of LMP1 TES1-stimulated K63-pUb chain attachment. LMP1 1–231 induced K63-pUb chain attachment to TRAF2 in 293 TRAF1 cells. K63-pUb chain attachment to TRAF2 may serve important roles in LUBAC recruitment and also in TAK1 activation (Fig 11). LUBAC has multiple zinc finger pUb-binding domains, and K63-pUb chains play an important role in LUBAC recruitment to TNFR1 [56,93,94,95]. LMP1-induced K63-pUb chain attachment to TRAF2 may play a similarly important role in LUBAC recruitment to LMP1 and TRAF1 complexes. Likewise, TNF-alpha induced K63-pUb chain attachment to TRAF2 is important for TAB/TAK1 complex recruitment and downstream MAP kinase and canonical NF-kB activation [99–100]. K63-pUb chain linkage to TRAF2 may play a similar role in activating TAK1 downstream of LMP1 TES1. Colocalization of M1- and K63-linked pUb chains may serve to juxtapose the IKK and TAB/TAK1 complexes at the level of LMP1 complexes, and thereby enhance MAP kinase and canonical NF-kB activation (Fig 11).
TRAF1:TRAF2 heterotrimers associate with cIAP1 and cIAP2 more tightly than TRAF1 or TRAF2 homotrimers [24]. Since cIAP1/2 catalyzes pUb chains that are essential for LUBAC recruitment to TNFR1, we investigated whether cIAP1/2 were likewise important downstream of LMP1. Surprisingly, cIAP1/2 depletion by SMAC mimetic did not impair LMP1 1-231-induced M1-pUb chain attachment to TRAF1 complexes. While residual cIAP activity could have been sufficient to enable LUBAC recruitment, TNFR1 and LMP1 signaling may differ in this regard. An area of future LMP1 investigation will be to identify the Ub ligase that attaches K63-pUb chains to TRAF2.
We were also intrigued to find that LMP1 may be a target of M1-linked pUb chain attachment in GM12878 LCLs and 293 cells. Western blot analysis of M1-pUb chains, immuno-purified under denaturing conditions from GM12878 or conditional 293 TRAF1 cells, demonstrated high-molecular weight bands reactive with an anti-LMP1 antibody. Since untagged LMP1 1–231 does not have a lysine residue, M1-pUb may therefore be attached to the LMP1 N-terminus or to an LMP1 non-lysine residue. While Kaposi sarcoma associated herpesvirus MIR1 can attach ubiquitin to a MHC class I cysteine residue, all of our western blot samples were treated with reducing agent, which disrupts cysteine-ubiquitin linkages [86]. We note that N-terminally FLAG-tagged LMP1 complexes purified from LCLs are modified by M1-pUb chains. Possible explanations include attachment of M1-pUb chains to FLAG tag lysine residues, to the FLAG N-terminus, to LMP1 K330, or to an LMP1-associated protein, such as TRAF1. Indeed, recombinant TRAF1 was found to be a LUBAC target in vitro.
We identified LUBAC as a potential therapeutic target in EBV-transformed B-lymphoblastoid cells. HOIP depletion by independent shRNAs or by CRISPR/Cas9 mutagenesis impaired GM12878 LCL growth and induced apoptosis, largely through activation of the intrinsic apoptosis pathway. Interestingly, LUBAC has recently been implicated in the pathogenesis of the ABC subtype of DLBCL, and a stapled alpha-helical peptide inhibitor that blocks HOIP and HOIL-1L association is toxic to DLBCL [87,88]. A goal of future studies will be to identify whether anti-LUBAC stapled peptides inhibit the growth of EBV-transformed B-cells. Similarly, LMP1 highly upregulates TRAF1 expression in transfected keratinocytes, and TRAF1 expression was detectable in 17 of 42 EBV+ undifferentiated nasopharyngeal carcinomas (NPC) [50]. Further studies are required to determine whether TRAF1 associates with LUBAC in the context of NPC, whether TRAF1 or LMP1 co-localize with M1- or K63-chains in NPC tumor samples, and whether HOIP depletion is toxic to EBV+ NPC cells in culture.
|
10.1371/journal.pgen.0030171 | Genome-Wide Expression Profiling of the Arabidopsis Female Gametophyte Identifies Families of Small, Secreted Proteins | The female gametophyte of flowering plants, the embryo sac, develops within the diploid (sporophytic) tissue of the ovule. While embryo sac–expressed genes are known to be required at multiple stages of the fertilization process, the set of embryo sac–expressed genes has remained poorly defined. In particular, the set of genes responsible for mediating intracellular communication between the embryo sac and the male gametophyte, the pollen grain, is unknown. We used high-throughput cDNA sequencing and whole-genome tiling arrays to compare gene expression in wild-type ovules to that in dif1 ovules, which entirely lack embryo sacs, and myb98 ovules, which are impaired in pollen tube attraction. We identified nearly 400 genes that are downregulated in dif1 ovules. Seventy-eight percent of these embryo sac–dependent genes were predicted to encode for secreted proteins, and 60% belonged to multigenic families. Our results define a large number of candidate extracellular signaling molecules that may act during embryo sac development or fertilization; less than half of these are represented on the widely used ATH1 expression array. In particular, we found that 37 out of 40 genes encoding Domain of Unknown Function 784 (DUF784) domains require the synergid-specific transcription factor MYB98 for expression. Several DUF784 genes were transcribed in synergid cells of the embryo sac, implicating the DUF784 gene family in mediating late stages of embryo sac development or interactions with pollen tubes. The coexpression of highly similar proteins suggests a high degree of functional redundancy among embryo sac genes.
| During the sexual reproduction of flowering plants, a pollen tube delivers sperm cells to a specialized group of cells known as the embryo sac, which contains the egg cell. It is known that embryo sacs are active participants in guiding the growth of pollen tubes, in facilitating fertilization, and in initiating seed development. However, the genes responsible for the complex biology of embryo sacs are poorly understood. The authors use two recently developed technologies, whole-genome tiling microarrays and high-throughput cDNA sequencing, to identify hundreds of genes expressed in embryo sacs of Arabidopsis thaliana. Most embryo sac–dependent genes have no known function, and include entire families of related genes that are only expressed in embryo sacs. Furthermore, most embryo sac–dependent genes encode small proteins that are potentially secreted from their cells of origin, suggesting that they may act as intracellular signals or to modify the extracellular matrix during fertilization or embryo sac development. These results illustrate the extent to which our understanding of plant sexual reproduction is limited and identifies hundreds of candidate genes for future studies investigating the molecular biology of the embryo sac.
| The life cycle of plants alternates between haploid gametophyte and diploid sporophyte generations. A central step in plant sexual reproduction is the transfer of sperm cells from the male gametophyte, the pollen grain, to the female gametophyte, the embryo sac, resulting in fertilization and the formation of a new sporophytic embryo.
In flowering plants, each embryo sac develops within the sporophytic tissues of the ovule, which is itself located within the ovary of the flower. Embryo sac development is preceded by meiosis, and consists of precise series of mitotic divisions, nuclear migrations, cellularizations, and cell deaths (reviewed in [1,2]). In Arabidopsis, the mature embryo sac consists of four cells: the egg cell, two synergid cells, and a large central cell. During fertilization, a pollen tube penetrates the sporophytic tissues of the ovule and terminates growth at one of the synergid cells of the embryo sac. Following the rupture of both the targeted synergid and the pollen tube cell, the two sperm cells fuse with the egg cell and the central cell to form the embryo and the endosperm.
Genes expressed within the embryo sac are responsible for many aspects of embryo sac biology. Embryo sac–expressed genes control the developmental program of the embryo sac, as evidenced by the large number of female gametophytic mutants that result in the arrest of embryo sac development at various stages [3–9]. Embryo sac–expressed genes are also central to the fertilization process. Ovules that do not contain functional embryo sacs do not attract pollen tubes [9,10] , an observation which led to the suggestion that the embryo sac produces signals that guide pollen tube growth. Both Arabidopsis genetics [11] and in vitro studies with Torenia ovules [12,13] have identified the synergid cells as a source of embryo sac–derived pollen tube attractants. In addition, embryo sac–expressed genes are required for the reception of pollen tubes by the synergids [6,14,15] and for the coupling the initiation of seed development to fertilization [16–19].
There has been success in the identification of numerous female gametophytic mutants. However, the total set of genes expressed in the embryo sac is poorly defined due to the fact that the embryo sac is embedded within the sporophytic tissues of the ovule, making it difficult to directly isolate embryo sac tissue for gene expression analysis. Here, we used genetic subtraction to identify embryo sac–expressed genes by comparing gene expression in wild type ovules to that in determinate infertile1 (dif1) and myb98 mutant ovules. DIF1 encodes a cohesin required for meiosis; sporophytic tissues of the ovule are unaffected in dif1 ovules, but gametogenesis is prevented by the failure of meiosis to produce functional megaspores [20,21]. dif1 ovules therefore represent a clean, genetic ablation of the entire embryo sac. MYB98 encodes a transcription factor expressed specifically within the synergid cells of the embryo sac [11]. The early stages of embryo sac development in myb98 ovules resemble wild type, and the only observable morphological differences of mature myb98 embryo sacs compared to wild type are abnormalities within the subcellular structures of the synergid cells [11]. In addition, myb98 ovules have an incompletely penetrant pollen tube guidance defect [11]. myb98 mutations therefore impact the last stages of embryo sac development and specifically impact interaction with pollen tubes.
We used high-throughput (454) cDNA sequencing and whole-genome tiling arrays to compare gene expression in wild-type ovules to that in dif1 and myb98 ovules. Importantly, these two techniques allow for genome-wide measurement of gene expression that is unbiased toward annotated genes. We identified 382 genes that were downregulated in dif1 ovules and 77 genes that were downregulated in myb98 ovules. The majority of genes downregulated in each mutant belonged to families of small, potentially secreted proteins. Because most embryo sac–dependent genes were unannotated or recently annotated, only 31% were by reported by recent studies of embryo sac gene expression using the annotation-based ATH1 microarray [22,23]. Our results identify a surprisingly large number of embryo sac–expressed, secreted proteins as candidate extracellular signaling molecules during embryo sac development and fertilization, and in particular implicate the poorly understood DUF784 gene family as potential mediators of the last stages of embryo sac development or signaling interactions between the embryo sac and the pollen tube.
To obtain an unbiased, genome-wide survey of ovule gene expression, we sequenced stage 14 ovule cDNAs from male sterile ms-1 plants (Landsberg ecotype) using the high-throughput 454 sequencing method [24]. We obtained 249,440 cDNA reads, comprising a total of 26.5 million bp, with the reads having a median length of 106 bp (Table 1). 225,499 reads (90%) could be confidently aligned to the Arabidopsis genome [25] using blat or blastn (Dataset S1), with the majority of unalignable reads consisting primarily of simple sequence repeats and/or PCR primer sequence. 28,732 reads matched equally well to more than one genomic location and thus represent transcripts from recent duplications. Eighty-five percent of alignable reads mapped to annotated exons, covering 12.5% of all annotated exonic sequence. In total, 15,312 annotated genes were matched by at least one cDNA read. The number of reads per gene ranged from 0 to 3,099 (to AtMg00020, a ribosomal protein encoded by the mitochondrial genome), with most genes having 0–5 reads, 15% of genes having ten or more reads, and 5% of genes having 25 or more reads (Figure S1). The aligned cDNA reads were 97.5% identical to the published genome sequence. The 2.5% of bases not matching genomic sequence were likely the result of cDNA sequencing errors as well as ecotype-specific polymorphisms due to the alignment of Landsberg cDNAs to the Columbia genome.
We also searched for embryo sac–dependent transcripts across the entire Arabidopsis genome by using whole-genome tiling microarrays to compare gene expression in wild-type (Columbia ecotype) ovules to that in dif1 and myb98 ovules. We dissected ovules from mature (stage 14) flowers, collecting sufficient material to yield at least 2 μg of total RNA for each of four biological replicates for each genotype (wild type, dif1, and myb98), resulting in a total of 12 samples. After reverse transcription, second-strand cDNA synthesis, and double-stranded random labeling, samples were hybridized to Genechip Arabidopsis Tiling 1.0F arrays (Affymetrix), which contain over 3,000,000 25mer perfect match probes spread across the Arabidopsis genome with a median gap of 10 bp between probes. The log2 transformed hybridization signals to the perfect match probes were quantile normalized across the twelve arrays; pairwise correlations ranged from 0.952 to 0.966 (Table S1).
Mutations in DIF1 result in the ovules that entirely lack embryo sacs [20]. To identify embryo sac–dependent transcripts de novo, without bias towards existing gene models, we used a simple algorithm to identify genomic regions differentially expressed between dif1 and wild-type ovules. In brief, a Welch's t-test was performed for each probe comparing the log2 scale expression values of the four wild-type replicates to the four mutant replicates. An arbitrary threshold was applied to define probes that correspond to differentially expressed messages (p ≤ 0.05, log2 fold change ≥ 1). Neighboring probe matches within 80 bp of each other that met this threshold were joined to define differentially expressed intervals, with the requirement that at least three differentially expressed probe matches were required to define an interval.
Using this algorithm, we identified 1,099 genomic intervals that were downregulated in dif1 ovules compared to wild type. Of these intervals, 969 mapped to an annotated gene. The remaining 130 seemingly intergenic intervals were compared to the genomic alignments of the ovule cDNAs. After joining adjacent intervals (those separated by less than 200 bp), 27 dif1 downregulated intervals that overlapped with ovule cDNA alignments were considered as putative unannotated genes. Using cDNA and EST sequences as guides, open reading frames (ORFs) were found for 22 of these putative genes (Tables 2 and S3). Sixteen of the newly identified ORFs were matched by at least three ovule cDNA reads with unique matches to the genome (Table 2). As an example of the array and cDNA data supporting one of the newly annotated genes, a region between the annotated genes At1g01300 and At1g01310 with substantially more expression in wild-type ovules than in dif1 ovules was detected as a differentially expressed interval, whereas surrounding genic and intergenic probes had highly similar expression values between all three genotypes (Figure 1A). In this case, the same interval was also differentially expressed between wild-type and myb98 ovules, and overlapped with the genomic alignments of 11 ovule cDNAs (Table 1; Figure 1A). The differentially expressed interval also overlapped well with a putative ORF and was assigned the name At1g01305 (Figure 1A).
We used the tiling array data to quantify changes in transcript levels between wild-type and dif1 ovules for the entire set of Arabidopsis genes, including both the 22 genes we identified as well as previously annotated genes (The Arabidopsis Information Resource [TAIR] release 7 annotations, containing 27,029 protein coding genes, 3,889 pseudogenes, and 1,123 noncoding RNA genes [http://www.arabidopsis.org]). For each gene, a t-test comparing wild-type and dif1 signal intensities was performed across all probes matching that gene (Table S4). We defined genes as having significantly different expression in dif1 compared to wild type by setting a p-value threshold of 0.001 and a log2 fold change threshold of 1 (which corresponds to a 2-fold change). At these cutoffs, we found 382 protein-coding genes that were expressed at lower levels, and 35 genes that were expressed at higher levels in dif1 ovules (Tables 3, S5, and S6).
To empirically assess the extent to which sampling error contributed to the observed differential expression, we estimated the false discovery rate (FDR) by shuffling the eight arrays (four wild type and four dif1) into two permuted groups of four and reanalyzed the data to identify the number of genes with seemingly differential expression between these arbitrarily grouped sets or arrays. To control for differences in gene expression relating to the dif1 phenotype, we considered only the 18 balanced permutations in which the two groups of arrays being compared each contained two wild-type arrays and two dif1 arrays. On average, 3.6 genes had the p-values less then 0.001 and log2 changes in expression greater than 1 for the balanced permutations of the wild-type and dif1 datasets. Therefore, we estimate the FDR to be approximately 1% for the 417 genes differentially regulated between dif1 and wild type at this threshold. Estimates of the FDR at more relaxed thresholds suggest that several hundred additional genes are differentially expressed in dif1 ovules with changes in expression less than 2-fold (Table S2).
Because the dif1 mutation only affects cells that undergo meiosis, the simplest interpretation of these data is that the 382 dif1 downregulated genes are expressed preferentially within the embryo sac as compared the sporophytic ovule. It is also possible that some of these genes require the presence of the embryo sac for expression within the sporophytic ovule.
To characterize the set of DIF1-dependent genes, we analyzed the abundance of protein domains in the sets of differentially expressed genes as compared to the total set of protein coding genes. Ten gene families were significantly overrepresented in the set of dif1 downregulated genes (Table 3); 241 of the 382 dif1-downregulated genes belonged to one of these ten gene families. Several of these families, such as Domain of Unknown Function 784 (DUF784), DUF1278, and DUF239, lack homology to any protein with a known function. Two families, the Defensin-Like (DEFL) genes and the thionin-like genes, have homology to small, secreted antipathogenic peptides, whereas the Papaver Self-Incompatibility-Like (PSIL) genes have homology the pistil-secreted S1 protein of Papaver. The functions of these six families within the context of the Arabidopsis ovule are unknown. The remaining overrepresented families encode proteins with presumed functions as catalytic enzymes (peptidases, lipases, and polygalacturonases) or as enzyme inhibitors (pectinmethylesterase inhibitors [PMEIs]). Many members of these gene families are encoded by tandemly arrayed, recently duplicated genes (Figures 1B, 1C, S2, and S3).
In addition to the overrepresentation of certain protein domains, the set of dif1 downregulated genes was highly enriched for genes encoding small proteins that contain putative signal peptides (Table 3). Seventy-eight percent of dif1 downregulated genes were predicted to encode for a signal peptide, as compared to 18% among all protein-coding genes, and 66% of dif1 downregulated genes were predicted to encode proteins that weigh less than 20 kilodaltons, as compared to 20% among the total set of annotated proteins (Table 3). This bias towards small proteins with signal peptides was related to the bias towards certain protein families; 91% of the 215 DUF784, DUF1278, PSIL, DEFL, PMEI, and thionin-like genes downregulated in dif1 encode proteins that contain putative signal peptides and that weigh less than 20 kD. The presence of a signal peptide can target a protein for one of several fates, such as localization to a membrane-bound organelle, localization to the cell membrane, or secretion from the cell. However, the abundance of putative signal peptides amongst DIF1-dependent proteins, as well as the fact that numerous DIF1-dependent genes have homology to proteins that are known to be secreted in other organisms or tissues (e.g., DEFL, thionin-like, and PSIL genes) [26–28], suggests that many embryo sac–dependent proteins have the potential to act outside of their cells of origin.
The 140 dif1 downregulated genes that did not belong to the ten gene families listed in Table 2 represented a broad range of functionalities (Table S5). Several DIF1-dependent genes have known roles in embryo sac biology, including the synergid-expressed transcription factor MYB98 as well as two genes, FERTILIZATION INDEPENDENT SEED2 and MEDEA, that regulate the development of the central cell and endosperm [16,17].
While 382 genes were downregulated at least 2-fold in dif1 ovules, some genes were more highly downregulated. Most of the genes with large differences in expression levels between dif1 and wild-type ovules belonged to multigenic families encoding small, potentially secreted proteins (Figure 2). For example, 83% of the 189 genes downregulated at least 8-fold in dif1 ovules belonged to the DUF784, DUF1278, DEFL, PSIL, PMEI, or thionin-like gene families (Figure 2). All 23 genes that were downregulated at least 64-fold in dif1 ovules belonged to the DUF784 or DUF1278 families (Figure 2).
We analyzed the expression patterns of 59 dif1 downregulated genes by reverse transcriptase PCR (RT-PCR), with a focus on members of the DUF784, DUF1278, PSIL, and DEFL gene families (Figure 3). In all 59 cases, including eight previously unannotated genes, expression was lower in dif1 ovaries than in wild type (Figure 3).
In addition to 382 protein-coding genes, 26 annotated pseudogenes were also significantly downregulated in dif1 ovules (Table S9). Most of these pseudogenes had a high degree of homology to adjacent, tandemly arrayed protein coding genes also downregulated in dif1 ovules (e.g., DUF784 pseudogenes). While some of the observed expression of these pseudogenes may have been due to cross-hybridization to transcripts from homologous protein-coding genes, the fact that 13 were uniquely matched by ovule cDNA reads indicates that many are in fact transcribed (Table S9). It seems that some recently duplicated, embryo sac–dependent genes have retained regulated, functional promoters despite having acquired frame shift or nonsense mutations within their ORFs. It is unclear as to what, if any, functional roles these expressed pseudogenes might play.
Many embryo sac–dependent genes have similarity to each other at the nucleotide sequence level, suggesting a common origin and function. The ORFs of 109 dif1 downregulated genes were >90% identical to another dif1 downregulated gene. In most cases, highly similar genes were present in tandem arrays of apparently recently duplicated genes. Seventy-five of the DIF1-dependent genes could be grouped into 17 clusters of highly similar genes that shared at least 50% of their tiling array probes with another gene. Twenty-six of these partially ambiguous genes could still be identified as significantly downregulated in dif1 ovules based solely on the expression values of probes with unique matches in the Arabidopsis genome. Another 14 were uniquely matched by ovule cDNA fragments, providing evidence that they are expressed in the ovule. Nonetheless, for approximately 50 genes, it is difficult to be certain that the differential expression observed on the tiling array truly reflected the expression of each individual gene or if only a subset of genes were differentially expressed.
The most extreme example of closely related embryo sac–dependent genes is that of 30 DUF1278 genes (as well as three DUF1278 pseudogenes) that are >95% identical to each other. Most probes on the tiling array that correspond to this cluster perfectly match multiple genes; moreover, it was not possible to design RT-PCR primers specific to any particular gene from this cluster. Fifty-six of the 60 cDNA reads matching genes from this cluster matched more than one gene equally well. It is therefore difficult to be certain that the expression observed for genes of this cluster corresponds to all 30 genes or to a subset of the 30 genes. However, the high number of cDNA reads mapping to this genes in this region, together with the high degree (40- to 70-fold) of DIF1 dependence detected for this region, make it clear that as a unit, this region is expressed in an embryo sac–dependent manner. Moreover, the RT-PCR primers to this region (i.e., to gene At5g36350) failed to detect any expression from this cluster in dif1 ovules despite being perfectly complementary to most of the 30 genes (Figure 3), further demonstrating that no gene from this cluster is highly expressed in ovules that lack embryo sacs.
In contrast to the set of dif1 downregulated genes, the 35 genes upregulated in dif1 ovules were not significantly enriched for any protein domains, nor for genes predicted to encode proteins with signal peptides or weighing less than 20 kD (Tables 3 and S6). Furthermore, the magnitude of upregulation was modest compared to changes in expression levels observed for dif1 downregulated genes. Only one gene was upregulated more than 8-fold in dif1 ovules (Figure 2).
Whole-genome tiling arrays allow for the comprehensive, genome-wide measurement of gene expression. However, because the Tiling 1.0F array has been developed only recently, few studies that use it to measure gene expression have been published. In contrast, the Genechip Arabidopsis ATH1 Genome array (Affymetrix), containing 22,500 probe sets that match to 23,688 genes, is a widely used tool to measure gene expression in Arabidopsis. Two recent studies used the ATH1 microarray to identify genes that are downregulated in ovules that lack embryo sacs: Yu et al. identified 249 genes (representing 225 probe sets) downregulated in sporocyteless/nozzle (spl/nzz) ovules [23], and Steffen et al. identified 104 genes (representing 86 probe sets) downregulated in dif1 ovules [22]. A comparison to these datasets illustrates the utility of using genome-wide expression measures to profile gene expression and also validates the sensitivity of the whole genome tiling array as a means of quantifying gene expression.
Only 31% of the dif1 downregulated genes identified by the whole genome tiling array analysis were reported as embryo sac–dependent by one or both studies using the ATH1 array (Figure S4). The large number of DIF1-dependent genes uniquely discovered by the tiling array is primarily due to the fact that a surprisingly large number of embryo sac–dependent genes were not measured by the ATH1 array. While 84% of all currently annotated Arabidopsis protein-coding genes had a corresponding probe set on the ATH1 array (at least six of 11 probes perfectly matching), a significantly smaller percentage (41%) of DIF1-dependent genes were represented by ATH1 probe sets (Table 4). In total, 224 dif1 downregulated genes did not have ATH1 probe sets (Table 4).
In addition to failing to detect the majority of embryo sac–dependent genes, the ATH1 array is specifically biased against certain gene families. Whereas over 90% of DIF1-dependent genes encoding lipases, subtilisins, or polygalacturonases were represented in the ATH1 array, less than 20% of the DIF1-dependent genes belonging to families encoding small, functionally uncharacterized proteins (DUF784, DUF1278, DEFL, PSIL, and thionin-like genes) had corresponding ATH1 probe sets (Table 4; Figure S4). The poor representation of these families can be attributed to the fact that many members of these families were annotated after the design of the ATH1 array [29–31]. For example, 65 DIF1-dependent DUF784, DUF1278, and thionin-like genes were unannotated prior to the TAIR7 annotation release of April, 2007 [30]. The combined analysis of tiling array data and high-throughput cDNA sequencing led to the finding that large families of poorly understood, potentially secreted proteins are embryo sac dependent, a finding that was not evident from the more limited and biased sets of embryo sac–dependent genes detected by the ATH1 array.
Considering only those genes with ATH1 probe sets, there was considerable overlap between the tiling array and ATH1 data; 76% of the 158 dif1 downregulated genes with ATH1 probe sets were reported as downregulated in at least one of the ATH1 analyses (Figure S4). Sixty-five genes were reported as downregulated in all three studies, and another 55 genes were reported as downregulated in both our analysis and in one of the other studies (Figure S4). Thus, the reproducibility and accuracy of gene expression quantification by the whole genome tiling array was at least roughly comparable to that of the more commonly used expression array.
114 spl/nzz downregulated genes were not identified as downregulated in either study using dif1 ovules (Figure S4.) Some of these genes were downregulated in dif1 ovules, but at levels below our statistical thresholds. However, 69 SPL/NZZ-dependent genes were expressed at similar levels in dif1 and wild-type ovules (p > 0.25) and seven were actually upregulated in dif1 ovules (p < 0.05). The fact that a large number of SPL/NZZ-dependent genes are not downregulated in dif1 ovules is most likely due to the different stages of ovule development at which SPL/NZZ and DIF1 act. SPL/NZZ is known to be required for the proper expression of several key genes during development of the somatic ovule, and spl/nzz ovules never initiate meiosis [32–35]. In contrast, the somatic development of dif1 ovules appears to be entirely wild type, and dif1 ovules initiate meiosis properly [20]. Therefore, it seems that approximately one-fourth of SPL/NZZ-dependent genes are not embryo sac–dependent but rather require SPL/NZZ for expression in the somatic ovule.
Although many embryo sac genes are not detected, the ATH1 array is capable of allowing quantitative comparisons to published studies using the same platform. We measured gene expression in three biological replicates from functionally wild-type ovules (from the male sterile ms-1 mutant) on the ATH1 array. Normalized probe set expression values were calculated via the RMA method [36] from probe level data from the ovule arrays, together with probe level data from 41 sets of Affymetrix gene chip experiments from various wild-type tissues and developmental stages that did not contain stage 12 or later ovules [37,38]. We analyzed the data to identify genes for which (1) ovule expression was at least two times higher than that of any other tissue and (2) ovule expression was at least three standard deviations above the mean expression level in non-ovule tissues, resulting in 155 ovule-enriched genes (Table S10). Of the 158 DIF1-dependent genes with ATH1 probe sets, 55 were identified as ovule enriched (Table 4). Certain gene families were highly represented amongst the set of ovule-enriched genes, including all DIF1-dependent DUF784, DUF1278, and thionin-like genes with ATH1 probe sets, as well as the majority of DIF1-dependent DEFL and PSIL genes with ATH1 probe sets (Table 4). In combination, the ATH1 array data and RT-PCR data (Figure 3) show that numerous members of the DUF784, DUF1278, DEFL, and PSIL gene families are expressed primarily within the ovule.
In contrast to the total ablation of embryo sac tissue in dif1 ovules, myb98 embryo sacs are morphologically similar to wild type with exception of the subcellular structure of the synergid cells [11]. myb98 embryo sacs are also impaired in mediating pollen tube guidance [11]. Genes with reduced expression levels in myb98 ovules are therefore likely to represent genes active during the final stages of embryo sac development and during the initial steps of the fertilization process.
Using the same significance thresholds as in the dif1 versus wild type comparison (p ≤ 0.001, log2 fold change > 1), we found that 77 genes were downregulated in myb98 mutants compared to wild type, whereas 40 were upregulated (Tables 3, S7, and S8). As would be expected from the more severe dif1 phenotype and the fact that DIF1 is required for MYB98 expression, the set of MYB98-dependent genes is largely a subset of the DIF1-dependent genes; 76 of the 77 myb98 downregulated genes were also downregulated in dif1 ovules (Figure 4). The set of myb98 downregulated genes was even more highly enriched for genes encoding potentially secreted proteins (92%) and proteins weighing less than 20 kD (84%) than was the set of dif1 downregulated genes (Table 3).
Thirty-seven of the 40 DUF784 genes encoded in the Arabidopsis genome were downregulated in myb98 ovules (Table 3). In total, DUF784 genes comprised nearly 50% of the myb98 downregulated genes, while other gene families overrepresented among dif1 downregulated genes were represented to varying degrees among the myb98 downregulated genes (Table 3). The magnitude of gene downregulation in myb98 ovules was less than that in dif1 ovules (Figure 2). Whereas all 40 DUF784 genes were downregulated at least 8-fold in dif1 ovules as compared to wild type, only 23 were downregulated 8-fold in myb98 ovules (Figure 2). No genes were downregulated 64-fold in myb98 ovules.
RT-PCR analysis confirmed the degree of myb98 dependence among the different gene families. Consistent with the microarray results, all of the 16 DUF784 genes tested were expressed at lower levels in myb98 ovaries than in wild type, but many were detected at higher levels than in dif1 ovaries (Figure 4). Also consistent with the array results, only a fraction of the DUF1278, PSIL, and DEFL genes tested were expressed at lower levels in myb98 ovaries (Figure 3).
Of the 40 genes significantly upregulated in myb98 ovules compared to wild type (Table 3), 27 were downregulated in dif1 ovules (Figure 3). These appear to be genes that are expressed within the embryo sac, perhaps during the early stages of embryo sac development, and that go down in expression in response to MYB98. However, the fold increase amongst myb98 upregulated genes was modest; only one was upregulated more than 4-fold relative to wild type (Figure 2).
To localize the expression of DIF1-dependent and MYB98-dependent genes within the embryo sac, we constructed 11 transgenic lines expressing a glucouronisidase (GUS) reporter gene under the control of a putative promoter sequence corresponding to the genomic region upstream of an embryo sac–dependent gene. The four DUF784 promoters tested (At5g35405, At4g08025, At5g34885, and At2g21727) corresponded to genes that had large numbers of ovule cDNA reads (Table S4) and that represent different subfamilies of the DUF784 phylogenetic tree (Figure S2). In T1 plants of all four DUF784::GUS lines, ∼50% of ovules had a single, punctate spot of GUS expression located at the extreme micropylar end of the embryo sac (Figure 5A–5E). This localization of GUS expression is most consistent with transcription in the synergid cells, although in some cases GUS expression appeared to extend into the egg cell. GUS staining was observed in ovules of stage 12c flowers of DUF784::GUS plants, but not in ovules from stages 12b or earlier (unpublished data).
The four DUF1278 promoter::GUS lines analyzed also all resulted in GUS expression in synergid cells (Figure 5F–5I). Unlike the DUF784 promoters, all of which corresponded to MYB98-dependent genes, only two of the DUF1278 promoters tested (At5g54062 and At5g42895) corresponded to myb98 downregulated genes as determined by array analysis or RT-PCR. The synergid specific expression of At5g36340::GUS and At2g24205::GUS, neither of which correspond to MYB98-dependent genes, demonstrates that some synergid specific markers are MYB98 independent.
In contrast to the expression in synergid cells observed for DUF784 and DUF1278 promoters, the DEFL promoter (Figure 5J) and PSIL promoters (Figure 5K and 5L) that were tested drove GUS expression within the central cell of the embryo sac.
Of the 382 DIF1-dependent genes, 241 (63%) belonged to one of ten gene families (Table 3). The subset of embryo sac–dependent genes that required the synergid-specific transcription factor MYB98 was even more enriched (80%) for these same families (Table 3). While many of these gene families are similar in that they encode for small, potentially secreted proteins, each has a unique sequence profile, evolutionary history, and, presumably, role in embryo sac biology.
More than 20% of the embryo sac–dependent genes encoded for defensin-like or thionin-like proteins (Table 3), two classes of small, secreted proteins with disulfide-linked cysteines. Members of both classes are known to have antimicrobial or antifungal properties [27,28]. Not counting pseudogenes, there are approximately 286 defensin-like genes and 62 thionin-like genes in the Arabidopsis genome. For the vast majority of these genes, no functional or biochemical data exists [31]. We found that 32% of all Arabidopsis DEFL genes and 19% of all Arabidopsis thionin-like genes are embryo sac–dependent (Table 3). While the embryo sac–expressed genes of these families have homology to antipathogenic peptides, it is unknown whether the role of these families in the embryo sac is related to defense against pathogens or whether they serve other roles as small secreted proteins.
22 embryo sac–dependent genes, including seven annotated in this work, have homology to the S1 self-incompatibility protein of the genus Papaver. In Papaver, pistil-secreted S1 proteins inhibit growth and trigger cell death of incompatible pollen [39]. While Arabidopsis thaliana is self-fertile, other Brassicaceae, including the near relative Arabidopsis lyrata, exhibit self-incompatibility, albeit through incompatibility factors that do not resemble those in the Papaver stigma [26]. We found that 40% of PSIL genes encoded in the Arabidopsis genome are embryo sac–dependent. Data from the AtGenExpress expression atlas [37] show that most PSIL genes that are not embryo sac–dependent are expressed most highly in anthers or pollen, suggesting that PSIL genes also play a role in the male gametophyte.
The DUF784 and DUF1278 gene families are unique in that the majority of genes belonging to these two families are embryo sac–dependent (Table 3). In the case of DUF784, all 40 genes encoded in the genome are downregulated in dif1 ovules, and 37 out of 40 are downregulated in myb98 ovules (Table 3). Neither family has apparent homology to any protein with a known molecular or biological function, although the DUF1278 genes are related to the EARLY CULTURE ABUNDANT1 gene identified in barley microspores [40] and to the EC1 gene that is expressed in wheat egg cells [41]. Both families are defined by the presence of six highly conserved cysteines that are present in almost all family members (Figures S2 and S3). While the pattern of conserved cysteines amongst DUF1278 proteins appears to be similar to that of DUF784, sequence homology between members of these two families was not detected by blastp, nor did HMMer searches using a HMM from one family find significant homology to members of the other family.
While several gene families were overrepresented among the set embryo sac–expressed genes, the extent of embryo sac specificity amongst DUF784 genes is particularly striking. All 40 genes belonging to this family were down regulated at least 8-fold in dif1 ovules, and no DUF784 gene detected by the ATH1 array or tested by RT-PCR showed high levels of expression in any tissue outside of the ovule. Furthermore, all four of the DUF784 genes tested were transcribed within the synergid, and most DUF784 genes are significantly downregulated in myb98 ovules. In total, DUF784 genes accounted for ∼50% of the myb98 downregulated genes. The fact that the DUF784 family is both synergid expressed and MYB98 dependent suggests that the pollen tube guidance defect in myb98 ovules may be due to the downregulation of DUF784 genes and implicates the DUF784 family as being important for the development of pollen tube–attraction competence in synergid cells or as potential signaling molecules that are perceived directly by pollen tubes.
Despite decades of research in plant sexual reproduction, the genetic mechanisms that underlie the development of the embryo sac and the interactions between the embryo sac and the pollen tube have remained poorly characterized. Through the use of truly genome-wide measures of gene expression that are capable of detecting unannotated genes and recently annotated genes, our analysis uncovered the embryo sac–dependent expression of hundreds of genes not analyzed by recent studies using annotation-based microarrays [22,23]. The finding that the majority of embryo sac–dependent genes are functionally uncharacterized underscores the limited extent of our understanding of embryo sac molecular biology. The finding that hundreds of embryo sac–dependent proteins are potentially secreted suggests that the number and complexity of intracellular communications, cell well modifications, and other extracellular events that take place during embryo sac development, fertilization, and the initiation of seed development may be even greater than previously realized.
We find that hundreds of related genes, comprising entire families and subfamilies of genes with unknown function, require the mature embryo sac for expression in ovules. The fact that so many paralagous genes have overlapping domains of expression in the embryo sac suggests that there is a high degree of functional redundancy between embryo sac genes. The coexpression of functionally redundant paralogs may explain why genes from these families have not been identified in forward genetic screens for female gametophytic mutants. Furthermore, many of these embryo sac–dependent genes are not expressed at high levels in tissues other than ovules, suggesting that they may be specialized for roles in female reproductive development and function. Future experiments to discover the potentially overlapping functions of embryo sac–dependent gene families will likely be crucial to building a more complete understanding of the genetic mechanisms that underlie plant sexual reproduction.
Seeds for ms-1 (CS75, Landsberg background), dif1 (SALK_091193, Columbia background), and myb98–1 (SALK_020263, Columbia background) were obtained from the Arabidopsis thaliana Biological Resources Center. Plants were grown in a growth chamber under long day (16 h light/8 h dark) conditions at 22 °C.
Total RNA (0.5 μg)from stage 14 ms-1 ovules was reverse transcribed and PCR amplified for 15 cycles using the BD SMART cDNA synthesis kit (Clontech) as per the manufacturer's instructions. The cDNA was fragmented and subjected to high-throughput 454 sequencing (454 Life Sciences) [24].
Primer sequence in the 454 reads was masked with Crossmatch (http://www.phrap.org), and each read was aligned to the Arabidopsis genome (January 2004 release) using blat with default settings [42]. Reads that had no blat hits were aligned to the genome with blastn (http://blast.wustl.edu) (parameters S = 100, S2 = 5, gapS2 = 200, X = 26, gapX = 55, W = 12, gapW = 18, gapall, Q = 11, R = 11, M = 5, N = −11, Z = 3e9, Y = 3e9, V = 1e6, B = 1e6, hspmax = 1000, hspsepqmax = 2e5, topcomboN = 200, wordmask = seg, maskextra = 10, hspsepsmax = 2000). For each match found by blat or blastn, more precise exon–exon boundaries were defined by running exalin [43] on the genomic region found by blat or blastn, with an additional 200 nucleotides of flanking sequence on each side. For each read, matches with submaximal exalin scores were discarded, as were matches which contained less than 20 aligning nucleotides, were composed of primarily (>75%) of a single nucleotide (usually A or T), or for which the read had less than 80% identity when compared to genomic sequence. Matches with overlapping genomic coordinates and which were not transcribed from opposite strands were grouped together to build consensus “contigs.” The genomic coordinates of matches to cDNA reads were compared to those of annotated genes (TAIR release 7). Each gene was assigned a normalized number of reads, where each match to a read was weighted relative to the number of genomic matches that read had (i.e., a match to read with a unique genomic match was given a weight of 1, whereas each match to a read with four equally good genomic matches was give a weight of 0.25).
Ovules were dissected from approximately one-month-old wild-type (Columbia ecotype), dif1, and myb98 plants. To obtain mature, unpollinated ovules, stage 12a flowers were emasculated 24 h before collection of ovules. RNA was purified by RNAqueous-micro spin columns (Ambion). Sufficient material was collected to yield at least 2 μg of total RNA (requiring ∼1600 ovules from ∼40 ovaries) for each of four biological replicate for each genotype (i.e., 12 samples total). Preparation of samples for hybridization to tiling arrays was carried out as previously described [44]. PolyA RNA was purified with Oligotex beads (Qiagen), and random hexamer-primed first-strand cDNA was reverse transcribed with Superscript III reverse transcriptase (Invitrogen) at 42 °C for 1 h. Second-strand cDNA was synthesized in second-strand reaction buffer (Invitrogen) with 40 U of E. coli DNA polymerase I (New England Biolabs), 10 U of E. coli DNA ligase (New England Biolabs), and 2 U of E. coli RNase H (Epicentre) at 16 °C for 2 h. cDNA samples were incubated with 10 U of RNase H, 0.5 U of RNaseA, and 20 U of RNaseT1 at 37 °C for 20 min and then purified on Qiaquick spin columns (Qiagen). Samples were biotin labeled using the Bioprime system (Invitrogen) and concentrated by ethanol precipitation. Samples were hybridized to Genechip Arabidopsis Tiling 1.0F arrays (Affymetrix) and probe intensities were scanned at the University of Chicago Functional Genomics Center. Expression data was analyzed using the R Project for Statistical Computing and the affy package [45]. Probe intensities were corrected for spatial abnormalities [46], the perfect match intensities were background corrected with the bg.adjust function of the affy package, and the log2 transformed perfect match intensities were quantile normalized across the 12 arrays.
All 25mer probes on the Arabidopsis Tiling 1.0F array were matched to the Arabidopsis genome using blastn. Probes perfectly matching the genome more than 30 times were removed from the analysis. For each probe match, a p-value and mean log2 fold change were calculated by a t-test comparing the expression values of the four replicate wild-type expression values against the expression values of the four dif1 replicates. Probes with log2 fold changes less than 1 or p-values greater than 0.05 were removed from the analysis. Probe matches that passed these thresholds and that were located within 85 bp of each other were iteratively grouped together to define differentially expressed intervals. Intervals containing less than three passing probe matches were removed from the analysis.
The set of dif1 downregulated genomic intervals was compared to genomic locations of existing gene annotations (TAIR release 7) to identify those not mapping to within 50 bp of an annotated gene. Intergenic intervals that overlapped the genomic alignments of ovule 454 cDNA contigs were considered as potential gene fragments. Adjacent contigs were considered to be part of the same gene if they were within 200 bp or overlapped the same cDNA contig. ORFs were predicted by extending the longest ORF within the interval into flanking genomic sequence. In cases where cDNA reads suggested splicing, the splice sites of the cDNA reads were used to guide ORF annotation.
The positions of probe matches were compared to the positions of annotated exons (TAIR7 release) and to the positions of newly identified protein coding genes. Probe matches that overlapped with more than one gene (i.e., regions for which both strands are transcribed) were removed from the analysis, as were genes having less than three probe matches. For each gene, the mean log2 fold change and corresponding p-value was calculated from a t-test of the wild-type expression values against the mutant expression values (dif1 or myb98) across all probes matching that gene. Genes were considered to be differentially regulated if the p-value was less than 0.001 and the log2 change in expression was greater than 1. The FDR at this threshold was estimated by permuting the groupings of the four wild-type and four dif1 arrays. The set of eight arrays was partitioned into two “balanced” groups of four such that each group of four contained two wild-type arrays and two dif1 arrays. The expression data was reanalyzed for each of 18 possible “balanced” permutations of array groupings, and the FDR estimated as the average number of genes passing the statistical thresholds for the permuted groupings as a percentage of the number passing the thresholds for the actual grouping of arrays. The decision to base FDR estimate on the 18 balanced permutations, rather than on all 35 possible permutation (including the actual permutation), was based on the observation that the large number of genes highly downregulated in dif1 ovules resulted in an unreasonably strict threshold when the nonbalanced permutations were included in the FDR analysis.
The list of PFAM domains in TAIR7 proteins was downloaded the TAIR website. For the DEFL, DUF784, DUF1278, and thionin-like families, the existing PFAM annotations were found to omit family members. The set of nonpseudogene DEFL genes was taken from Silverstein et al. [31]. For the DUF784 and DUF1278 families, the HMMs downloaded from the PFAM website were used to iteratively search the annotated peptides using HMMer (version 3.2, http://hmmer.janelia.org/) to identify additional family members. Most genes annotated as “plant thionins” were found not to correspond to the “plant thionin” (PF00321) PFAM domain. An HMM was built using all proteins annotated as encoding a “thionin” or “thionin-like” protein, which was then used to search for additional family members. The genes found by this HMMer search are referred to as “thionin-like” in this paper.
Gene families were considered to be overrepresented in the set of dif1 downregulated genes if at least five dif1 downregulated genes encoded that domain and the number of down regulated gene encoding that domain was significantly greater than the expected number (based on the frequency of that domain amongst all proteins) as determined by a chi-squared test. The presence of putative signal peptides was predicted with SignalP [47].
To allow for the comparison of previously published gene sets from studies using the ATH1 array, we mapped ATH1 probe sets to the most recent gene annotations (TAIR7) by blasting the probe sequences against the annotated cDNA sequences. A probe set was considered to match a gene if at least six of the 11 probes perfectly matched that gene. Previously published sets of SPL/NZZ-dependent genes [23] and DIF1-dependent genes [22] were retabulated based on the mappings of the published ATH1 probe sets to the TAIR7 annotations. In cases for which a single ATH1 probe set mapped to multiple genes, all matched genes were considered as downregulated for purposes of comparison to the whole-genome tiling array data.
Ovules were microdissected out of mature (∼stage 14) ovaries from ms1/ms1 homozygotes (Landsberg ecotype), and RNA was purified using RNeasy columns (Qiagen) as per the manufacturer's instructions. Three replicate samples were collected, each of which yielded sufficient RNA (>7 μg) to allow unamplified preparation of cRNA for Affymetrix analysis. Preparation of labeled cRNA, hybridizations to ATH1 Genechip arrays (Affymetrix), and scanning of probe-level scores were carried out at the Keck Foundation Biotechnology Resource Laboratory (Yale University, New Haven, Connecticut). Probe level data from the three ovule arrays, along with probe level data from multiple tissues and developmental stages contained in AtGenExpress (samples ATGE_1, ATGE_10, ATGE_12, ATGE_13, ATGE_14, ATGE_15, ATGE_16, ATGE_17, ATGE_2, ATGE_24, ATGE_25, ATGE_26, ATGE_27, ATGE_28, ATGE_29, ATGE_3, ATGE_31, ATGE_32, ATGE_34, ATGE_35, ATGE_36, ATGE_4, ATGE_40, ATGE_41, ATGE_42, ATGE_43, ATGE_5, ATGE_6, ATGE_73, ATGE_76, ATGE_77, ATGE_78, ATGE_79, ATGE_8, ATGE_81, ATGE_82, ATGE_83, ATGE_84, and ATGE_9) [37], as well as seedling and stigma data from Swanson et al. [38], were normalized using the RMA method [36] as implemented in the affy R package [45], and the mean expression value was calculated for each probe set in each tissue type. The “ovule-specificity factor” (Ovsp) was calculated for each probe set as Ovsp = (Expovule − Expmean)/SD, where Expovule is the mean expression in stage 14 ovules, Expmean is the mean log2 expression value across all tissue types (excluding those that contain stage 12 to stage 14 ovules), and SD is the standard deviation of mean expression values across all tissue types (again excluding tissues containing stage12 or stage 14 ovules). ATH1 probe sets were mapped to the most recent gene annotations (TAIR7) by blasting the probe sequences against the annotated cDNA sequences. A probe set was considered to match a gene if at least six of the 11 probes perfectly matched that gene.
Samples from different tissues and developmental stages were collected from 1-month-old ms-1 plants, except for anther samples, which were collected from wild-type Landsberg plants. For wild-type Columbia, dif1, and myb98 ovary samples, stage 12a flowers were emasculated 24 h before ovary tissue was collected. RNA was purified by RNeasy spin columns (Qiagen). For each tissue, 0.5 μg of RNA was treated with DNase before first-strand cDNA was reverse transcribed using Protoscript II RT-PCR kit (New England Biolabs) and diluted to 100 μl. Each RT-PCR reaction used 2 μl of first-strand cDNA (or 2 μl of no reverse transcriptase control) as template in a 30 μl reaction with 0.33 μM of each primer. Primer sequences are contained in Table S11.
Genomic sequences upstream of embryo sac–dependent genes were PCR amplified and cloned into pCAMBIA-1381Z upstream of the GUS ORF (Table S12). Arabidopsis thaliana plants (Columbia ecotype) were transformed by the floral dip method [48]. Seeds were grown on MS media containing 25 mg/ml hygromycin for 12 d before seedlings with true leaves were transferred to soil. Stage 12c flowers were emasculated 24 h before ovules were dissected into GUS staining solution (50 mM sodium phosphate buffer pH 7.0, 10 mM EDTA, 0.1% Triton X-100, 2 mM potassium ferrocyanide, 2 mM potassium ferricyanide, and 1 mg/ml X-Gluc) on ice before incubation at 37 °C for 45 min (12 h for At5g34885::GUS and At4g24974::GUS). Samples were cleared in 20% methanol/4% concentrated HCl at 55 °C for 15 min followed by 60% ethanol/1.8 M NaOH at 25 °C for 10 min. Samples were washed with 30% ethanol and 10% ethanol and transferred to 50% glycerol for mounting on slides. Samples were viewed on a Zeiss Axioscope using DIC optics, and images were captured on a Zeiss AxioCam HRc digital camera.
Raw and processed microarray data is available at the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo/), with the series identifier GSE8392. Sequences of ovule cDNA reads are available at NCBI dbEST (http://www.ncbi.nlm.nih.gov/dbEST/), with the identifier numbers 45453167–45702604. |
10.1371/journal.pntd.0000589 | Thrichomys laurentius (Rodentia; Echimyidae) as a Putative Reservoir of Leishmania infantum and L. braziliensis: Patterns of Experimental Infection | The importance of the genus Thrichomys in the retention of infection and transmission of Leishmania species is supported by previous studies that describe an ancient interaction between caviomorphs and trypanosomatids and report the natural infection of Thrichomys spp. Moreover, these rodents are widely dispersed in Brazil and recognized as important hosts of other tripanosomatids. Our main purpose was to evaluate the putative role of Thrichomys laurentius in the retention of infection and amplification of the transmission cycle of Leishmania infantum and L. braziliensis. Male and female T. laurentius (n = 24) born in captivity were evaluated for the retention of infection with these Leishmania species and followed up by parasitological, serological, hematological, biochemical, histological, and molecular assays for 3, 6, 9, or 12 months post infection (mpi). T. laurentius showed its competence as maintenance host for the two inoculated Leishmania species. Four aspects should be highlighted: (i) re-isolation of parasites 12 mpi; (ii) the low parasitic burden displayed by T. laurentius tissues; (iii) the early onset and maintenance of humoral response, and (iv) the similar pattern of infection by the two Leishmania species. Both Leishmania species demonstrated the ability to invade and maintain itself in viscera and skin of T. laurentius, and no rodent displayed any lesion, histological changes, or clinical evidence of infection. We also wish to point out the irrelevance of the adjective dermotropic or viscerotropic to qualify L. braziliensis and L. infantum, respectively, when these species are hosted by nonhuman hosts. Our data suggest that T. laurentius may act at least as a maintenance host of both tested Leishmania species since it maintained long-lasting infections. Moreover, it cannot be discarded that Leishmania spp. infection in free-ranging T. laurentius could result in higher parasite burden due the more stressing conditions in the wild. Therefore the tissular parasitism of the skin, infectiveness to the vector, and amplification of the transmission cycle of both Leishmania species could be expected.
| For Leishmania, one genus among several genera belonging to the parasitic Trypanosomatidae family, many nonhuman mammals are known to be hosts in addition to humans. Most studies that describe Leishmania wild reservoirs are based on isolated descriptions of infection that can lead to misinterpretation of information. The definition of the epidemiological importance of a putative reservoir host depends on adequate data on the dynamics and peculiarities inherent to the host-parasite interactions and their involvement in the transmission cycle of these parasites. Our objectives were to sort out the features displayed by nonhuman mammal populations (the caviomorph rodent Thrichomys laurentius) which, with an insect host, perpetuate Leishmania transmission cycles. This rodent species had the ability to act as maintenance and/or amplifier host of both tested Leishmania species. The similar pattern of infection displayed by T. laurentius infected by these two Leishmania species shows that the definition of dermotropic or viscerotropic based on the clinical features observed in humans should not be applied to natural hosts, and it emphasizes that the search for Leishmania reservoirs should consider all possibilities of the infection course, independent of current knowledge in other mammal hosts.
| Although recognized as one of the most important and widespread parasitic diseases in the world, leishmaniasis prevention and control remains a challenge for health authorities in some countries [1]. In Brazil, human cutaneous leishmaniasis occurs in association with different Leishmania species, but Leishmania (Viannia) braziliensis is the most frequent and widespread species in the country. The visceral form is exclusively associated with Leishmania (Leishmania) infantum (syn. L. (L.) chagasi).
The Leishmania genus comprises more than 20 vector-borne species, primarily enzootic parasites, which includes species capable to infect a broad range of mammalian hosts and to be transmitted a variety of phlebotomine vectors. The transmission cycles of Leishmania spp. involves a variety of phlebotomine vectors and mammalian hosts. Failure to interrupt human transmission and prevent new epidemics are related, among others, to the involvement of wild and synanthropic hosts, mainly rodents and marsupials, that can colonize peri-urban areas [2]–[4].
Till now, the majority of studies that point out Leishmania spp. wild reservoirs are based on punctual observations of infection, most of them by molecular methods (PCR) rather than by parasite isolation and characterization. This can conduct to misinterpretation of concepts since the mere finding of Leishmania DNA in a given mammal species is not sufficient to consider this species a reservoir host [5],[6]. Reservoir is better defined not as a single species, but as an assemblage of species responsible for the long lasting maintenance of a parasite in a given environment [7],[8]. This concept does not include target species (human or domestic mammals) neither does it consider the eventual symptoms displayed by the reservoir hosts. Natural Leishmania sp. infection in wild rodents was already reported in different parts of the world [2],[9],[10], and some of them were also successful in demonstrating the persistence of infection up to two years [11]–[13]. Laboratory studies using natural hosts as experimental models provide a suitable indication of the importance of these hosts as reservoirs, since it allows a better understanding of the dynamics of infection, especially concerning the ability to retain the infection and amplify the parasite populations in a given environment, due to a feature that favors the parasite transmission (e.g., presence of parasites in the skin). There are only rare studies that follow up experimentally infected wild hosts by Leishmania species, mostly due to the difficulties of managing wild mammals in captivity.
Thrichomys laurentius is a South American caviomorph rodent formerly included in a monospecific genus. The formerly recognized species, Thrichomys apereoides, was recently split into five species: T. apereoides, T. laurentius, T. pachyurus, T. inermis and T. sp [14],[15]. The recently described species within this genus comprise crepuscular and scansorial rodents that inhabit open vegetation in various Brazilian biomes: savannah (“Cerrado”), white shrub (“Caatinga”) and marshland (“Pantanal”), widespread from western to northern Brazil [16].
Some reasons point to the importance of the Thrichomys genus as a putative reservoir for Leishmania species: 1) the probable ancient association between caviomorph rodents and the trypanosomatids. It was proposed that the entry of new species of Leishmania (Leishmania) subgenus was the consequence of the arrival of infected caviomorph rodents during the Oligocene [17]; 2) the detection of Leishmania sp. DNA in free ranging Thrichomys sp. These rodents were found infected by Leishmania species from different complexes – L. mexicana and L. donovani – in an endemic area for both visceral and tegumentar leishmaniasis in Minas Gerais state, Brazil [3]; 3) the importance of these rodents as reservoirs of other trypanosomatids – Trypanosoma cruzi and T. evansi. This feature is confirmed by both experimental [18],[19] and field work studies [20],[21]; and 4) Thrichomys spp are widely dispersed throughout Brazil, comprising one of the most abundant species in the three Brazilian biomes where they occur [16]. Moreover, they are habitat generalists, found even in degraded areas, and can also frequent human dwellings [20],[22].
In the present work, we investigated the experimental infection of Thrichomys laurentius with Leishmania infantum and L. braziliensis. Our main purpose was to evaluate the putative role of T. laurentius for the retention of infection and amplification of the transmission cycle of these Leishmania species. To achieve this aim, we: (i) studied the differences on the course of infection on L. infantum and L. braziliensis experimentally infected T. laurentius; (ii) followed up the health status of experimentally infected rodents by hematological and biochemical parameters, in order to evaluate the consequence on rodents' health of the experimental infection; and (iii) analyzed the parasitism distribution in the host.
Twenty-four Thrichomys laurentius of both sexes born in captivity were kindly supplied by Dr. Paulo D'Andrea. The colony of T. laurentius was derived from 9 males and 38 females captured in Piauí state (northeast region of Brazil) in 2000. The animals are free from other parasites, provided from food and water ad libitum and kept under conventional conditions (temperature 24±2°C, natural daylight) at animal facilities of the Laboratory of Biology and Parasitology of Small Reservoir Mammals, Oswaldo Cruz Institute. Animals were individually housed in 41-34-17 cm polycarbonate cages with sawdust as bedding and fed with NUVILAB CR1 mouse pellets (Nuvital nutrients S.A., Colombo/PR, Brazil) [23].
The rodents were divided into two groups and intradermically inoculated into the right ear pinna (0.05mL maximum volume) by either Leishmania infantum – MHOM/BR/2001/HP-EMO = IOC-L2504 (n = 12) or L. braziliensis – MHOM/BR/2000/LTCP13396 = IOC-L2483 (n = 12) obtained from the Oswaldo Cruz Institute Leishmania collection (Coleção de Leishmania do Instituto Oswaldo Cruz, CLIOC). At 60 day-old, animals were infected with 106 promastigotes derived from stationary phase culture starting from freshly amastigotes and followed up for 3, 6, 9 or 12 months post infection (mpi). The age of the animals at the time of inoculum was based on calculations from the weightless T. laurentius caught in nature, i.e., when young rodents starts to be exposed to infections outside their nest (personnal observations). The parasites (isolated no more than 2 weeks before Thrichomys infection) were maintained by in vivo passage in golden hamsters (Mesocricetus auratus) derived from the animal facilities of Oswaldo Cruz Foundation. In this case, promastigotes were intradermically inoculated in hamsters footpads and re-isolated from inoculation site (L. braziliensis) and spleen (L. infantum) 4–5 months after infection. Hamsters were also used for control of the infectivity of the inocula.
The study design was carried out according to the protocol approved by the Institutional Committee for Experimentation and Care of Research Animals (CEUA-Fiocruz: P0076/01 and P0269/05) and animal facilities are supported by the Brazilian Institute of Environment and Renewable Natural Resources (IBAMA license 02022.002062/01-04). The present study was conducted from November 2005 to December 2008.
Blood samples were collected in heparinized and nonheparinized tubes from the retro-orbital plexus of animals previously intramuscularly (IM) anesthetized with 100 mg/kg ketamine hydrochloride and 2 drops of local anesthesia with colirium (0.5% solution of proximetacaine chloridrate) every 3 weeks. The following parameters were determined: (i) red (RBC) and white (WBC) blood cell count, using a Neubauer hemocytometer; (ii) hematocrit, by the centrifugation of microcapilar tubes; (iii) hemoglobin levels, using a commercial test kit (Labtest, Lagoa Santa/MG, Brazil); (iv) percentage of leukocyte cells, by microscopical observation of thin blood smears stained with Panótipo Rápido (derived from the Romanowski stain). Heparinized blood was also collected onto filter paper (Whatman 5, Maidstone, UK) for the molecular assay, while the serum obtained from non-heparinized blood samples was used for the biochemical and serological follow-up. Medium corpuscular volume (MCV), medium corpuscular hemoglobin (MCH) and medium corpuscular hemoglobin volume (MCHV) were also calculated. Values of all parameters obtained for each group one-day before the inoculum were considered normal and used to calculate the reference values.
The ability to produce nitric oxide was evaluated by the nitrite level in rodent sera, using the Griess Reagent System (Promega, Madison, USA). Only for rodents infected by L. infantum, albumin and total protein levels were determined using commercial test kits (Labtest, Lagoa Santa/MG, Brazil). All of these assays were done according to the manufacturers recommendations.
The kinetics of the humoral immune response was evaluated by indirect immunofluorescence test (IFAT) and enzyme-linked immunosorbent assay (ELISA) using Leishmania antigen deriving from axenic promastigotes of the same strain used for the experimental infection (homologous) and/or deriving from a mixture of L. infantum and L. braziliensis formalin-treated promastigotes (mix), the latter only for the rodents followed for more than 6 months. IFAT was performed assaying two-fold sera dilutions (1∶10–1∶1,280) against Leishmania promastigotes and the reactions conducted using a specific in-house intermediary antibody anti-Thrichomys sera produced in rabbits. The reaction was visualized using a commercial anti-rabbit IgG-FITC (Sigma-Aldrich, St. Louis, USA), according to Camargo [24]. Standard micro-ELISA was conducted according to Rosario et al. [25], using a commercial anti-rat IgG-peroxidase (Sigma-Aldrich, St. Louis, USA). We established 1∶20 and 1∶30,000 for the sera and conjugate dilutions, respectively, after the analysis of different serum dilutions derived from experimentally infected and non infected captivity Thrichomys using the ROC Curve (BioStat 5.0 software). For each rodent, the cut-off value was determined using sera collected before the experimental infections and the absorbance at 492 nm was measured in an EMax Microplate Reader (Molecular Devices, Ramsey, USA).
The DNA extraction from filter paper was conducted by boiled water, according to Marques et al. [26]. The DNA product amplifications were conducted using pureTaq Ready-To-Go PCR beads (Amersham Biosciences, Buckinghamshire, UK) and primers directed to the conserved region of the Leishmania kDNA minicircle, as follows: forward: 5′-GGGGAGGGGCGTTCTGCGAA-3′ and reverse: 5′-GGCCCACTAT ATTACACCAACCCC-3′. The PCR conditions were as follows: initial denaturation at 94°C for 5 min, followed by 30 cycles at 94°C for 1 min, 60°C for 1 min, 72°C for 30 s, and a final extension at 72°C for 5 min [27]. Blood from uninfected rodents and uninfected blood samples to which promastigotes axenically cultured were added, were used as control for both extraction and amplification processes. The amplified polymerase chain reaction (PCR) products were analyzed in polyacrylamide gel electrophoresis (4%) and the negative samples were re-analyzed by electrophoresis in 12.5% polyacrylamide gels using the Genephor electrophoresis system apparatus (Pharmacia Biotech). All of the gels were stained using the DNA Silver Staining Kit (GEHealthcare, Chalfont St. Giles, UK).
Euthanasia was performed by CO2 inhalation on months 3, 6, 9 and 12 post inoculation (n = 3 for each batch). Procedures were undertaken in a Class II biosafety cabinet: (i) inoculation of fragments of spleen, liver, inoculation site (right pinna skin) and bone marrow in biphasic culture mediums (NNN/Schneider's) supplemented with 10% fetal bovine serum (v/v) and antibiotics (350 IU penicillin and 150 µg/mL streptomycin), which was examined every 3–4 days for 1 month; (ii) slide imprints of spleen, liver and inoculation site, which were Giemsa-stained and microscopically observed at ×400 magnification; (iii) collection of tissue fragments – spleen, liver, inoculation site, skin and bone marrow – in 1.5 mL tubes containing ethanol and stored at −20°C, which were used for the molecular analyses; (iv) fixing of tissue fragments – spleen, liver, skin, lymph nodes and both ears separately – in 10% neutral buffered formalin for histological studies. After dehydration and paraffin-embedding, 4 µm sections in thickness were made, routinely stained with hematoxylin and eosin (H&E), and the sections examined by light microscopy. For the slide imprints, histological and molecular tests, liver tissue samples were performed considering two fragments from different lobes.
For the molecular diagnosis, tissue fragments were washed three times with Milli-Q water and DNA extraction realized using the Wizard Genomic DNA Purification Kit (Promega, Madison, USA) according to the manufacturer's recommendations. The PCR was conducted as described above for the blood collected on filter paper.
Normal ranges for the hematological and biochemical values were determined in relation to medium values and two-fold standard errors obtained for each group one-day before the inoculum. The differences on the hematological and biochemical kinetics between rodents infected by either L. braziliensis or L. infantum were evaluated by the non-parametric Mann-Whitney test. The differences between the normal values and each point of the hematological and biochemical follow-up were evaluated by the Kruskal-Wallis and Student-Newman-Keuls tests. All of the data were analyzed using the BioStat 5.0 software (Instituto Mamirauá, Tefé, Brazil) and considering p<0.05 significant.
Thrichomys laurentius was able to control and retain the infection for both inoculated Leishmania species (L. braziliensis and L. infantum), and four aspects should be highlighted: (i) the long-term retention of T. laurentius infection, at least 12 months; (ii) the low parasitic burden displayed by T. laurentius tissues on the necropsy; (iii) the early onset and maintenance of important humoral response, demonstrated by significant serological titers; and (iv) the similar pattern of infection displayed by T. laurentius infected by these two Leishmania species usually associated to distinguishable manifestations of human disease. This latter aspect was mainly demonstrated by the parasite recovery from internal viscera and detection of Leishmania DNA in all sampled tissues of L. braziliensis infected rodents. This is the first report of an experimental infection in a putative wild rodent reservoir species, where L. braziliensis and/or L. infantum could be re-isolated.
The Leishmania sp. inoculated were shown to be infective as demonstrated by the parasite recovery from liver, spleen and inoculation site of hamsters. However, only L. braziliensis could be isolated from the inoculation site. As expected for outbred animals, a great individual variability among infected T. laurentius was noted. Despite that, all Thrichomys rodents were able to efficiently control the infection without presenting lesion or clinical evidence of disease. Growth development, determined by weekly body mass measure, was not affected by the Leishmania infection and no expressive alterations of health markers were observed. Amastigotes of Leishmania spp. were absent in the thin blood smears and Leishmania DNA could not be detect in blood samples collected in filter paper in 3 weeks interval.
Leishmania infection did not result in anemia and all of the rodents displayed values that were inside the normal range during the complete follow-up. Nevertheless, L. infantum infected T. laurentius tended to display lower red blood cell counts and hemoglobin levels when compared to those infected by L. braziliensis. During the follow-up, most of the values obtained for RBC counts and hemoglobin levels showed significant differences (p<0.05) between T. laurentius infected by L. braziliensis and L. infantum (Figure 1). This same feature was also observed for the hematocrit values (data not shown). Significant leucopenia from the 120 dpi day on (p<0.05), was observed in rodents inoculated with L. braziliensis. A similar, but not significant, picture was also observed in the rodents infected by L. infantum (Figure 2). No differences before and after the inocula were observed for MCV, MCH, MCHV and differential counts of WBC (data not shown).
Albumin and total protein levels were not affected by L. infantum infection, excepting for the rodent (7548) where re-isolation of parasites was possible. This rodent displayed a marked decrease in albumin levels and increase in total protein levels after 200 dpi, resulting in declined in albumin/total protein level (Figure 3). The nitrite level in rodent sera displayed a great individual variability and could not be correlated to any other hematological, serological or parasitological parameter.
All infected T. laurentius were able to produce a humoral response that could be detected during the three initial weeks by both IFAT and ELISA assays. The IFAT showed no differences on serological titers among assays performed with homologous or a mixture of antigens. The response onset and magnitude of titers were very homogeneous and similar among the infected rodents. Rodents infected by L. braziliensis displayed serological titers that were always slight higher than those observed for the rodents infected by L. infantum (Figure 4). The ELISA assay revealed an individual variability on the response onset and magnitude of titers that varied from negligible to strong responses. In common, infected T. laurentius displayed a peak of absorbance values on 100 dpi that were in medium four times higher than the day 0 and kept constant until the end of follow-up (data not shown).
Both Leishmania species demonstrated the ability to invade and maintain itself on viscera and skin of the infected T. laurentius, although this parasitism was not expressive since isolation of parasites was rare: from liver, spleen and inoculation site 3 mpi from one rodent infected by L. braziliensis; and from liver and spleen 12 mpi from one rodent infected by L. infantum.
Leishmania DNA was detected in all experimental batches, independent of the Leishmania species (Table 1). Nevertheless, the low parasitic burden was evidenced by the large amount of positive PCR (73%) observed only when electrophoresis was conducted on the GenePhor electrophoresis system apparatus. Up to 77% and 50% of the L. braziliensis and L. infantum infected T. laurentius, respectively, displayed Leishmania DNA in at least one of the tissues collected on the necropsy. The individual variability, peculiarities of the host, parasite and host-parasite interaction, and the time of the infection seems to be the major factors that influenced the different percentage of positive reactions. Parasite distribution in viscera was not homogeneous and 30.4% of the tissues (spleen, liver and inoculation site) that had two different fragments examined, displayed a positive and a negative result for the presence of Leishmania DNA.
No sign of parasitism was observed in tissue imprints or histological sections. Comparative histologic analysis did not detect any inflammatory or degenerative changes in T. laurentius infected with L. infantum or L. braziliensis. Our study on the pathology of Leishmania sp. in golden hamster (Mesocricetus auratus) demonstrated that this rodent developed definite evidence of infection, characterized by extensive spleen necrosis and inflammation associated to high number of amastigotes (Figure 5A). No histological abnormalities or other histological differences were observed between positive and negative culture tissues obtained from infected T. laurentius (Figure 5B–F). Discrete differences in the cellularity of primary splenic follicles and periarterial lymphoid sheath seem not related to infection and were seen in seven T. laurentius from both groups.
The genus Thrichomys comprises recently described cryptic species that are undergoing a process of allopatric and/or parapatric differentiation [14]. Within this widespread rodent genus, T. laurentius is distributed in northeast Brazil, from Ceará to Bahia state, a region that reports numerous cases of both visceral and tegumentar human leishmaniasis [28],[29]. For demonstrating the potential to act as a maintenance host, a given mammal species must be able to control and retain the parasite infection. In our experimental conditions this rodent species showed to be able to retain long term infections by the main etiological agents of human leishmaniasis in Brazil, Leishmania infatum and L. braziliensis. This ability was undoubtedly demonstrated by the parasite re-isolation in liver and spleen of rodents experimentally infected by both Leishmania species.
Asymptomatic infection is usually considered as an essential attribute to be considered a reservoir host. This is currently not considered as a rule; in fact, it is the transmission strategy of the parasite that is positively selected in a successful host-parasite system, independent of the damage caused by the parasite or health status displayed by the host. Even ancient host-parasite interaction may not necessarily evolve in the direction of less damage or lower virulence, but instead of that, to the maximum transmissibility of the parasite [30],[31]. According to the concept proposed by McMichael [32] and Roque et al. [19], maintenance host is the one who retain the infection (where a given parasite persists) while an amplifier host displays an infection course that favors the transmissibility of the parasite. Taken together our data suggest that T. laurentius may act at least as a maintenance host of both tested Leishmania species since it maintained long-lasting infections. Moreover, it cannot be discarded that in nature, infected rodents display higher parasite burden and tissular parasitism on skin, acting then as an amplifier hosts of Leishmania species.
Anemia, a characteristic trait in L. infantum infected humans, dogs and laboratory rodents [33]–[35], was not observed in the L. infantum infected T. laurentius. Although animals infected by L. infantum displayed a significant decrease of the hematological parameters in comparison to those infected with L. braziliensis, this decrease still did not characterize anemia. Leucopenia, another common trait observed on L. infantum, but not on L. braziliensis infections [36], was only observed in the L. braziliensis infected T. laurentius from the 120 dpi onwards. This finding was not surprising in the light of the parasite disseminated to liver and spleen. Surprising was (i) the absence of leucopenia in L. infantum infected T. laurentius; and (ii) the later presence of that leucopenia, only after 120 dpi, while the L. braziliensis isolation occurred before that.
Hypoalbuminemia and hypergamaglobulinemia, the most common biochemical alterations in L. infantum symptomatic infection in humans and dogs [34],[37], were only observed on the rodent from which the re-isolation of parasites was possible, probably due to a higher parasite burden in this animal. Considering that biochemical alterations are not described as being associated to L. braziliensis infections, these parameters were not tested in the animals infected by this Leishmania species. Moreover, given the similar pattern observed in the rodents infected with both Leishmania species we question whether analyses of further parameters could display alterations in the animal from which L. braziliensis was isolated in the necropsy.
Experimental T. laurentius infection by two different Leishmania species did not result in important damage for rodents, but this can be quite different for naturally infected T. laurentius. Captivity rodents are free from other pathogens, are provided food and water ad libitum and maintained in controlled environmental conditions. In nature, rodents are constantly facing out stress (search for food and escape from predators), competitions (intra and inter-specific), and infection by other parasites and Leishmania re-infections. All of these factors will directly influence the course of any parasitic infection and be reflected by higher virulence and/or host damage.
The effectiveness of the serological assays was demonstrated even for the rodents infected by L. braziliensis, an infection usually not associated to an important humoral response [36],[38]. This is probably due to the visceralization of this parasite species in T. laurentius. This data emphasizes that the search for Leishmania reservoir should consider all possibilities of the infection course, which includes a broad range of diagnostic methods independent of the current knowledge in other mammal hosts. The efficacy of the IFAT assay for serological screening of Thrichomys sp. was already demonstrated for Trypanosoma cruzi and T. evansi infections [19],[39]. In the present study, ELISA showed to be a promising tool, since it was able to detect a humoral response production in all of the infected rodents. The use of an intermediate anti-Thrichomys antibody and the determination of cut-off values based on a great number of positive and negative serum samples might result in a standardized and efficient assay to diagnose Leishmania infection in wild Thrichomys sp.
In this study, we were not able to detect Leishmania DNA in any of the blood samples examined; even considering that 24 infected T. laurentius were analyzed and blood samples were collected every 21 days post infection, totalizing 236 samples evaluated. These data show that whole blood is not a reliable sample to detect Leishmania infection, at least in this mammal host species. The persistence of both Leishmania species with an extremely low burden in T. laurentius could only be demonstrated by the use of a more sensitive technique: PCR targeting a high copy number DNA sequence coupled to a high resolution electrophoresis (in this study, the Genephor electrophoresis system). Unfortunately, the elevated cost of some commercial kits makes the routine use still unfeasible.
In T. laurentius, L. braziliensis can invade and maintain itself in other tissues in addition to the skin. The parasite's ability to invade and maintain itself in internal organs, such as spleen and liver, in non-human hosts was described several times since the 1950ths [40],[41]. Despite that, the description of L. braziliensis as a dermotropic parasite is widespread throughout the scientific community. Our results demonstrated that the definition of dermotropic or viscerotropic based on the clinical feature observed in humans should not be applied to the natural hosts of that Leishmania species.
Studies based only on molecular probes are successful to determine parasite hosts, but lack the capacity to determine the transmissibility of that parasite, and thus the importance of that putative reservoir host on the transmission cycle. Moreover, contamination in isolation attempts in field conditions seriously hampers the successful isolations. For that reasons only few studies were capable to isolate Leishmania parasites in naturally infected wild rodents [2],[9],[42]. The polymerase chain reaction (PCR) methodology is undoubtedly a great advance for the diagnosis of Leishmania infection, but it cannot be associated to parasite transmissibility.
The scarce studies on the L. infantum experimental infection of wild rodents report the failure to re-isolate the parasite [43]–[45], and only in one of them, parasite DNA could be detected [45]. We were able to re-isolate L. braziliensis and L. infantum from experimentally infected T. laurentius. Moreover, we detected L. infantum DNA in bone marrow samples of another species of Thrichomys, T. pachyurus, one year after the experimental infection (unpublished data). The ability to maintain and disseminate to different organs (which include bone marrow, spleen, liver and skin) during long term infections by Leishmania species and their wide and abundant distribution in Brazilian endemic leishmaniasis areas point to the importance of Thrichomys spp. at least as maintenance host for Leishmania species. Future studies concerning the natural infection of Thrichomys spp. becomes crucial to understand the role of these caviomorph species on the wild transmission cycles of Leishmania species.
|
10.1371/journal.pcbi.1006798 | Multi-Cell ECM compaction is predictable via superposition of nonlinear cell dynamics linearized in augmented state space | Cells interacting through an extracellular matrix (ECM) exhibit emergent behaviors resulting from collective intercellular interaction. In wound healing and tissue development, characteristic compaction of ECM gel is induced by multiple cells that generate tensions in the ECM fibers and coordinate their actions with other cells. Computational prediction of collective cell-ECM interaction based on first principles is highly complex especially as the number of cells increase. Here, we introduce a computationally-efficient method for predicting nonlinear behaviors of multiple cells interacting mechanically through a 3-D ECM fiber network. The key enabling technique is superposition of single cell computational models to predict multicellular behaviors. While cell-ECM interactions are highly nonlinear, they can be linearized accurately with a unique method, termed Dual-Faceted Linearization. This method recasts the original nonlinear dynamics in an augmented space where the system behaves more linearly. The independent state variables are augmented by combining auxiliary variables that inform nonlinear elements involved in the system. This computational method involves a) expressing the original nonlinear state equations with two sets of linear dynamic equations b) reducing the order of the augmented linear system via principal component analysis and c) superposing individual single cell-ECM dynamics to predict collective behaviors of multiple cells. The method is computationally efficient compared to original nonlinear dynamic simulation and accurate compared to traditional Taylor expansion linearization. Furthermore, we reproduce reported experimental results of multi-cell induced ECM compaction.
| Collective behaviors of multiple cells interacting through an ECM are prohibitively complex to predict with a mechanistic computational model due to its highly nonlinear dynamics and high dimensional space. We introduce a methodology where nonlinear dynamics of single cells are superposed to predict collective multi-cellular behaviors through a developed linearization method. We represent nonlinear single cell dynamics with linear state equations by augmenting the independent state variables with a set of auxiliary variables. We then transform the linear augmented state equations to a low-dimensional latent model and superpose the linear latent models of individual cells to predict collective behaviors that emerge from multi-cellular interactions. The method successfully reproduced experimental results of cell-induced ECM compaction.
| Cell-induced compaction of fibrous extracellular matrix (ECM) is an important mechanism for numerous processes such as wound healing and tissue development [1–3]. During wound healing, for example, traction forces exerted by fibroblasts and myofibroblasts result in ECM compaction at the site of injury [2, 3]. In vitro experiments using cell-populated collagen gel reveal global compaction of the matrix as a result of cooperative effect of multiple cells at the boundaries as well as propagation through the bulk [4–6]. Furthermore, matrix densification is observed in the regions around [7] and in-between cells. Here we examine the mechanical aspect of intercellular communication through the ECM and how contractile cells can induce emergent mechanical changes leading to matrix compaction. From a simplified mechanics point of view, compaction results when the traction forces exerted by the contractile cells embedded within the ECM overcome the resistive forces of the ECM structure, including viscoelastic forces and elastic energy forces. As a result the matrix is deformed from its original stress-free state and the elastic modulus increases [4–7].
In reality, the compaction process is far more complex. The ECM forms a network of cross-linked fibers that is highly nonlinear and intricate, but is critical for predicting large compaction and long-range transmission of forces [4]. As a large deformation is induced by contractile cells, the standard linear mechanics model yields substantial errors. The ECM fiber network is anisotropic and causes irreversible deformations as a large compaction takes place. This prominent nonlinearity prohibits use of simple methods for predicting the ECM compaction by a multitude of cells. In addition, cells can internally modulate their state in response to local mechanical stresses within the ECM, which influences cell polarity, contractility, stiffness and strength of focal adhesion’s [8–11]. These cell properties are highly nonlinear and complex. Consideration of these nonlinear physical and physiological properties involved in the cell-ECM mechanics often result in differential equations that are intractably complex due to high-dimensional, nonlinear coupled dynamics.
Many in silico modeling approaches in the areas of wound healing and fibrotic disease have helped elucidate and explore the underlying phenomena involved in cell-induced ECM compaction, and have been used to supplement in vitro experiments for fast and inexpensive methods of evaluation. Approaches in previous works include: i) a hybrid continuum-discrete framework consisting of the macroscopic finite element domain and local microscopic fiber network [12], ii) rule-based models with deformable cells and ECM fibers to explain matrix remodeling and durotaxis [13, 14], iii) a discrete fiber model of cell populated fibrous matrix [15], and iv) continuum models of ECM gel compaction [7, 16, 17]. Even though these works provide many insights, they also simplify the ECM gel compaction mechanism by: a) 2-D representation of a 3-D system, b) exclusion of intracellular mechanics such as mechanobiology of actin stress fibers, focal adhesions, and remodeling of cellular and nuclear membranes, and c) consideration of linear elastic spring model of ECM fibers without including the viscoelastic nature of the fibers. Consequently, these prior models abstract detailed cell-ECM interactions, resulting in limitations to understanding how these interactions enable characteristic gel compaction. In addition, models examining the complex dynamics surrounding cell morphology, contractility, and polarity based on finite element methods do exist for 2 dimensional cases [10]. And, focal adhesion-stress fiber dynamics have been modeled for 2-D PDMS substrates based on non-equilibrium thermodynamics [11].
In the current work, the ECM is modeled as a 3-D cross-linked network of discrete, viscoelastic fibers, and detailed mechanistic cell dynamics, including focal adhesion dynamics, cytoskeleton remodeling, actin motor activity and lamellipodia protrusion, are derived from basic principles. The resultant model is computationally complex, especially for a larger number of cells. The governing differential equations are highly nonlinear, coupled, and of high dimension. Here, we solved this difficulty by introducing a methodology having its disciplinary basis spanned in system dynamics, machine learning, and statistics.
It is known that a nonlinear system can behave more linearly when recast in a larger space [18]. In our approach, the original nonlinear dynamics derived from physical and physiological principles are recast in an enlarged state space by augmenting independent state variables with auxiliary variables that inform input-output characteristics of the nonlinear elements involved in the system. Once recast in the augmented space, the nonlinear system can be represented as an augmented set of linear dynamic equations. The linear representation facilitates model reduction using latent variable analysis, which can be shown is difficult to apply to highly nonlinear systems [19–22]. Furthermore, linearization in the augmented space allows for superposition of multiple subsystems. In the current work, collective behaviors of multiple cells are predicted via superposition of single cell subsystems through the linearization in the augmented state space. The proposed methodology is general, and is applicable to a broader class of problems where large-scale, collective behaviors must be predicted while retaining sufficient mechanistic details.
We construct a computational model for predicting cell-mediated gel compaction by multiple (ncell) cells having a uniform phenotype and interacting through a surrounding 3-D ECM fiber network. The ECM is modeled as a network of many fibers connected at a large number of nodes (Ne ≈ 2000), whereas each cell is represented with a mesh structure consisting of multiple nodes (Nc ≈ 200) which forms the cell outer membrane (see Fig 1A). The cell outer membrane deforms and gains traction as the nodes on the membrane bond to the nodes of the surrounding ECM fiber network and form focal adhesions, which occur when bonding molecules (or integrins) on the cell membrane bind to ligands on ECM.
Consider the i-th outer membrane node of the k-th cell with three dimensional spatial coordinates x i c , k ∈ ℜ 3 × 1 (See Fig 1B). The forces acting on it include the cell’s cortical tension force and elastic energy force (collectively denoted as F C o r t - E l a s , i c , k ∈ ℜ 3 × 1), focal adhesion force (denoted as F F A , i c , k ∈ ℜ 3 × 1), lamellipodium force (F L , i c , k ∈ ℜ 3 × 1), and frictional damping force (F D a m p , i c , k ∈ ℜ 3 × 1) [23, 24]. Assuming that the mass of the node is negligibly small and the damping force is given by F D a m p , i c , k = - D c d x i c , k / d t, where Dc is damping constant, the equation of motion is given by:
F C o r t - E l a s , i c , k + F F A . i c , k + F L , i c , k - D c d x i c , k d t = 0 i = 1 , ⋯ , N c , k = 1 , ⋯ , n c e l l (1)
The generation of lamellipodium force pertains to the polarity of the cell. Namely, lamellipodia extend in a particular direction of the cell determined by the cell’s polarity [23–26]. The cell polarity and the lamellipodium forces can be treated as a cell’s decision or, in the system dynamics terminology, control inputs. Let x c , k = ( x 1 c , k T ⋯ x N c c , k T ) T ∈ ℜ 3 N c × 1 be a vector containing the 3-D coordinates of all the cell membrane nodes. Here the superscript in XT represents the transpose of matrix or vector X. The above equation of motion can be written collectively as:
d x c , k d t = W C E c F C o r t - E l a s c , k + W F A c F F A c , k + L c u k k = 1 , ⋯ , n c e l l (2)
where F C o r t - E l a s c , k ∈ ℜ 3 N C × 1 is a vector comprising cortical tension and elastic energy forces for all the cell nodes (i = 1, ⋯, NC), F F A c , k ∈ ℜ 3 N C × 1 is a vector of focal adhesion forces at all the cell nodes, u k ∈ ℜ 3 N C × 1 is an input vector containing all the lamellipodium forces (F L , i c , k), and W C E c , W F A c and Lc are constant matrices of consistent dimensions.
The equation of motion of the surrounding ECM fiber network can be represented in a similar manner. The forces acting on the j-th node of the fiber network are the elastic energy forces, including both lateral restoring forces and the one associated with bending moments, (F E l a s , j e ∈ ℜ 3 × 1), focal adhesion forces from the shared attachment with the cell (F F A , j e ∈ ℜ 3 × 1) and damping forces (F D a m p , j e ∈ ℜ 3 × 1) [23–26]. The equation of motion can be written as:
F E l a s , j e + F F A , j e - D e d x j e d t = 0 , j = 1 , ⋯ , N e (3)
Let x e = ( x 1 e T ⋯ x N e e T ) T ∈ ℜ 3 N e × 1 be a vector containing the 3-D coordinates of all the ECM nodes. Then Eq 3 can be written as:
d x e d t = W E l a s e F E l a s e + W F A e F F A e (4)
The ECM elastic energy force is a nonlinear function of ECM coordinates xe. The cortical tension and elastic energy force of the k-th cell is a nonlinear function of its membrane coordinates xc,k. Here xe and xc,k are independent state variables of the multi-cell ECM system.
F E l a s e = F E l a s e ( x e ) , F C o r t - E l a s c , k = F C o r t - E l a s c , k ( x c , k ) k = 1 , ⋯ , n c e l l (5)
The focal adhesion force is modeled as a stochastic binding process between nodes on the cell membrane and those on the ECM. Using Monte Carlo simulations it has been found that focal adhesion forces can be approximated to a nonlinear algebraic function of cell membrane and ECM nodes as well as the biochemical parameters involved in integrin-ligand binding [23, 24].
F F A c , k = F F A c , k ( x c , k , x e ) , k = 1 , ⋯ , n c e l l (6)
Assuming that no two cells bind to the same ECM node, we can find that the focal adhesion force of the i-th membrane node of the k-th cell attached to the j-th ECM node must satisfy:
F F A , i c , k + F F A , j e = 0 (7)
Namely, F F A , i c , k and F F A , j e have the same magnitude with the opposite signs. Therefore, all the focal adhesion forces of the k-th cell can be mapped to the corresponding ECM nodes. Collectively, the focal adhesion forces of all the nodes within the ECM may be written as:
F F A e = ∑ k = 1 n c e l l P m a p k F F A c , k (8)
where P m a p k ∈ ℜ 3 N e × 3 N c is a parameter matrix (consisting of either 0 or -1 elements) that maps the membrane focal adhesion forces of the individual cells (F F A c , k , k = 1 , … , n c e l l) to the corresponding ECM focal adhesion forces (F F A e). The focal adhesion connections between the membrane nodes and ECM nodes change over time as the cell membrane deforms, gains traction and generates lamellipodial protrusions. Therefore, the mapping matrix P m a p k is updated at each time step. Details on the formation and structure of P m a p k are given in the Methods Section. The ncell cells interact with each other through the surrounding ECM by generating focal adhesion forces, which propagate through the ECM fiber network and influence the other cells. The resultant collective behavior of the multiple cells is complex due to coupled, nonlinear dynamics.
Although the governing equations derived above are rigorous and based on basic principles, they are complex and can become computationally expensive as the number of cells increases. Computational complexity is a key challenge in predicting collective behaviors of multiple cells. The number of state variables for the given system is 3Ne + 3Nc ncell, which is on the order of 7,000 for ncell = 2 and 9000 for ncell = 5. We aim to transform the governing equations into a linear latent variable representation in order to considerably reduce the number of state variables but also facilitate prediction of collective behaviors of the multiple cells through superposition of individual cell dynamics.
Model reduction is a challenging problem particularly for highly nonlinear, dynamical systems [21, 22, 27–29], as in the presented problem of collective behaviors of multiple cells within an ECM. If the system is linear or near linear, model reduction is more amenable and simple methods, such as Principal Component Analysis and Partial Least Squares, can reduce dimensionality. Here, we propose a unique linearization method, termed Dual-Faceted (DF) Linearization, and then apply a model reduction method to the linearized model. In DF Linearization, we represent the nonlinear dynamical system in an augmented space consisting of independent state variables (xe and xc,k) and nonlinear forces (F E l a s e , F C o r t - E l a s c , k , F F A c , k) as the additional variables, termed auxiliary variables. Standard linearization, such as Taylor series expansion, is limited in accuracy, which may be valid only in the vicinity of a reference point. In DF Linearization, instead of taking “algebraic” linearization of these nonlinear terms, we consider “dynamic” linearization by representing their dynamic transitions using linear regressions.
Before formally introducing DF Linearization, let us consider a simple example that delineates the basic principle of the method. Suppose that the system consists of one spring and a damping element with negligibly small mass, F - D x ˙ = 0. If the spring is a linear spring, F = kx, there is absolutely no difference between the equation in terms of state variable x, x ˙ = ( k / D ) x, and the one in force F, F ˙ = ( k / D ) F. However, it is not the case if the spring is nonlinear, for example a hard spring: F = ax + bx3 where a > 0, b > 0. Representing the differential equation in two variables, one with the state variable x and the other with the auxiliary variable F, provide different equations.
d x d t = a D x + b D x 3 (9) d F d t = 1 D ( a + 3 b ( g ( F ) ) 2 ) F (10)
where x = g(F) s the inverse function of F = ax + bx3. Both equations represent the same nonlinear system, yet the expressions are different, hence Dual-Faceted representations. Linearizing these differential equations lead to two linear differential equations viewed from the augmented space. Note that Eq 9 can be represented as a linear equation by using both state and auxiliary variables:
d x d t = W F F (11)
where WF = 1/D. The augmented state Eq 10 can be approximated to a linear regression:
d F d t ≃ S x x + S F F (12)
where Sx, SF are regression parameters.
The expression given by Eq 12 differs from the one based on the first order Taylor expansion (or “algebraic” linearization) which yields:
F ( x ) ≃ F ( x ¯ ) + d F d x Δ x (13)
Furthermore, if we evaluate the derivative J(x)≡dF/dx at a particular point, J ¯ ( x ¯ ), and then use Eq 13 to express the augmented state equation, it reduces to F ˙ = J ¯ x ˙. This implies that F ˙ and x ˙ are proportional. Using an “algebraic” linearization yields a differential equation representing the transition of F that is collinear to the one representing the transition of x, and thereby an auxiliary state equation would not provide any new information. Conversely, the regression model in Eq 12 provides us with a diverse view of the original nonlinear system, thus providing a richer representation of the nonlinearity than the standard first order Taylor expansion.
Applying the above principle of Dual Faceted Linearization to our problem, we note that the original state equations governing the transition of the independent state variables, 2 and 4, are linear in the augmented state space. All we need is to obtain the transition of the auxiliary variables. Let the regression of the dynamic transition of auxiliary variable F E l a s e, be expressed as:
d F E l a s e d t ≈ R x e x e + R F E l a s e F E l a s e (14)
where R x e , R F E l a s e ∈ ℜ 3 N e × 3 N e are parameter matrices. If an “algebraic” linearization using the Jacobian J ¯ = ∂ F E l a s e / ∂ x e | x ¯ e was utilized, the above equation would be: d F E l a s e / d t = J ¯ · d x e / d t .
This state transition equation through “algebraic” linearization is equivalent to one of the original independent state Eq 4 because d F E l a s e / d t and d x e / d t are collinear within this formulation which renders it redundant. In contrast, the state transition equation presented in Eq 14 is not collinear, providing a diverse facet of the nonlinear system. Similarly, for the auxiliary variables F C o r t - E l a s c , k , F F A c , k, let the regression equations be written as:
d F C o r t - E l a s c , k d t ≃ Q x c x c , k + Q F C E c F C o r t - E l a s c , k + Q u u k (15) d F F A c , k d t ≃ H x c x c , k + H x e x e + H F F A c F F A c , k + H u u k (16)
where Q ⋆ ⋆ , H ⋆ ⋆ (⋆ -corresponding subscript and superscript) are parameter matrices with consistent dimensions. The high-dimensional parameter matrices (R ⋆ ⋆ , Q ⋆ ⋆ , H ⋆ ⋆) do not need to be determined explicitly as discussed in the subsequent sections. DF Linearization represents a nonlinear dynamical system with two sets of differential equations. One set is the original state equations governing the transition of the independent state variables and the other set is the regression of the dynamics of auxiliary variables. The original state Eqs 2 and 4, are apparently linear in terms of the auxiliary variables and inputs. In these equations, all the forces acting on each node sum to zero. These are linear expressions when the nonlinear forces are treated as auxiliary variables. In addition, the auxiliary state transitions (Eqs 14–16) are given by linear regressions in the augmented space. Therefore, both differential equations are linear. The two linear differential equations represent different (or dual) facets of the original nonlinear system viewed from the augmented space, thus providing a richer representation of the nonlinearity.
Now that the original nonlinear system has been represented as a linear dynamical system in the augmented space, we can apply a latent variable modeling method to reduce model order. Represented in the augmented space, the differential equations may contain similar modes, or some variables are close to collinear. These similar modes and collinear variables can be eliminated by model order reduction methods.
Let ζ c , k be the augmented variable vector containing membrane node coordinates and forces of the k-th cell.
ζ c , k = ( x c , k F C o r t - E l a s c , k F F A c , k ) ∈ ℜ 9 N c × 1 (17)
Here uk (the cell’s lamellipodial force) is treated as input variables that are excluded from the augmented state space. Similarly, let ζe be the augmented variable vector containing ECM node coordinates and forces:
ζ e = ( x e F E l a s e ) ∈ ℜ 6 N e × 1 (18)
Focal adhesion forces F F A e are determined by the individual cells by Eq 8 and, thereby, excluded from the augmented space of the ECM.
We apply latent space analysis to vectors ζc,k and ζe, respectively. First we generate data by using Eqs 2 and 4–7. Computation of the nonlinear state equations is amenable for a single cell interacting with ECM. A data set can be created by simulating those nonlinear equations by placing a single cell at diverse locations, i.e. repeating the simulation with different initial conditions. Let C ζ ζ c , C ζ ζ e be the covariance matrices of simulation data sets of augmented states ζc,k and ζe, respectively. Each covariance matrix contains both independent state and auxiliary variables, where the latter is nonlinear functions of the former. If auxuliary variables were linear functions of the state variables, then the rank of the covariance matrix would be equal to the number of independent state variables. However due to the nonlinearity, the rank is higher. Details on the formation of C ζ ζ c , C ζ ζ e are given in the Methods Section.
The covariance analysis also reveals that the system represented in the augmented space contains many components that may be negligibly small. Using Principal Component Analysis, the original data of ζc,k and ζe can be represented with latent variables of truncated dimension mc ≪ 9Nc and me ≪ 6Ne, respectively [29]:
ζ c , k = ( V x c V F C E c V F F A c ) ︸ V c z c , k , ζ e = ( V x e V F E l a s e ) ︸ V e z e (19)
where matrices V c ∈ ℜ 9 N c × m and V e ∈ ℜ 6 N e × m are orthogonal matrices comprising the eigenvectors of the covariance matrices, and
z c , k = ( V x c V F C E c V F F A c ) T ( x c , k F C o r t - E l a s c , k F F A c , k ) ∈ ℜ m c × 1 z e = ( V x e V F E l a s e ) T ( x e F E l a s e ) ∈ ℜ m e × 1 (20)
Differentiating the latent variable state vector zc,k and substituting Eqs 2, 15, 16 and 19 yields:
d z c , k d t = V x c T d x c , k d t + V F C E c T d F C o r t - E l a s c , k d t + V F F A c T d F F A c , k d t = A z c , k + B u k + C z e , k = 1 , ⋯ , n c e l l (21)
where:
A = V x c T ( W C E c V F C E c + W F A c V F F A c ) + V F C E c T ( Q x c V x c + Q F C E c V F C E c ) + V F F A c T ( H x c V x c + H F F A c V F F A c ) B = V x c T L c + V F C E c T Q u + V F F A c T H u C = V F F A c T H x e V x e (22)
Eq 21 provides the latent variable state equation of the k-th cell interacting with the ECM. Given the latent variable state of ECM ze and the input uk reflecting the cell’s decision, the transition of the cell’s latent variable state is determined locally without directly including the states of the other cells. Cells interact indirectly through the strain field created by other cells over the ECM fiber network.
The ECM dynamics can be represented in the latent variable space spanned by Ve. Differentiating the latent variable state vector and substituting Eqs 4, 14 and 19 yield:
d z e d t = V x e T d x e d t + V F E l a s e T d F E l a s e d t = G z e + ∑ k = 1 n c e l l D k z c , k (23)
where:
G = V x e T W E l a s e V F E l a s e + V F E l a s e T ( R x e V x e + R F E l a s e V F E l a s e ) D k = V x e T W F A e P m a p k V F F A c (24)
Fig 2 shows numerical examples of the DF linearization and subsequent latent variable transformation in reproducing accurate cell morphologies of the original nonlinear computational model over time. Remarkably, the DF linearized latent variable model can correctly reconstruct the complex cell membrane topology with m = mc + me = 50 + 50 = 100 total latent variables. Fig 2C quantifies the root mean square error and computation time as a function of latent variable model dimension. As can be seen, the computation time for the latent variable cell-ECM system increases with increased latent variables while the root mean square error decreases. Conversely, the standard Taylor expansion linearization is not capable of representing cell morphologies without marked error which is quantified in Fig 2B. We also compare our DF linearization approach to a more sophisticated method for approximation of nonlinear systems termed trajectory piece-wise linear (TPWL) [30, 31]. This method uses collection of (algebraic) linearizations of the original nonlinear system about suitably selected states to approximate the nonlinear system. As can be seen from Fig 2B, although the TPWL method yields lower error than the first order taylor expansion, our DF linearized model still leads to significantly lower prediction error. This is because in DF linearization, instead of taking “algebraic” linearization of nonlinear terms, we consider “dynamic” linearization by representing their dynamic transitions using linear regressions. These results demonstrate the effectiveness of DF linearization and model reduction in reconstructing simulations from a high dimensional complex nonlinear model.
This latent space model provides not only a low-dimensional structure for efficient computation, but also contains natural insights into the interactions among the multiple cells. Fig 3A shows the dynamic interactions in block diagram form based on Eqs 21 and 23. The ECM changes its latent variable state ze with the autoregressive feedback through matrix G as well as with the forward path that collects the latent variable states of all the individual cells, as shown by the summing junction Σ. Each single cell changes its latent variable state zc,k (k = 1, …, ncell) with autoregressive feedback through A as well as with two forward paths. The first path (fed through C) represents global feedback from the ECM (ze). The second path (fed through B) represents the updated lamellipodial forces uk determined from ECM state ze. The lamellipodial forces can be thought of as the individual cell’s decision based on its position and updated ECM properties as explained more in the following section. The actions taken by all the cells are integrated into the global ECM state transition, which is fed back to the individual cells. Therefore, each cell is connected to other cells through the global feedback of the ECM latent variable state ze. Fig 3A manifests the control-theoretic interpretation of multiple cells interacting through ECM.
Since the system is represented in a lower dimensional space, the high dimensional regression coefficient matrices (R ⋆ ⋆ , Q ⋆ ⋆ , H ⋆ ⋆) are not computed explicitly. Instead, the lower dimension coefficient matrices A, B, C, G are computed from numerical simulation data that can be transformed into latent variable space. Details are given in the Methods Section.
As discussed previously, input vector uk pertains to the lamellipodial forces generated at each membrane node within the leading edge of the cell. The cell continuously updates its lamellipodial protrusions depending on the orientation of the leading edge as the cell’s polarization (or polarity) changes. The polarity of a cell is important to determine the orientation of the leading edge and is influenced by the direction of local maximum stiffness in the ECM [23–26]. Here we aim to extend the dynamics model of cell polarity developed in [23–25] to predict the formation of lamellipodia.
Let d P o l k ∈ ℜ 3 × 1 be a 3-dimensional unit vector indicating the direction of polarity in the k-th cell and d M a x - S t i f f e , k ∈ ℜ 3 × 1 be a 3-dimensional unit vector pointing in the direction of the maximum stiffness of ECM in the vicinity of the k-th cell’s current location. According to [25], the cell polarity rotates dynamically in response to ECM’s local stiffness in such a way that the polarity vector may align with the direction of the maximum stiffness:
d d P o l k d t = κ d P o l k × ( d M a x - S t i f f e , k × d P o l k ) (25)
where × indicates vector product, and κ is a scalar parameter. Fig 3B illustrates this relationship. The polarity vector d P o l k tends to align with the maximum stiffness direction, d M a x - S t i f f e , k.
The leading edge of the cell is indicated by a right circular cone with apex angle 2 α L k having its centerline aligned with the polarity direction. The membrane nodes of the k-th cell within the cone have nonzero lamellipodial forces (F L , i c , k ≠ 0). Membrane nodes outside this cone have zero lamellipodial forces (F L , i c , k = 0). The direction of maximum ECM stiffness d M a x - S t i f f e , k depends on the stress field within the ECM, which pertains to the latent state vector of ECM ze. The details are given in Appendix C in S1 Text.
Compaction of the ECM by the collective efforts of multiple cells is numerically analyzed based on the model reduction and superposition of the nonlinear cell-ECM dynamics via DF Linearization. We first consider the case where two cells placed 30 μm apart are embedded in a 3-D cylindrical ECM that measures 40 μm in diameter and 100 μm in length as seen in Fig 4A. The boundary conditions of the ECM fiber network are set such that the two flat planes on sides are fixed to space (constrained), while the curved surface surrounding the ECM is kept free (unconstrained). The volume of the cylindrical ECM shrinks over time from its initial unstressed state as the cells interact with the surrounding ECM. To quantify the spatiotemporal compaction process, the original ECM cylinder is segmented into 10 slices of 10 μm thickness along its longitudinal axis as shown in Fig 4B. The volumetric changes to the individual slices are plotted in Fig 4C. The prediction of decreased cell volume by the latent variable superposition model (blue) agrees well with the ground-truth, full-scale nonlinear simulation results (green). This is further verified by the corresponding cross-sectional images of the 2-cell cylindrical ECM simulations in Fig 4C. The polarity directions of both cells (shown by red arrows initially pointing in arbitrary directions) shift to point inward, indicating that larger stresses are detected in the area between the cells. A video of the simulation comparison is shown in S1 Video.
The proposed model is able to reproduce collective behaviors of multiple cells causing the characteristic compaction of ECM gel, which is not observed for single isolated cell models. This is further verified in Fig 4D which compares the ECM compaction results between single cell and two cell models. As can be seen, the single cell model predicts more localized shrinkage of the ECM volume whereas the two cell model shows more global shrinkage extended to within the region between the cells. A video of the simulation comparison is shown in S2 Video.
Fig 4D suggests the presence of more than one cell is necessary for the pronounced ECM compaction leading to emergent changes within the ECM. However, the emergence of pronounced compaction entails not only plurality of cells but proper cell spacing. Fig 5A shows that, as the spacing between cells increases, compaction is less pronounced between them, indicating decreased interaction and integration of cell induced propagated forces. This is summarized in Fig 5B which quantifies the average ECM elastic force in-between cells against cell spacing. From Fig 5B, we see that the average ECM elastic force in-between cells spaced at 100 μm is an order of magnitude less than that of the cells spaced at 30 μm. A video of this simulation is shown in S3 Video.
The above computational results shown in Figs 4 and 5 verify that the proposed method can capture collective behaviors of multiple cells. The verification was made by comparing the reduced-order superposition model using DF Linearization against the full-scale, nonlinear model. To further verify the capability of the reduced-order superposition model, a comparison is also made against in vitro experimental data of ECM compaction by a larger number of cells. As shown in Fig 5C and 5D, the computational model successfully reproduces the in vitro experiment conducted by Fernandez, et al [5], in which a heterogeneous planar distribution of MC3T3-E1 osteoblasts were plated in 3-D rectangular prism collagen gel of 50 μm height, 100μm width and 250 μm length. The boundary conditions of ECM in the computational model were set to be consistent with experimental conditions in [5].
The multi-cell latent variable simulation is able to predict the characteristics of the ECM over time. Whereas the group of 5 cells at the left edge exhibit anisotropic contraction of the ECM at the boundary, the single isolated cell at the right edge barely contracts the gel. Fig 5C compares the isolated cell to the group of cells in terms of maximum edge displacement. The isolated cell’s ECM edge displacement is so small that five times its displacement is still substantially lower than the displacement of the group of cells. A video of this simulation is shown in the S4 Video. The presented method for predicting collective behaviors of cell-mediated ECM gel compaction is scalable. Since the individual cell-ECM interactions are local computations, as given by Eq 21, the computational complexity does not increase exponentially, although the number of cells increases.
Implementation of the presented method is enabled by four key constructs: 1) generation of simulated training data, 2) formation of the data covariance matrix necessary for latent variable transformations, 3) estimation of the parameter matrices involved in the latent variable space state equations (Eqs 21 and 23), and 4) association of cell and ECM focal adhesion force variables using mapping matrix. Each method is detailed below.
The nonlinear state Eqs 2 and 4–6 were computed with custom C-code based on references [23, 24]. The computation of a single isolated cell embedded in the cylindrical ECM took approximately 24 hours for a single simulation of physical time T ≈ 3600 seconds with sampling interval of 1 second. The simulation was repeated for N ≈ 10 times at different initial cell locations each time. The simulations with a single isolated cell embedded in the large rectangular ECM for reproducing the experimental result were run for approximately 5 days (120 hours). The physical time of simulation was T ≈ 3600. The simulation was repeated for N ≈ 10 for various initial locations of a cell. For each simulation, the total number of sample points for all the variables was over 5,000,000. The number of sample points was 5 × 107. The computation was performed on Intel Xeon CPU E5-2687W @ 3.10 GHz (2 processors) with 32 logical cores. More details on the formulation of the full-scale nonlinear state equations are summarized in Appendix A in S1 Text.
We create the covariance matrix using simulated training data. The training data consists of 3,600 time points of both state and auxiliary variables of a cell embedded in an ECM environment. The simulation is repeated N = 10 times, each time with the cell embedded in distinct locations within the ECM. Then the data covariance matrices may be formed:
C ζ ζ c = 1 K · N · T ∑ k = 1 K ∑ n = 1 N ∑ t = 1 T ζ ˜ c , k , n ( t ) ζ ˜ c , k , n ( t ) T C ζ ζ e = 1 N · T ∑ n = 1 N ∑ t = 1 T ζ ˜ e , n ( t ) ζ ˜ e , n ( t ) T (26)
where ζ ˜ c , n , k ( t ) represents the mean centered t-th time sample (of augmented variable vector ζc, k) for the k-th cell (here K = 1 or 2) embedded within for the ECM at the n-th simulation, and ζ ˜ e , n ( t ) represents the mean centered t-th time sample of the augmented variable vector ζe in the n-th simulation. By performing eigen-decomposition on the covariance matrices we obtain the orthogonal matrix Vc, Ve comprised the eigenvectors of the data covariance matrix:
C ζ ζ c ≈ V c Λ c V c T ∈ ℜ m c × m c C ζ ζ e ≈ V e Λ e V e T ∈ ℜ m e × m e (27)
where Λc, Λe are diagonal matrices containing the largest mc and me eigenvalues of the covariance matrices, respectively. It is important to check whether the covariance matrices contain sufficiently rich data, and their first mc and me components are sufficient to capture the cell-ECM dynamics at any cell location within the ECM. Standard techniques can be applied to validate the data and truncation of components [29]. With these, the ECM dynamics of a cell embedded within the ECM at an arbitrary location will be well represented in linear latent variable space which is critical to the success of the method. Covariance matrix calculation were conducted by using Matlab.
The following outlines the steps to compute coefficient matrices A, B, C, G by Eqs 21 and 23:
Parameter matrices A ∈ ℜ m c × m c , B ∈ ℜ m c × 3 N c , C ∈ ℜ m c × m e , G ∈ ℜ m e × m e are substantially lower in dimension than the regression coefficient matrices R * * , Q * * , H * * given in Eqs 14, 15 and 16. Therefore, fewer data points allow us to determine these parameter matrices in the latent variable space. It should be noted that matrices Dk’s are of high dimension, but are not computed with regression since they consist of known matrices as shown in Eq 24. Matlab was used for estimation of parameter matrices and subsequent computations of the latent variable model. 3-D visualization of simulation data was conducted using Tecplot 360.
When a focal adhesion is formed between the i-th node of the k-th cell and the j-th node of ECM, the two focal adhesion forces sum to zero, as described previously (F F A , i c , k + F F A , j e = 0 where F F A , i c , k , F F A , j e ∈ ℜ 3 × 1). Representing this relationship in terms of the collective focal adhesion force vectors, F F A c , k ∈ ℜ 3 N c × 1 and F F A e ∈ ℜ 3 N e × 1, requires a matrix P m a p k ∈ ℜ 3 N e × 3 N c. Let F F A e , k be the forces acting on the ECM nodes caused by focal adhesion between the k-th cell and the ECM nodes. This can be written as
i ↓ F F A e , k = ( ⋮ F F A , j e , k ⋮ ⋮ ) = j → ( ⋯ ⋯ ⋯ ⋯ ⋯ ⋯ - I 3 × 3 ⋯ ⋯ ⋯ ⋯ ⋯ ⋯ ⋯ ⋯ ⋯ ) ︸ P m a p k ( ⋮ ⋮ F F A , i c , k ⋮ ) (28)
where I3×3 is the 3-dimensional identity matrix. Obtaining this mapping matrix P m a p k for all the cells, the complete focal adhesion forces in the ECM can be expressed in relation to the cell’s focal adhesion forces.
F F A e = ∑ k = 1 K F F A e , k = ∑ k = 1 K P m a p k F F A c , k (29)
As previously mentioned, the focal adhesion connections between the cell membrane nodes and ECM nodes can vary over time as the cell membrane deforms, gains traction, and generates lamellipodial protrusions. Therefore, P m a p k is updated to reflect the new focal adhesion attachments and detachments at each time step. The original nonlinear computational model has developed a functional relationship between the focal adhesion force, number of integrins, and distance between the membrane and ECM node (see details supplementary materials Appendix A in S1 Text). In the presented framework, the change in focal adhesion attachments can be derived from simulated training of the nonlinear computational model.
The collective ECM compaction by multiple cells is predicted through superposition of individual cells’ contributions in latent variable space. This is made possible by DF Linearization, latent variable transformation and subsequent superposition of single-cell models to predict the collective behavior among multiple cells.
As shown in Fig 3A, the DF linearization has two-order-of-magnitude higher accuracy than the first-order Taylor expansion, and can approximate the original full scale model with a reasonable root-mean-square error. This representation of nonlinear dynamics is markedly different from standard linearization methods. The DF linearization was also compared to the TPWL method. The figure shows an order-of-magnitude better result for the DF linearization compared to the sophisticated technique. Note that the TPWL does not yield a linear model since the state equation includes a product of two state functions. Therefore, superposition as applied with our DF linearization approach cannot be applied.
As applied to the analysis of multi-cell ECM compaction, linear augmented equations describing single cell-ECM interactions were derived from DF linearization, and then converted to a reduced-order linear representation by transformation onto a basis of eigenvectors derived from simulated data set. Unlike model reduction of nonlinear dynamical systems, which still remains a challenging problem in the field [19–22], the model reduction of a linear system through DF Linearization is straightforward. It allows for the evolution of independent and auxiliary states to be described within a lower dimensional linear manifold. The resulting reduced order latent variable model is capable of reproducing nonlinear dynamics, and the linearized structure of individual models facilitated their integration to describe multi-cell behaviors. The prediction of collective behaviors of a group of cells was achieved by superposing contributions of individual cells represented by latent variables zc,k, which evolves based on their own dynamics in response to the global ECM state represented by latent variable ze.
The linear representation of the collective multi-cell-ECM interactions manifests the two types of feedback actions by the individual cells. As shown in the block diagram in Fig 3A, the individual cells are exposed to the ECM forces represented by latent variable vector ze in two separate paths. The path through the cell polarity block and matrix B, leading to lamellipodia formation, can be viewed as an “active input” as addressed in [5]. This feedback path includes a cell’s internal decision as to which direction it extends lamellipodia. In contrast, the other feedback path through a gain matrix C does not have a high-level cell decision, but is reactive, playing a “passive role” [5]. These feedback interactions support the prior experimental work [5]. It is interesting to note that ECM compaction begins almost instantaneously, but the magnitude of compaction is rather limited. Once the “active” feedback loop is initiated in, the ECM compacts further, resulting in a large deformation. As the polarity dynamics are rather slow, the second stage ECM compaction does not start immediately. The time scale is determined by the constant κ involved in the polarity dynamics Eq 25. Using the proposed methodologies, we are able to reproduce intercellular mechanical interactions consistent with published experimental observations. In particular, the global compaction of gel volume via collective cell-contractile activities is characteristically different from local deformations of single isolated cells embedded within the same gel. Through study of emergent behaviors of groups of cells embedded in a 3-D ECM fiber network, we can advance our understanding of intercellular mechanical signaling during tissue formation [1–7]. There are a few limitations to our method, however. While the presented method can predict complex nonlinear behaviors, the method is still a type of approximation. Care must be taken with the validity period. In Fig 4C at the sample time of t = 50 minutes, the latent variable superposition simulation over predicts the volume shrinkage by 12%.
With the current mathematical formulation, we have not yet incorporated the degradation of ECM fibers through matrix metalloproteinases. ECM degradation would be necessary to reproduce sustained movement and migration of the cells particularly in 3-D embedded matrices [32]. Since ECM degradation continuously changes the fiber connectivity through ECM remodeling, a methodology to update the node grid structure describing the ECM field would need to be developed. However, ECM degradation may not be necessary for predicting gel compaction since a cluster of cells remains stationary when contracting the surrounding gel [5]. Finally, in the current work, it was assumed that the cell’s polarity mechanism is a dominating internal response to mechanical cues. Cells change their internal state through a complex process of mechanotransduction and intracellular signaling. Incorporating these more complex mechanisms is an exciting avenue for future research. While the method has been developed and demonstrated for multi-cellular interactions with 3D ECM, the basic methodology is applicable to a broad range of systems where nonlinear dynamics of many interacting subsystems are prohibitively complex to compute.
|
10.1371/journal.pbio.1002253 | Vibrissa Self-Motion and Touch Are Reliably Encoded along the Same Somatosensory Pathway from Brainstem through Thalamus | Active sensing involves the fusion of internally generated motor events with external sensation. For rodents, active somatosensation includes scanning the immediate environment with the mystacial vibrissae. In doing so, the vibrissae may touch an object at any angle in the whisk cycle. The representation of touch and vibrissa self-motion may in principle be encoded along separate pathways, or share a single pathway, from the periphery to cortex. Past studies established that the spike rates in neurons along the lemniscal pathway from receptors to cortex, which includes the principal trigeminal and ventral-posterior-medial thalamic nuclei, are substantially modulated by touch. In contrast, spike rates along the paralemniscal pathway, which includes the rostral spinal trigeminal interpolaris, posteromedial thalamic, and ventral zona incerta nuclei, are only weakly modulated by touch. Here we find that neurons along the lemniscal pathway robustly encode rhythmic whisking on a cycle-by-cycle basis, while encoding along the paralemniscal pathway is relatively poor. Thus, the representations of both touch and self-motion share one pathway. In fact, some individual neurons carry both signals, so that upstream neurons with a supralinear gain function could, in principle, demodulate these signals to recover the known decoding of touch as a function of vibrissa position in the whisk cycle.
| Animals interrogate the world around them with actively moving sensory organs, resulting in a blend of sensory inputs: one input is from the object under study, while the second is from self-generated movement of the sensor. The detection of an object thus depends on the ability of the animal to distinguish among internally versus externally generated sensations. Nervous systems employ various signaling mechanisms to reference inputs from a sensory organ relative to its position. A well-known example is proprioception, in which receptors in the limb muscles and joints are used to infer the position of tactile sensors in the hands. In this case, the signals of limb position are encoded in areas of the brain that are distinct from those encoding touch. Here, we investigate the analogous problem of encoding the position of the vibrissae, or whiskers—essential orofacial sensorimotor organs in rodents. In contrast to the case for limbs, we find that vibrissa position is encoded along the same neuroanatomical pathway as vibrissa touch. The seeming ambiguity that results from the mixed representation of position and touch can be resolved by a nonlinear neuronal input-output relation that demodulates touch with respect to vibrissa position. This scheme enables the rodent to determine where an object is located relative to its body axis.
| Animals navigate the world around them with actively moving sensory organs [1]. This process results in a blend of sensory input from the presence of two underlying sensory signals. One input is from the environment or object under study, while the second is from self-generated movement of the sensor [2]. The detection of an external stimulus with confidence, as well as the ability to confirm the position and trajectory of the sensor, depends on the ability of the animal to distinguish among internally versus externally generated sensations. Ambiguity among these sources leads to unpleasant outcomes, such as vertigo [3] and motion sickness [4] for the case of vestibular control.
To resolve this ambiguity, nervous systems use three complementary signaling mechanisms to reference input from a sensory organ relative to the position of the sensors [5]. One is to encode self-generated sensor movement by the exo-receptors that also encode changes in the external environment; this is denoted peripheral re-afference. A second mechanism is to use muscular endo-receptors to encode elongation and contraction force, as performed by spindle fibers and Golgi tendons, respectively; this is denoted proprioception. A third mechanism is to generate a central copy of the motor commands for the intended sensor position; this is denoted corollary discharge. These three mechanisms report complementary, but not necessarily complete [6], information on sensor position.
While movement of a limb involves proprioceptive and corollary discharge reference signals, current evidence suggests that facial muscles, which bridge attachment points across soft tissue as opposed to bone, contain neither spindle fibers nor Golgi tendons [7–11]. Additional evidence demonstrates that despite the presumed lack of proprioceptors in the vibrissa musculature, neuronal signals related to rhythmic self-generated vibrissa motion, i.e. whisking, are encoded predominantly through peripheral sensory mechanisms [12–14]. Together, these observations lead to the hypothesis that self-generated vibrissa motion is encoded through re-afferent activation of mechanoreceptors. Specifically, activation of lanceolate- and/or Merkel-ending trigeminal neurons could presumably encode both re-afferent and ex-afferent input. These primary sensory neurons have identical, broad axonal arborizations across nuclei in the trigeminal brainstem [15,16]. Vibrissa self-motion signals are thought to inform the rodent about the position of its vibrissae upon tactile contact with an object [17–20], though an alternative possibility based on contact forces has been proposed [21] and critiqued [22].
How might the animal determine the location of objects that it contacts with its moving vibrissae? Past work shows that the strength of vibrissal ex-afferent touch responses, as measured in cortex, are strongly modulated by the phase in the whisk cycle at the moment of contact [20]. The responses of these units, therefore, contain the information necessary to determine object location through self-motion, but the underlying neuronal architecture required to achieve this cortical representation of object location remains unknown. Elements of signal detection theory [23] suggest two scenarios to demodulate touch relative to phase in the whisk cycle. One scenario is that the whisking and touch signals are encoded by different populations of peripheral receptors and are maintained as separate whisking and touch pathways to somatosensory cortex. A plausible scheme for demodulation involves gating of the touch signal by the separate whisking signal [20]. A second scenario is that both whisking and touch signals are encoded by the same sensory receptors and central neurons to cortex. In this case, a gain function with an accelerating nonlinearity [24] can serve to demodulate the touch signal.
As a means to gain insight into the particular scenario used by rodents to merge touch and self-motion of the vibrissae, we examine the response of neurons along the two dominant ascending somatosensory pathways [25,26]. Our investigation is motivated by the pioneering work of Ahissar and colleagues [27], who addressed the issue of pathways at the level of thalamus. These investigators made use of anesthetized animals, in which whisking was induced by electrical stimulation of the buccal motor branch of the facial nerve [28]. Under these conditions, the neuronal spikes rates are much reduced by the effects of anesthesia and the concurrent loss of neuromodulation. Furthermore, the process of electrical stimulation leads to the preferential activation of motoneurons with large caliper axons, as opposed to physiological recruitment, which begins with fibers of small caliper and progresses to those of larger caliper [29]. Thus there is a need for a thorough reexamination of the signaling of vibrissa input along ascending somatosensory pathways.
The more familiar of the two pathways, the lemniscal somatosensory pathway, includes trigeminal nucleus principalis (PrV) and the upstream dorso-medial division of ventral-posterior-medial (VPMdm) thalamic nucleus (Fig 1). Neurons along this pathway spike vigorously in response to stimulus-induced deflection of one or multiple vibrissa [30–33]. Yet there is limited information on the nature of the response to vibrissa self-motion [34]. A second pathway, the paralemniscal pathway, encompasses the rostral aspect of spinal trigeminal nucleus interpolaris (SpVIr), the upstream posterior medial (PO) thalamic nucleus, and includes collaterals to the ventral aspect of zona incerta (ZIv), a region that further provides feedforward inhibition to PO thalamus (Fig 1) [35,36]. Neurons along this pathway in PO thalamus spike, albeit less prominently, in response to deflection of the vibrissae [31,37], yet there is apparently contradictory data on the nature of the self-motion response [27,38]. Lastly, we consider an alternate origin for whisking-related re-afference and ask if whisking is encoded by mechanoreceptors in the mystacial pad, which moves in phase with the vibrissae during whisking [39]. Encoding of self-motion in these receptors would represent re-afferent signals that are, in principle, independent of vibrissa touch. The result of these measurements defines the utilization of different pathways for sensorimotor signaling and constrains computational models of vibrissa-based object location [19,40].
Although current evidence suggests a lack of proprioceptive innervation of most facial muscles in a number of species, data specific to the innervation of the rodent vibrissa musculature are more limited [8]. We therefore used three complementary anatomical techniques to determine whether vibrissa muscles contain endo-receptors (Fig 2).
First, a classic measure to observe endo-receptors is via the labeling of spindle-like proprioceptive afferent endings [42]. Spindles appear as helical-shaped fine processes that surround intrafusal muscle fibers. Spindles are well known to be prominent in the masseter muscle [43], as confirmed by immunostaining of neurofilament proteins from tangential sections of the muscle (Fig 2a). We thus searched for spindle-like endings in the mystacial pad, in both intrinsic and extrinsic muscles, as compared to sections of masseter muscle from the same animals. The number of motoneuron endplate claws in the same sections serves to normalize our counts. We observed spindles in the vibrissa musculature in only one of three animals (2,480 endplates across 23 sections) (Fig 2c), which correspond to 0.0012 ± 0.0007 (mean ± SD) spindles/plate compared to 0.0279 ± 0.0054 for the masseter muscle (970 endplates across 36 sections) (Fig 2b). Thus the vibrissa muscles contain over 20-fold fewer spindles than a muscle with known proprioceptive control (Fig 2d).
As a second measure, we prepared transverse sections of both the mystacial pad and the masseter muscles and directly stained both the intrafusal and extrafusal fibers. The intrafusal fibers are identified by their small size and bundling of multiple fibers within a capsule (arrows in Fig 2e and 2f). Here the total number of extrafusal fibers in a section serves as the normalization. We observed intrafusal fibers in the vibrissa musculature in two of three animals (53,800 extrafusal fibers across 13 sections) (Fig 2f), which corresponds 0.00011 ± 0.00005 intrafusal to extrafusal fibers compared to 0.00160 ± 0.00017 for the masseter muscle (56,960 extrafusal fibers across 13 sections) (Fig 2e). Thus the vibrissa muscles contain 15-fold fewer intrafusal fibers than a muscle with known proprioceptive control (Fig 2d).
As a final measure, we asked if γ-motoneurons, which innervate intrafusal fibers, are present in the lateral facial nucleus. This nucleus contains the motoneurons for the vibrissa musculature [44,45]. As a positive control, we compared staining in the lateral facial nucleus to the trigeminal motor nucleus, which innervates the masseter and other jaw muscles and, consistent with the presence of spindles in the masseter muscle (Fig 2a), is known to contain γ-motoneuron efferents [46,47]. Recently, it has been demonstrated that γ-motoneurons can be distinguished from α-motoneurons based on their size and the relative intensity of anti-ChAT and anti-NeuN staining. Specifically, both α- and γ-motoneurons are labeled intensely with anti-ChAT, but α-motoneurons have larger somata and are labeled by anti-NeuN, whereas γ-motoneurons are smaller and are not labeled by anti-NeuN [48].
We analyzed immunohistochemical labeling on rat brainstem sections for ChAT and NeuN (Fig 3) and considered only neurons whose nucleus was contained in the section as indicated by a DAPI counterstain (Fig 3a and 3b). Qualitatively, the trigeminal motor nucleus contained two populations of motoneurons. Larger motoneurons were labeled both by anti-ChAT and anti-NeuN, whereas smaller motoneurons were labeled only by anti-ChAT (Fig 3c–3e). In the facial nucleus, we observed only one population of medium-sized motoneurons, presumably α-motoneurons, that were labeled both by anti-ChAT and anti-NeuN (Fig 3f–3h). To quantify these observations, we calculated the area of each motoneuron and the average intensity of anti-ChAT and anti-NeuN labeling within the labeled area. We observed two clusters of neurons in the trigeminal motor nucleus, putatively corresponding to α- and γ-motoneurons (Fig 3i and 3k). Approximately one-third of the motoneurons fell into the putative γ-motoneuron cluster, consistent with spinal motoneuron pools that innervate muscles with spindles [48]. In the facial motor nucleus we observed a unimodal distribution of motoneuron sizes and anti-NeuN intensities, putatively corresponding to α-motoneurons (Fig 3j and 3l). These results imply that the innervation of intrafusal fibers by the lateral facial motor nucleus represents at most a small fraction of its total output, and are consistent with past reports that most facial muscles lack proprioceptive signaling [7–11]. Together, these anatomical analyses of neuronal endings, muscle fibers, and motoneuron types imply that classic propriception makes a negligible contribution to the encoding of vibrissa self-motion.
Given the relatively poor proprioceptive innervation of the vibrissa musculature, re-afferent activation of trigeminal mechanosensory afferents, including lanceolate and Merkel ending neuron types, is a likely source of the sensory signal of whisking phase. We thus monitored neuronal activity in two of the target nuclei in the brainstem for these neuron types [15,16], nuclei PrV and SpVIr, that provide the majority of the ascending projections to thalamus (Fig 1). These nuclei anchor the lemniscal and paralemniscal pathways, respectively, and the literature is unequivocal about the presence of vibrissa touch responses in both nuclei. We recorded single and multi-unit activity in nucleus PrV (25 putative single unit and 31 multi-unit spiking signals) and nucleus SpVIr (14 putative single unit and 10 multi-unit spiking signals) (Methods). As illustrated by the example of Fig 4, the spike rates of units in nucleus PrV are substantially modulated on a cycle-by-cycle basis during rhythmic whisking in air (Fig 4a and 4b). To quantify this modulation of the spike rate, we isolated individual whisk cycles (Eq 1 and Eq 2) and aligned spike events relative to the instantaneous phase in the whisk cycle (Fig 4c,d) [49]. We next computed the distributions of whisking phases and of whisking phases at which spikes occurred (Fig 4e). From these distributions, we estimated the spike rate as a function of phase in the whisk cycle, (black line in Fig 4f) and fit a sinusoid rate function (Eq 4) to the data as a means to parameterize the modulation depth (Eq 5) and preferred phase. For the unit in Fig 4, the majority of spikes, i.e., 373/404 (92%) of spikes across 303 whisks, occurred during the retraction phase of the whisk cycle, when the vibrissae were moving in the caudal direction (Fig 4f).
Tactile receptive fields were established for a subset of the recorded units by briefly anesthetizing the rat with isoflurane and manually stimulating different vibrissae (Methods). The unit in the example of Fig 4 was located among many units that responded to vibrissa E3. The firing rates of additional example units as a function of phase in the whisk cycle, along with their local receptive fields, are shown in Fig 5. These include units in the sub-region of nucleus PrV that corresponds to the macro-vibrissae (Fig 5a) and units in sub-regions that correspond to the skin and fur around the mouth and nose (Fig 5b). Furthermore, we observed units in nucleus SpVIr that were significantly, albeit modestly, modulated by whisking (Fig 5c). As a population, 49/56 PrV units (88%) and 16/24 SpVIr units (67%) were significantly modulated by whisking (Kuiper test, p < 0.05). Units in nucleus PrV tended to fire more spikes when the animal was whisking as opposed to not whisking (Wilcoxon signed rank test, p = 1.0 x 10−5), whereas spike rates were not significantly different between whisking and not whisking in nucleus SpVIr (Wilcoxon signed rank test, p = 0.39; Fig 5d).
We characterized the sinusoidal fits of spike rates across all units (Figs 4f and 5a–5c) by two measures. The first measure is the modulation depth, MWhisk (Eq 5), which reports the fraction of the unit’s response that is locked to whisking. The second measure is the signal-to-noise ratio, SNRWhisk, over a time interval (T) chosen to be the average period between whisks for head-fixed rats, i.e., T = 165 ms (Eq 6) [39,50]. We observe a greater modulation depth for lower mean spike rates, with a SNRWhisk that peaks at a mean rate of <λ> ~20 Hz (Fig 6a). As a population, units in brainstem nucleus PrV were more strongly modulated than those in nucleus SpVIr (Wilcoxon ranked sum test, p = 0.03), with median MWhisk values of 1.0 versus 0.6 for nucleus PrV versus SpVIr, respectively. Furthermore, units in nucleus PrV had a greater SNRWhisk than those in nucleus SpVIr (Wilcoxon ranked sum test, p = 0.0016, with median values of SNRWhisk = 1.6 versus 0.8 for nucleus PrV versus SpVIr, respectively.
Different units preferentially spiked at different phases of the whisk cycle, denoted ϕPreferred (Eq 4). All phases are represented for units in both nuclei PrV and SpVIr (Fig 6b). There is a significant bias in the preferred phase across all units in nucleus PrV, with a vector average <SNRWhisk> = 0.6 and <ϕPreferred> = 4.9 radians (Hotelling’s one-sample test; p = 0.02); this phase corresponds to retraction from the fully retracted position. There was no bias for units in SpVIr (Hotelling’s one-sample test; p = 0.3) [51]. In toto, these data show that self-motion is represented along the primary nuclei of the lemniscal and paralemniscal pathways, but more robustly along the lemniscal pathway.
To determine the encoding of self-generated whisking in the thalamic nuclei that receive inputs from PrV and SpVIr (Fig 1), we recorded spiking activity of individual neurons using the juxtacellular configuration in VPM and PO thalamus (74 neurons). Occasionally, extracellular recordings of nearby units were obtained on the same micropipette; these units had negative initial deflections, as opposed to the initial positive spike deflections of the juxtacellularly recorded neurons (3 of 71 records). Similarly, we recorded spiking activity of individual neurons using the extracellular or juxtacellular configuration in ZIv (15 neurons). We next consider the spiking dynamics of individual neurons in VPM and PO thalamus, as well as in ZIv, in response to self-generated whisks (Figs 7–9) and external vibrissa deflections with air-puffs (Fig 10).
As illustrated by the example of Fig 7, in which the neuron was located among units that responded to vibrissa C4, neurons in VPM thalamus are substantially modulated on a cycle-by-cycle basis during whisking (Fig 7a and 7b). The analysis of the spike rate as a function of phase in the whisk cycle for thalamic neurons (Fig 7c–7f) proceeded similarly to that for units in the brainstem (Figs 4 and 5). For the neuron in Fig 7, the majority of spikes occurred during the protraction phase of the whisk cycle. The spike rate from this neuron was particularly well described by a sinusoidal modulation as a function of phase (Fig 7f).
Additional example neurons from VPM thalamus, the adjacent sub-region in PO thalamus, and ZIv, along with their anatomical locations of the recording sites, are shown in Fig 7, S1, S2 and S3 Figs. Qualitatively, neurons in the sub-region of VPM thalamus that corresponds to the macro-vibrissae (Fig 8a), as well as units in sub-regions that correspond to the skin or fur around the mouth and nose (Fig 8b), were modulated. PO thalamus also contained a minority of neurons that were modulated (Fig 7c), while modulation appeared absent in neurons in ZIv (Fig 8d). As a population, neurons in VPM and PO thalamus tended to fire more spikes when the animal was whisking as opposed to not whisking (Wilcoxon signed rank test, p = 10−9 and p = 0.04, respectively (Fig 8e). This is consistent with past results [52]. Neurons in ZIv tended to fire fewer spikes when the animal was whisking (Wilcoxon signed rank test, p = 0.02) (Fig 8e). Overall, neurons in VPM thalamus tended to have significantly higher spike rates than in those PO thalamus during whisking epochs (Wilcoxon ranked sum text, p = 0.0057), but not during non-whisking epochs (Wilcoxon ranked sum text, p = 0.15).
Similarly to the analysis for units in nuclei PrV and SpVIr, we characterized the population response for neurons in VPM and PO thalamus and ZIv in terms of the modulation depth, MWhisk (Eq 5), and the signal-to-noise ratio, SNRWhisk with T = 165 ms (Eq 6). The majority of neurons in VPM thalamus were significantly modulated by whisking phase (49/57; Kuiper test p < 0.05) (Fig 9a), whereas only a minority of PO neurons were significantly modulated (4/17; Kuiper test p < 0.05) (Fig 9a) and no neurons in ZIv were significantly modulated (0/15; Kuiper test p < 0.05). Of the VPM neurons located among units that had receptive fields corresponding to the micro-vibrissae or peri-mystacial fur, 12/15 of these neurons were significantly modulated. As in the case for brainstem (Fig 6b), different VPM neurons preferentially spiked at different phases of the whisk cycle. All phases are represented for neurons in VPM thalamus (Fig 9b), but with no significant bias in the preferred phase (Hotelling’s one-sample test; p = 0.72). In toto, these data show that self-motion is represented in thalamic nuclei of the lemniscal and paralemniscal pathways but, as with the case of brainstem, only robustly along the lemniscal pathway.
We computed the perpendicular distance between the Chicago sky blue spot and the VPM/PO border for each labeled recording site, based on cytochrome-oxidase stained sections. The location of the VPM/PO border, determined by visual inspection, was estimated to be accurate to approximately 80 μm (S4 Fig and S4 Data). There does not appear to be a clear systematic relationship between the signal-to-noise ratio for whisking and the distance to the border between VPM and PO thalamus at this spatial resolution. Neurons with high values of SNRWhisk occur in VPM thalamus both close to the border as well as deeper in the nucleus (Fig 9c). To further clarify whether there is a potential segregation of function within VPM thalamus, we reconstructed the locations of the labelled recording sites in three dimensions (Fig 9d and 9e). Again, there is no clear spatial relationship between the location of a neuron within VPM and its SNRWhisk. The lack of topography would imply that self-generated motion and touch are signaled within the same anatomical pathway.
To determine whether the same neurons respond to ex-afferent and re-afferent stimuli, we next consider how the same neurons along the lemniscal and paralemniscal pathways respond to external deflections of the vibrissae. The case for touch-based responses in the VPM thalamus, along the lemniscal pathway, is unequivocal. However, the case for touch-based responses in PO thalamus, along the paralemniscal pathway, is the subject of conflicting reports as to whether external stimuli can drive neurons in PO thalamus independent of feedback activation from deep layers in cortex. As past work involved anesthetized animals [37,38,53–55], we undertook a re-analysis of the response of neurons in VPM and PO thalamus along with the somatosensory region of ZIv (Fig 1).
As illustrated by the examples of Fig 10a–10c, neurons in all three areas were modulated by air-puff deflections to multiple vibrissae and peri-mystacial fur, with neurons in VPM thalamus responding vigorously, those in PO thalamus the least responsive (Fig 10b), and those in ZIv responding with short latency, precisely timed spikes (Fig 10c). Across the population, 49/54 VPM neurons (91%), 11/17 PO neurons (65%), and 12/15 (80%) ZIv neurons were significantly modulated by air-puffs (p < 0.05) (Fig 10d). These data imply that nucleus SpVIr indeed drives ascending targets and that neurons in PO thalamus are responsive to stimulation in alert rats.
We next consider the responses of these same neurons to self-motion of the vibrissae (inserts in Fig 10a–10c). Consistent with the notion of a single anatomical pathway for re-afferent whisking and ex-afferent touch, the majority of VPM units that were modulated by self-generated whisking tended to also be modulated by air-puff deflections. Of the neurons in VPM thalamus, 42/54 (78%) were significantly modulated by both air-puffs and whisking, five were modulated by whisking only, and seven were modulated by air-puffs only. Yet there does not appear to be a relationship between the fidelity of modulation for VPM neurons that are significantly modulated by both whisking and air-puffs, as measured by the correlation between signal-to-noise ratio for whisking and the peak modulation upon air-puff (Fig 10e), (r = 0.05 with p = 0.76 for VPM units).
We report the representation of self-generated whisking in subcortical somatosensory brain regions (Fig 1). First, we assess the potential contribution of proprioceptive endings in the facial musculature to somatosensation. We find a small number of previously unreported spindle-like endings and intrafusal fibers within the vibrissa musculature. However, these endings fibers are relatively scarce in comparison with the nearby masseter muscle (Fig 2). Furthermore, using a recently developed immunohistochemical strategy [48], we find that the lateral facial motor nucleus contains few, if any, intrafusal fiber-innervating γ-motoneurons (Fig 3). While we cannot rule out the possibility that intrafusal fibers are instead innervated by the hypoglossal [56] and trigeminal mesencephalic nuclei [57], this finding, together with comparatively low density of spindles and intrafusal fibers, would suggest that vibrissa position is unlikely to be sensed by proprioception [58]. Nonetheless, we find a robust representation of self-generated vibrissa motion, i.e. whisking, in nucleus PrV (Fig 4), which receives primary afferent input from mechanoreceptors in the vibrissa follicles and the face. These results, together with previous findings that the representation of rhythmic vibrissa motion in somatosensory afferents are derived from peripheral sensors [14,59,60], leads us to conclude that vibrissa position during whisking is encoded through re-afferent activation of mechanical exo-receptors (Figs 4–6).
We next established the modulation of spiking activity of neurons in the lemniscal and paralemniscal pathways by the phase in the whisk cycle. At the level of the trigeminal brainstem, lemniscal neurons in nucleus PrV are substantially more reliable encoders of phase than paralemniscal neurons in spinal trigeminal nucleus SpVIr (Figs 5 and 6). At the level of the thalamus, lemniscal neurons in VPM thalamus are again substantially more reliable encoders of phase than paralemniscal neurons in PO thalamus (Figs 7–9). In particular, the majority of PO thalamic neurons do not significantly encode phase (Figs 7–9). Consistent with the lack of whisking-related modulation in PO thalamus, neurons in ZIv, which receive inputs from axon collaterals of cells in SpVIr that primarily project to PO-thalamus (Fig 1), are also not modulated by whisking (Figs 8 and 9). Together these data indicate that the lemniscal pathway from brainstem to cortex contains both neurons with the highest acuity for passive vibrissa deflections and neurons with the greatest reliability for encoding phase in the whisk cycle. Some single units are reliable encoders of both signals (Fig 10a and 10e), as proposed by studies that utilized electrically induced whisking in anesthetized animals [27,61].
Whisking-phase responses observed in VPM thalamic neurons in the present study substantially extend the results of past studies performed with both alert [34] and anesthetized [27,61] rats. We observe phase-dependent spiking modulation throughout the depth of VPMdm thalamus, which presumably comprises units in both the “head” and “core” regions of the barreloids [62]. This finding is consistent with results in which artificial whisking was induced by electrical stimulation of the facial nerve in anesthetized rats [27]; however, we find that units are tuned to all phases of the whisk cycle rather than to protraction onset. These broader distributions of preferred phases, which are observed in both PrV and VPM thalamus (Figs 6b and 9b), are consistent with the range of phase preferences observed somatosensory cortex during natural whisking [12,13,20,63,64]. We were unable to assess whether there is a finer systematic map of the encoding of self-motion on the scale of individual barreloids [62,65]. Interestingly, in addition to the barreloids, we observe modulation in phase with whisking in some units that encode distortions to the skin or fur outside of the vibrissa follicle in both PrV and VPM thalamus (Figs 5b, 6a, 8b, 8e and 9).
The observation that the majority of whisking responses are encoded within the lemniscal pathway raises the question of how phase-dependent touch signals, which were previously observed in somatosensory cortex [20], arise from the observed thalamic inputs. There are at least two potential schemes that could produce these cortical phase-dependent touch signals (Fig 11). One scheme is that whisking and touch are encoded by different populations of peripheral mechanoreceptors and central neurons. In this scheme, thalamic neurons that predominantly encode the whisking signal could change the slope of the gain function of cortical neurons, i.e., the proportionality of spike rate to input current [20], in a phase-dependent manner (Fig 11a), analogous to heterodyne detection [23]. Contrary to previously proposed hypotheses [20,27], our data indicate that paralemniscal inputs are unlikely to be the source of this cortical gain modulation. However, lemniscal units that encode skin or fur distortions during whisking, which we observed in nucleus PrV and VPM thalamus, could in principle contribute to a re-afferent signal of vibrissa position that is independent of vibrissa touch (Figs 5b and 8b). It remains to be determined whether such signals can influence phase-dependent touch responses in the barrels of somatosensory cortex.
A more parsimonious scheme is that the same mechanoreceptors, PrV neurons, and VPM thalamic neurons encode both whisking and touch signals. In this scheme, a gain function with an accelerating nonlinearity [24] could enhance the spike rate at the peak of the whisking signal relative to other positions (Fig 11b), in analogy to homodyne detection [23] and the effect of a threshold nonlinearity [66]. Based on the present results, units that encode both whisking and external vibrissa deflections could provide the relevant inputs to somatosensory cortex (Fig 10a and 10e). According to this scheme, if touch occurs at preferred phase of the whisk cycle, the response is enhanced, while touch at the non-preferred phase leads to a diminished response. Such non-linear gain functions could be present at multiple stages along the sensory processing stream, including at the mechanoreceptors themselves. In fact, modulation of touch by self-motion can occur even if self-motion signal alone is sub-threshold, and the resulting threshold nonlinearity can further enhance the difference between touches at different phases.
The potential role of the paralemniscal pathway in sensing vibrissa motion is controversial [27,38,53,67–70]. The majority of neurons in nucleus SpVIr are similarly tuned to upward vibrissa deflections of many vibrissae in anesthetized rats [71], but are only weakly tuned to phase during whisking relative to neurons in nucleus PrV (Figs 5 and 6). Neurons in PO thalamus respond only weakly to external vibrissa deflections as a consequence of feed-forward inhibition from the output of ZIv neurons in ketamine-anesthetized rats [36]. Electrical stimulation of vibrissa motor cortex inhibits activity in ZIv, which disinhibits neurons in PO thalamus and thereby increases its responsiveness to deflections [69]. This observation led to the hypothesis that whisking-related activity in primary motor cortex [49,72] might gate PO thalamus so that it is sensitive to whisking. Our data suggest that while the overall firing rates of ZIv neurons decrease slightly during whisking, this decrease is not sufficient to elicit whisking-phase dependent responses in PO thalamus. The lack of phase-dependent responses in the majority of PO units in our study is consistent with a past report [38] but inconsistent with results obtained with electrically induced whisking in urethane-anesthetized rats [27]. Nonetheless, it is interesting that PO thalamic neurons have been shown to respond to vibrissa movements in the latter condition. In this respect, it remains possible that PO thalamic neurons are able to respond in a similar manner to SpVIr during a currently unknown behavioral context.
In the absence of proprioception (Figs 2 and 3) and corollary discharge [12], encoding of self-generated vibrissa movement through re-afferent activation of mechanoreceptors is a means for the animal to compute the position of its vibrissae [17]. This can be used to modulate the sensory response to touch depending on phase in the whisk cycle (Fig 11). Why does self-generated movement appear to be represented differently in the vibrissa system than in the limbs, where proprioceptive and cutaneous signals are encoded in separate thalamocortical pathways [73,74]? One possible explanation is that the limbs, which support the body, are likely to carry a variable load. Accurate positioning therefore requires sensory information related to muscle length and force that is independent of tactile sensation. This may also be true for jaw muscles, which are innervated by muscle spindle fibers [43] and corresponding γ-motoneurons (Fig 3) [46,75]. The vibrissa muscles, on the other hand, support only a small, relatively constant load that consists solely of the vibrissae, which readily flex upon the application of external forces [76,77]. While proprioception appears to exist in the extraocular muscles [78,79], other facial muscles that carry a small, relatively constant load are thought to be devoid of proprioceptive innervation [7–11]. We can only conjecture that facial expression control may follow similar mechanisms. In the case of other facial movements in which self-motion is encoded by exo- as opposed to endo-receptors, any position-dependent signal may serve as a reference signal for computing sensation in terms of sensor position.
This study was performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocol was approved by the Committee on the Ethics of Animal Experiments of the University of California at San Diego (Protocol numbers: S02173 and S02174R).
Fifty-four female Long Evans rats, 250 to 350 g in mass (Charles River), were used for combined anatomical, behavioral, and electrophysiological experiments. All behavior and electrophysiological data were obtained from head-restrained rats [80,81].
Rats were transcardially perfused with 0.1 M phosphate buffered saline (PBS) followed by 4% (w/v) paraformaldehyde in PBS. Whole rat heads were post fixed for 4 to 12 h at 4°C. Muscles were dissected off of the fixed heads and cryoprotected in 30% (w/v) sucrose in PBS for 8 to 12 h at 4°C.
Both mystacial pad and masseter muscles were sectioned tangentially at a thickness of 60 μm with a sliding microtome. Sections were incubated in 2% (w/v) goat serum (S-1000, Vector) block for 30 min and then the primary antibody rabbit anti Neurofilament H 1:500 (Ab 1991, Millipore) overnight at room temperature. For fluorescent staining, secondary antibodies raised in goat were used (rabbit anti-594, A-11012, Life technologies). For dark product staining, sections were incubated in biotinylated rabbit secondary antibody (BA-1000, Vector) for 90 min followed by processing with an ABC kit (PK-6100, Vector) and the SG peroxidase kit (SK-4705, Vector). Sections were either initially counterstained with cytochrome oxidase or a solution of 0.25% (w/v) Eosin Y in 79% ethanol and 21% water.
Mystacial pad and masseter muscles were frozen in blocks of OCT (25608–930, Tissue-Tek) and sectioned transversely at a thickness of 10 μm with a cryostat. Sections were directly mounted on slides to maintain the integrity and orientation of the muscle fibers. They were left to dry for a minimum of 1 h. Slides were rehydrated and, sequentially, incubated in Mayer’s Hematoxylin Solution (MHS15-500, Sigma-Aldrich) for 8 min, washed with running tap water for 5 min, differentiated in a 1% (v/v) hydrochloric acid in distilled water for 30 s, further washed with running tap water for 2 min, “blued” in a saturated lithium carbonate solution (1.4% [w/v] lithium carbonate in distilled water) for 30 to 60 s, washed for 5 min in running tap water, rinsed by dipping 5 to 7 times in 95% (v/v) ethanol in water, counterstained with a 0.25% (w/v) Eosin Y solution in 79% ethanol and 21% tap water for 2 min, finally dried in air, and cover slipped using mounting media (06522, Sigma Aldrich).
Confocal stacks of images of spindle fibers were obtained with a Leica Sp5. Dark product, hematoyxlin, and eosin stained slides were imaged with a slide scanning microscope (Nanozoomer 2.0 HT, Hamamatsu). Fibers were counted using Photoshop (CS4, Adobe).
Rats were perfused and fixed and the brains were extracted and sectioned at a thickness of 30 μm, as above. Sections containing trigeminal and facial motor nuclei were incubated overnight in anti-ChAT (1:100 AB144P, Millipore) and anti-NeuN (either 1:100 MAB377, Millipore, or 10 μg/mL of a custom anti-NeuN directly conjugated to Alexa 594(Chemicon [82]). Sections were then rinsed and incubated for 90 min in anti-goat Alexa 488 (1:200 A11055, Invitrogen) and anti-mouse Alexa 647 (1:200 A31571, Invitrogen), rinsed again, mounted, and coverslipped.
Slides were scanned as described above. Motoneurons in the trigeminal and facial motor nuclei that contained a DAPI-stained nucleus were manually outlined based only on the anti-ChAT label (green channel) using Neurolucida software. The area and average intensity of the anti-NeuN label (red channel) within the outlined perimeter was then calculated.
Vibrissae were clipped to approximately 2/3 of their original length and vibrissa position was monitored simultaneously with neuronal spiking activity under two behavioral conditions. First, as the rats were coaxed to whisk in air by presenting food or bedding from their home cage [83,89]. Second, as vibrissae were deflected externally by brief puffs of air applied to the face [81,83]. We monitored vibrissa position with a Basler A602f camera and a white light emitting diode backlight [50]. We chose a spatial resolution of 120 μm/pixel, a field of 360 × 250 pixels, a frame rate of 250 Hz, and a trial time of 10 s. The pixel intensity in the image was thresholded and the mean position of the full set of vibrissae was tracked by computing the center of mass of the thresholded pixels in each frame. The data were then converted into whisking angle versus time, denoted θ(t). Lastly, a Hilbert transform was used to decompose the whisking angle, θ(t), into the phase within the whisk cycle, ϕ(t), with
θ(t)=θAmplitudecos[ϕ(t)] + θMidpoint
(1)
where θAmplitude and θMidpoint are slowly varying parameters and the whisking frequency, fwhisk, is given by [49]:
fWhisk=12π dϕ(t)dt.
(2)
Lastly, we recall that the vibrissae tend to move in phase with one another during free-air whisking [49]; thus the phase, but not the amplitude or midpoint, of all vibrissae may be taken as identical.
|
10.1371/journal.ppat.1003877 | Alphavirus Mutator Variants Present Host-Specific Defects and Attenuation in Mammalian and Insect Models | Arboviruses cycle through both vertebrates and invertebrates, which requires them to adapt to disparate hosts while maintaining genetic integrity during genome replication. To study the genetic mechanisms and determinants of these processes, we use chikungunya virus (CHIKV), a re-emerging human pathogen transmitted by the Aedes mosquito. We previously isolated a high fidelity (or antimutator) polymerase variant, C483Y, which had decreased fitness in both mammalian and mosquito hosts, suggesting this residue may be a key molecular determinant. To further investigate effects of position 483 on RNA-dependent RNA-polymerase (RdRp) fidelity, we substituted every amino acid at this position. We isolated novel mutators with decreased replication fidelity and higher mutation frequencies, allowing us to examine the fitness of error-prone arbovirus variants. Although CHIKV mutators displayed no major replication defects in mammalian cell culture, they had reduced specific infectivity and were attenuated in vivo. Unexpectedly, mutator phenotypes were suppressed in mosquito cells and the variants exhibited significant defects in RNA synthesis. Consequently, these replication defects resulted in strong selection for reversion during infection of mosquitoes. Since residue 483 is conserved among alphaviruses, we examined the analogous mutations in Sindbis virus (SINV), which also reduced polymerase fidelity and generated replication defects in mosquito cells. However, replication defects were mosquito cell-specific and were not observed in Drosophila S2 cells, allowing us to evaluate the potential attenuation of mutators in insect models where pressure for reversion was absent. Indeed, the SINV mutator variant was attenuated in fruit flies. These findings confirm that residue 483 is a determinant regulating alphavirus polymerase fidelity and demonstrate proof of principle that arboviruses can be attenuated in mammalian and insect hosts by reducing fidelity.
| Chikungunya (CHIKV) is a re-emerging mosquito-borne virus that constitutes a major and growing human health burden. Like all RNA viruses, during viral replication CHIKV copies its genome using a polymerase that makes an average of one mistake per replication cycle. Therefore, a single virus generates millions of viral progeny that carry a multitude of distinct mutations in their genomes. In this study, we isolated CHIKV mutators (strains that make more errors than the wildtype virus), to study how higher mutation rates affect fitness in arthropod-borne viruses (arboviruses). CHIKV mutators have reduced virulence in mice and severe replication defects in Aedes mosquito cells. However, these replication defects result in selective pressure for reversion of mutators to a wildtype polymerase in mosquito hosts. To examine how mutators would behave in an insect model in absence of this genetic instability, we isolated mutators of a related virus, Sindbis virus (SINV). SINV mutators had no replication defect in fruit fly (Drosophila) cells, and a SINV mutator strain was stable and attenuated in fruit flies. This work shows proof of principle that arbovirus mutators can exhibit attenuation in both mammalian and insect hosts, and may remain a viable vaccine strategy.
| During replication, RNA viruses generate approximately 1 error per 104 nucleotides copied, giving rise to an immense population of genetically distinct but closely related variants [1], [2], [3], [4]. The genetic diversity of these “mutant swarms” is not detected by consensus sequencing, which to-date has been the basis for most studies of viral infection. However, this lack of information on genetic diversity has obscured crucial aspects of virus biology. Although RNA-dependent RNA polymerases (RdRp) have a high intrinsic error rate, their mutation rates can be altered to generate both higher and lower fidelity variants (antimutators and mutators, respectively) [5], [6], [7], [8], [9], [10], [11], [12], [13], [14]. Thus far, antimutator variants are thought to replicate more slowly, making fewer genomes with greater accuracy; in contrast, mutator variants have been shown to replicate more quickly, synthesizing more viral genomes but introducing many errors during the replication process [8], [15], [16], [17], [18]. Despite this, overall growth and titers of polymerase fidelity variants are not significantly different when grown in isolation in cell culture; for mutators the negative effects of accumulating deleterious mutations are only noticeable after several rounds of replication [6], [19]. In recent works, these variants have been useful in exploring how the course of viral infection is affected by either restricted or expanded population diversity [4], [10].
Current evidence indicates that mutation frequencies of RNA viruses have been optimized over time to be neither too accurate nor too erroneous [4], [6], [16], [20], [21], [22], [23]. It is thought that error-prone replication allows the virus to explore sequence space to gain adaptability and accumulate potentially advantageous mutations. For several RNA viruses, limiting viral population diversity has fitness costs in vivo. Despite similar in vitro growth phenotypes, variants that make fewer errors have reduced titers and exhibit restricted tropism in animal models [19], . This restriction in tropism may be due to cooperative inter-variant interactions or beneficial minority variants that are missing in a situation with restricted population diversity [19]. It is also proposed that high mutation rates of influenza A may contribute to altered tropism, allowing infection of new hosts [26]. Therefore, it seems that the relatively high error rates of RNA viruses generate a level of diversity that facilitates adaptive fitness advantages.
In contrast, there is also an upper threshold to mutation frequencies; if crossed, extreme error rates lead to the accumulation of deleterious mutations and loss of genetic integrity. Evidence for this is demonstrated by treatment of numerous RNA viruses with nucleoside analog mutagens, which increase mutation frequencies and result in extinction by lethal mutagenesis [27], [28], [29], [30], [31]. Although thus far RdRp mutators have not exhibited growth defects in isolation in vitro, a recent paper showed that HIV mutator and antimutator strains were less fit than wildtype in competition assays [32]. In addition, several studies recently report in vivo attenuation of mutator strains: Coxsackie virus B3 mutator strains present reduced viral titers in key organs and fail to establish persistent infections in mice [6], and a severe acute respiratory syndrome (SARS) coronavirus mutator strain exhibits reduced pathogenesis in several mouse models [7].
Antimutator and mutator variants are valuable tools to study where the threshold of advantageous polymerase error exists for viruses facing different selective pressures. In this respect, arboviruses represent a special evolutionary position due to their need to replicate in disparate hosts, which is accompanied by distinct selective pressures. Arbovirus fitness is not necessarily reduced due to obligate host-cycling (alternating passages of CHIKV did not limit viral fitness), yet it has been shown that evolvability may be reduced due to these evolutionary constraints [33], [34], [35], [36], [37], [38]. For alphaviruses, evidence suggests that viral diversity is most restricted in the insect host, due to more stringent population bottlenecks and selective pressures [33], [34], [35], [37], [39], [40]. Since minority variants are thought to play important roles in arbovirus pathogenesis, transmission, and emergence [25], [41], [42], [43], [44], [45], the implications of altered polymerase fidelity and mutation rates merit further study. Recently, this question was partially addressed using a chikungunya virus antimutator variant [25].
Chikungunya virus (CHIKV) is a re-emerging arbovirus, transmitted by Aedes species mosquitoes. This positive-stranded RNA virus (family Togaviridae, genus Alphavirus) has an 11.8 kB genome, of which the first 7.5 kb encode four nonstructural proteins (nsP1-4) involved in diverse processes including RNA synthesis, immune evasion, and host tropism [46], [47], [48], [49], [50], [51], [52]. In most cases, functions of these proteins are putative in CHIKV and have only been shown in related model viruses, such as Semliki forest virus and Sindbis virus [53], [54], [55], [56]. Nsp4 is the RdRp, responsible for nucleotide incorporation during replication [57]. Previously, we isolated an antimutator strain of CHIKV by passaging virus in ribavirin, an RNA nucleoside analog. Ribavirin causes nucleotide misincorporation by the RdRp, adding selective pressure for an intrinsically more faithful polymerase [14], [58], [59]. This antimutator strain harbored a single amino acid change (483Y) in nsp4. Although 483Y showed no growth defects in vitro, the variant was moderately attenuated in vivo in both mammalian and mosquito hosts [25]. However, no arbovirus mutators have been isolated thus far. To this end, we mutated the conserved cysteine residue at position 483 to obtain several mutators in the arboviruses CHIKV and SINV, confirming this position's importance in determining alphavirus fidelity. We used these novel mutator strains to examine how increased polymerase error affects arbovirus fitness in vitro and in vivo; interestingly, mutator strains presented distinct cell- and host-specific phenotypes.
Mammalian cell lines Vero, HeLa, and BHK-21 were maintained in DMEM (Gibco) supplemented with 10% newborn calf serum (NCS, Gibco) and 1% penicillin-streptomycin (P/S, Sigma), at 37°C with 5% CO2. Mosquito cell lines C6/36 and U4.4 (Aedes albopictus) and Aag2 (Aedes aegypti) were grown in L-15 media, supplemented with 10% fetal bovine serum (FBS, Gibco), 1% P/S, 1% tryptose phosphate, and 1% non-essential amino acids (NEAA), at 28°C with 5% CO2. Drosophila melanogaster S2 cells were grown in Schneider's Drosophila media (Gibco), supplemented with 10% FBS, 1% L-Glutamine, and 1% P/S at 25°C.
Wildtype CHIKV was generated from the La Reunion strain 06-049 infectious clone, previously described [33]. Nsp4 position 483 mutants were generated by site-directed mutagenesis of the infectious clone using the QuikChange II XL Site-Directed Mutagenesis kit (Stratagene). All newly generated DNA plasmids were Sanger sequenced in full (GATC Biotech) to confirm mutagenesis of position 483 and to ensure no second-site mutations were introduced. Select SINV mutants were constructed in the same fashion from the pTR339 wildtype infectious clone [60]. CHIKV and SINV expression plasmids were linearized with NotI or XhoI respectively, purified by phenol-chloroform extraction and ethanol precipitation, and subsequently used for in vitro transcription of viral RNAs using the SP6 mMESSAGE mMACHINE kit (Ambion). RNAs were then purified by phenol:chloroform extraction and ethanol precipitation, quantified, diluted to 1 µg/µl and stored at −80°C.
For RNA transfections, BHK-21 cells were trypsinized, washed twice with ice-cold PBS, and resuspended at a concentration of 2×107 cells/ml in ice-cold PBS. Cells (0.390 ml) were mixed with 10 µg of in vitro transcribed viral RNA, placed in 2 mm cuvette and electroporated at 1.2 kV, 25 µF with infinite Ω in a XCell Gene Pulser (BioRad). Cells were allowed to recover for 10 minutes at room temperature then mixed with 6 ml of pre-warmed media and placed into a T-25 flask. After 48 hours incubation at 37°C, viral titers were determined by standard plaque assay. In brief, 10-fold serial dilutions of each virus in DMEM were incubated on a confluent monolayer of Vero cells for 1 hour at 37°C. Following incubation, cells were overlaid with 0.8% agarose dissolved in DMEM and 2% NCS and incubated at 37°C for 72 hours. The cells were then fixed with 4% formalin for 1 hour, the agarose plugs were removed, and plaques were visualized by the addition of crystal violet. Plaque size was quantified by scanning the crystal violet-stained cell monolayer, then quantifying the size of each individual plaque in square millimeters using ImageJ (http://rsbweb.nih.gov/ij).
Each virus was then passaged once over a 70–80% confluent monolayer of BHK-21 cells, titered as described above, aliquoted, and stored at −80°C until use. To analyze each virus for reversion at position 483, viral RNA was extracted for each electroporation and BHK-21 passage using TRIzol reagent (Invitrogen). For CHIKV, this RNA was used to amplify a 3184 bp region corresponding to nucleotides (4522–7706), which included position 483, using the forward primer (5′-GATGAGCACATCTCCATAG-3′) and the reverse primer (5′-GTTTGGGTTGGGATGAACT-3′) and the Titan One Tube RT-PCR Kit (Roche). For SINV, a 2225 bp region (nucleotides 6556–8781) was amplified in the same fashion using forward primer (5′-ACCAGGCACGAAACACACAGAA-3′) and reverse primer (5′-ACTGGGCGGAAGTCTGTATGCG -3′). Each PCR product was cleaned using the Nucleospin PCR and Gel Extraction Kit (Macherey-Nagel) and Sanger sequenced at position 483/482 to confirm genetic stability. At passage 3, all viruses used were fully sequenced to ensure no second site mutations.
HeLa cells (250,000 cells/well in 12-well tissue culture plates) were pre-treated for two hours with either media containing no mutagen, or media containing 200 µM or 400 µM ribavirin (Sigma). Post-treatment, media was removed and the cells were inoculated with virus in DMEM at an MOI 0.1 for one hour at 37°C. Following incubation, mutagen-containing media was replaced and cells were incubated for 72 hours at 37°C. Virus was harvested at 72 hours and mean titers were obtained by TCID50. In brief, a 96-well tissue culture plates was plated for each virus with 1×104 Vero cells/well. Viruses were serially diluted in 8 ten-fold dilutions in DMEM. Each dilution was distributed in a row of the 96-well plate, with each well receiving 100 µl of diluted virus. Viruses and cells were incubated 5–7 days at 37°C with 5% CO2. Following incubation, cells were fixed with 50 µl of 4% formalin for 30 minutes. All media were removed, and 50 µl of crystal violet was added to each well. Viruses that exhibited significant sensitivity or resistance compared to wildtype at P<0.05 or greater at either 200 µM or 400 µM ribavirin were considered potential fidelity variants, and mutation frequencies were estimated (Table 1).
To determine mutation frequencies, all mutants were electroporated in tandem into BHK-21 cells. Supernatants were collected 48 hours later and viral RNA was extracted. For CHIKV, an approximately 800 bp region corresponding to nucleotides 9943–10726 was amplified of the E1 region of the genome using forward primer 5′-TACGAACACGTAACAGTGATCC-3′ and reverse primer 5′-CGCTCTTACCGGGTTTGTTG-3′. For SINV, the analogous region was amplified using forward primer 5′-TACGAACATGCGACCACTGTTC-3′ and reverse primer 5′-CGCTCGGAGCGGATTTACTG-3′, and approximately 500 bases of this fragment was included in the analysis. Amplified fragments were purified as described above, and 3 µl of each product was modified by a 3′ A-overhang addition reaction (1 µl AmpliTaq Gold 10× buffer, 1 µl 10 mM dATP). Modified products were cloned using the TopoTA cloning kit (Invitrogen), and single colonies were picked for sequencing. Mutation frequencies were determined as previously described [61]. Mutation frequencies in mosquito cells were obtained in the same fashion using the samples obtained from C6/36 growth curves (we determined mutation frequency for a wildtype sample electroporated into C6/36 cells, and there was no difference between samples generated by infection or electroporation; the nonviability of mutators transfected into mosquito cells made it impossible to estimate mutation frequencies in C6/36 by electroporation). We sequenced approximately 75 clones per viral population in C6/36 cells. Mutation frequencies from mouse muscle were determined using RNA extracted from homogenized muscle samples from mice that most closely represented the median titer for that variants. For estimating in vivo mutation frequencies, a minimum of 50 clones were sampled per population. To confirm that the presence of RNA or aberrant viral particles in supernatants/homogenates did not affect mutation frequencies, we purified virus on 20% sucrose cushion and re-estimated mutation frequencies; no differences were observed.
To estimate the population diversity of variants by deep sequencing, cDNA libraries were prepared by Superscript III from RNA extracted from virus generated in BHK-21 or C6/36 cells, and the viral genome was amplified using a high fidelity polymerase (Phusion) to generate 5 overlapping amplicons 2–3 kb in length. PCRs were fragmented (Fragmentase), multiplexed, clustered, sequenced in the same lane with Illumina cBot and GAIIX technology and analyzed with established deep sequencing data analysis tools and in house scripts. Briefly, per-base Phred quality scores were utilized to trim bases with error probabilities higher than 0.001, and sequences with less than 16 bases after trimming were discarded. For this purpose we used the fastq-mcf tool from the ea-utils toolkit at http://code.google.com/p/ea-utils [62]. The alignment step is performed using Burrows Wheeler Aligner [63] and Pileup is performed using SAMtools [64]. Once the pileup is done, an in-house script collects the data per-position and calculates the variance at each nucleotide position by root mean square deviation (RMSD) and determines the mean variance and standard error across the whole genome [65].
To estimate population diversity in a phenotypic assay, we performed neutralization assays using viruses which had been passaged 3 times on BHK-21 cells, using the n Neutralizing antibody CHK-102 (a kind gift from Dr. M.S. Diamond [66]). 100 pfu of wildtype and mutator CHIKV strains were incubated for 1 hour at 37°C with serial dilutions of antibody, ranging from 2 µg/ml to .0001 µg/mL, or left untreated. Virus-antibody complexes were added to pre-seeded confluent monolayers of Vero cells, and allowed to bind at 37°C for 1 hour. Assays were then overlayed with agarose and developed as described above for a plaque assay. Plaques were counted and normalized to the untreated control for each virus.
Virus growth was evaluated for WT and all mutant viruses in BHK-21, C6/36, U4.4, Aag2, and S2 cells and titers were determined by TCID50 on Vero cells as described above. Using the 24 hour time point from the C6/36 growth curve, we also performed a cytopathic effect (CPE) assay on C6/36 cells on all viruses (CellTiter 96 AQueuos One Solution Cell Proliferation Assay (MTS) kit; Promega). We obtained similar titers by standard TCID50 and CPE assay, indicating that viruses amplified on mosquito cells were still equally infectious when titered on Vero cells. For CHIKV, genome copy number was determined by extracting viral RNA from the supernatant at each time point using the TRIzol reagent and performing quantitative RT-PCR (qRT-PCR) using the TaqMan RNA-to-Ct kit (Applied Biosystems). Ct values were determined in duplicate based on amplification of nsp4 transcripts using forward (5′-TCACTCCCTGCTGGACTTGATAGA-3′) and reverse (5′-TGACGAACAGAGTTAGGAACATACC-3′) primers and probe 5′- [6-FAM] AGGTACGCGCTTCAAGTTCGGCG-3′ as previously published [33], [67]. To determine genome copy number for SINV, viral RNA was extracted in the same manner and quantitative PCR was performed based on amplification of nsp3 transcripts using forward (5′-AAAACGCCTACCATGCAGTG-3′) and reverse (5′-TTTTCCGGCTGCGTAAATGC-3′) primers and the SYBR green PCR master mix (Applied Biosystems). Standard curves were performed in each run using samples of in vitro transcribed CHIKV or SINV RNA.
In vitro transcribed RNA was transfected in BHK-21 cells in duplicate, as described above or at 28°C, including RNA from a construct in which the polymerase active site (GDD) was replaced with GNN by site-directed mutagenesis to abrogate replication and alongside a mock transfection where no RNA was added. Transfections in C6/36 and U4.4 cells were modified by pulsing with 250 V, 50 µF, and 550 Ω. Forty-eight hours post-transfection, supernatant containing progeny virus was collected. Cells were washed twice in PBS and RNA was TRIzol (Invitrogen) extracted, quantified and diluted to the same concentration. Samples were prepared in NorthernMax formaldehyde loading dye (Ambion) with 1 µl of ethidium bromide, heated to 65°C for 10 minutes, then separated on a 1.2% LE agarose (Lonza) gel containing 1× morpholinepropanesulfonic acid (MOPS) running buffer (Ambion) and 6.7% formaldehyde. RNA was transferred onto nitrocellulose membrane, cross-linked by ultraviolet irradiation (UVP), and prehybridized at 68°C for 1 hour in ULTRAhyb ultrasensitive hybridization buffer (Ambion). A plasmid used for the expression of CHIKV RNA probes corresponding to the 3′ portion of the E2 glycoprotein was generated by first amplifying the region of the CHIKV genome from 8703 (5′-GAAGCGACAGACGGGACG-3′) to 9266 (5′-GTTACATTTGCCAGCGGAA-3′) by PCR and subsequently TOPO-TA cloning the PCR product into the pCRTOPO-II vector. RNA probes complementary to positive strand RNA were labeled with 32P using the MAXIscript SP6 In Vitro Transcription Kit (Ambion), unincorporated nucleotides were removed using illustra MicroSpin S200 HR columns (GE healthcare), and probe was hybridized to the membrane overnight at 68°C. Membranes were washed several times at 68°C with 0.1× SSC with 0.1% SDS, then imaged using Amersham Hyperfilm MP autoradiography film (GE Healthcare). Quantification was done using ImageJ (http://rsbweb.nih.gov/ij).
C57BL/6 mice (Janvier) or CD-1 mice (Charles River) were housed according to Institut Pasteur guidelines in biosafety level 3 isolators, with the approved experimental protocol #10.620, reviewed by the Institut Pasteur ethics committee under dossier #CETEA 2013-0021. At 8-days old, litters of C57BL/6 were inoculated with 200 pfu of wildtype or mutant CHIKV viruses subcutaneously (n = 4/variant). Eight-day old CD-1 litters were inoculated with 100 pfu of wildtype or mutant SINV strains in the same fashion, and monitored for symptoms of hind limb paralysis and survival. In addition, seven days post-infection, CHIKV and SINV-infected mice were sacrificed and brains, thigh muscles, livers and blood were harvested and homogenized in 300 µl of PBS at 30 shakes/second for 2 min (MM300 Retsch). RNA was extracted and viral genome copies were determined by qRT-PCR as described.
Principal CHIKV vectors Ae. albopictus Providence (ALPROV, F8 generation) from La Reunion and Ae. aegypti Paea (PAE, a lab colony at Institut Pasteur since 1994) from Tahiti, in French Polynesia were fed on artificial bloodmeals containing 106 pfu/ml of virus in PBS-washed rabbit blood [68]. CHIKV wildtype and mutators were fed to both Ae. albopictus and Ae. aegypti, and SINV wildtype and mutator 482G were fed to Ae. aegypti. The blood meals were warmed to 37°C and presented to 10 day-old females in membrane feeders, and engorged mosquitoes were incubated for 7 days. Seven days post infection, mosquitoes were dissected to obtain legs and wings, and saliva was obtained by in vitro transmission assay; in brief, mosquitoes were salivated for 30–45 min by placing the proboscis in a pipette tip containing FBS. Following salivation, bodies were frozen. To confirm ingestion, a sample of engorged mosquitoes was immediately homogenized at time 0. Samples were homogenized as described for mouse tissues, RNA was extracted, and qRT-PCR was performed. A standard curve was generated using serial dilutions of a CHIKV bloodmeal of known titer.
Drosophila melanogaster flies (strain w1118) were reared on standard medium at 25°C. Three- to four-day-old female flies were injected with 50 nL of a virus dilution containing 400 pfu in 10 mM Tris-HCl (pH 7.5) using a Drummond nanoject injector as previously described [69]. Fly mortality at day 1 was attributed to damage produced by the injection, and these flies were excluded from further analyses. Mortality was monitored daily for 10 days, and every 3–4 days flies were transferred to fresh vials. In all experiments, 30–60 flies per genotype group were injected. Homogenates of individual flies were titrated on by plaque assay on Vero cells, as described above.
All experiments were performed in triplicate unless noted otherwise. Statistics, noted where applied, were performed in Microsoft Excel and GraphPad Prism. P-values>0.05 were considered non-significant (ns).
We previously described a CHIKV antimutator variant that possessed a single amino acid change from a cysteine to a tyrosine at position 483 (C483Y) of the RNA-dependent RNA polymerase nsp4 [25]. Since Coxsackie virus B3 mutator strains are situated in a structurally analogous area, we hypothesized that this position plays important roles in modulating intrinsic CHIKV RdRp fidelity [6]. To address this, we substituted each amino acid at position 483 of the CHIKV full-length infectious clone (Table 1). After three passages in BHK-21 cells, viruses were Sanger sequenced to determine genetic stability. Of the 19 substitutions, 12 were viable and genetically stable (Table 1). This high number of viable variants indicates that position 483 has structural plasticity and can tolerate a wider range of substitutions than in previous attempts at generating fidelity variants of other RNA viruses [6], [19]. Interestingly, unstable viruses did not readily revert to wildtype, but mutated to other variants, including the antimutator form of the protein, 483Y (Table 1). The only strict biochemical requirement we observed was a necessity for uncharged residues, as all variants with charged residues (483D, E, H, K, or R) were unstable or not recoverable. In addition, we observed a general correlation between hydrophobicity of the substituted amino acid and stability or viability of the variant, where hydrophobic amino acids were preferred. Finally, as a first characterization of virus fitness, we measured the mean size of plaques. Variants 483A, G, L, N, Q, T, and W had significantly smaller plaques than wildtype (Table 1).
Because polymerase fidelity variants have altered intrinsic rates of (in)correct nucleotide incorporation, they have often been identified by their relative resistance or sensitivity to nucleoside analog RNA mutagens [14], [19], [25]. Therefore, we addressed the sensitivity of all 12 genetically stable variants to ribavirin (Table 1 and Figure 1A). Viruses were grown in the presence of either 200 µM or 400 µM ribavirin, or left untreated. We expect antimutator variants (such as 483Y) to demonstrate resistance, and mutator variants to demonstrate sensitivity when compared to wildtype. As previously described, the antimutator 483Y demonstrated significantly higher survival than wildtype (P<0.001, two-way ANOVA) as did 483M and 483N (P<0.001 for both, two-way ANOVA). Additionally, we identified several mutator candidates that were significantly more sensitive to ribavirin (483A, G, W, T, Q; P<0.05 for all, two-way ANOVA). All ribavirin-sensitive variants presented small-plaque phentoypes, as well as variant 483N (P<0.01 for all, Student's t-test). Though these variants presented small plaque phenotypes, virus stocks reached wildtype-like titers, with the exception of 483N and 483Q (Table 1).
As observed previously for picornaviruses [5], [6], [14], the ribavirin-resistant and -sensitive phenotypes of these CHIKV variants suggested altered polymerase fidelity. To address this further in a genetic assay, we estimated the mutation frequencies of each variant that demonstrated significantly altered ribavirin sensitivity at either concentration of ribavirin. Viral RNA from the supernatants of BHK-21 cells was extracted, and an approximately 800 nucleotide fragment of the E1 genome was amplified by RT-PCR and TOPO cloned as previously described [61]. We sequenced approximately 150 individual clones per viral population (corresponding to an average of 122,200 nucleotides) to calculate the mutation frequencies (Figure 1B and Table 1). Since previous studies with 483Y required >350 clones per population to distinguish more subtle differences in mutation frequencies [25], we could only statistically confirm the altered fidelities of three mutator strains (483A, G and W; P<0.05, P<0.001, P<0.01, respectively, χ2 test) (Figure 1B). We excluded variants that did not exhibit significant fidelity differences compared to wildtype (483M, N, and Q). As a complementary approach, we performed deep sequencing on these same virus populations to characterize the relative diversity in these virus populations. In accordance with the mutation frequency data, the mean variance across the whole genome was significantly lower for the antimutator 483Y variant (P = 0.0006, Mann-Whitney u test) and significantly higher for the 483A, G and W mutator variants, compared to wildtype virus (P<0.0001 for all, Mann-Whitney u test; Figure 1C).
Next, we examined growth of these variants in mammalian cells. As seen previously, the antimutator 483Y presented no significant difference in amount of progeny virus (Figure 2A) or number of genome copies (Figure 2B). As observed with Coxsackie virus mutators, CHIKV mutator strains (483A, G, and W) generated the same or more genomes than wildtype virus (Figure 2B), but slightly fewer infectious progeny (Figure 2A). Consequently, these mutator variants have a lower specific infectivity than wildtype in mammalian cells (Figure 1C). This is consistent with previously published results showing that mutator variants make more lethally mutagenized RNA [6], [22], [28], [70].
Recently, low fidelity polymerase mutators of Coxsackie virus and exonuclease activity deficient mutators of coronaviruses were shown to be attenuated in mice [6], [7]. To determine whether this holds true for alphaviruses, we administered a sublethal infection of either wildtype or 483A, G and W viruses to 8-day old C57BL/6 mice. At 7 days of infection, when titers peak and virus is rapidly cleared thereafter, viral loads were determined in different compartments (muscle, blood, brain, liver). Viral loads were significantly lower for all three mutator strains in each tissue (Figure 3A). Since the in vivo mutation frequencies of mutator strains had not been previously reported, we examined the virus populations in the muscle of the wildtype- or the mutator-infected mouse that presented the median viral load. Although we cannot predict whether selection will act differently on these variants in mice to potentially skew the mutation frequencies, they remained elevated to varying degrees for the mutator strains. Interestingly, higher mutation frequencies in vivo correlated with increased attenuation (Figure 3B).
Because arboviruses must cycle through both vertebrate and arthropod hosts, and since mutator strains of other RNA viruses were only examined in mammalian systems [6], [7], [19], [24], we addressed viral replication in three mosquito cell lines: Ae. albopictus C6/36 cells, Ae. aegypti Aag2 cells and Ae. albopictus U4.4 cells. The replication profile for the antimutator 483Y was indistinguishable from wildtype in all conditions. On the other hand, the mutator strains 483A, G, and W presented significantly lower infectious progeny in C6/36 (P<0.001 for all, two-way ANOVA; Figure 4A), Aag2 (P<0.05 for 483A and G, two-way ANOVA; Figure 4B) and U4.4 (P<0.05 for 483A and W, P<0.01 for 483G, two-way ANOVA; Figure 4C) cells. Unexpectedly, we observed unprecedented reduction in genomic RNA released into the supernatant in all three mosquito cell types (Figure 4D–F). These results are discordant with the existing literature that found mutator polymerases synthesize RNA at faster rates than wildtype [6], in which case decreases in virus titer resulted directly from the increased mutational burden.
Here, the reduced viral titers obtained in mosquito cells seem to result from a host-specific replication defect, rather than the effect of lethal mutation. To further distinguish between these two effects, we examined whether mutation frequencies differed in mosquito versus mammalian cells, comparing wildtype CHIKV to the mutator strains. It is important to note that because mutator strains replicate so poorly in mosquito cells, these strains may present artificially low mutation frequencies. Unfortunately, it is not possible to uncouple replication from mutation frequency in this model. Nevertheless, the mutation frequencies of all viruses, including wildtype), were lower in C6/36 cells (Figure 5) than in BHK-21 cells (Figure 1B). Furthermore, the significant differences that existed between mutators and wildtype in mammalian cells were negated in mosquito cells, as evidenced by molecular clone sequencing (Figure 5A) and whole-genome deep sequencing (Figure 5B). We thus hypothesized that the negative fitness cost of mutator polymerases in mosquito cells is more closely linked to replication defects. To further confirm this, we generated genetically homogenous in vitro transcribed RNA corresponding to each variant, which do not present the differences in mutation frequencies of virus stocks generated in cell culture. Following transfection of mammalian BHK-21 cells, there were no significant differences in RNA synthesis (Figure 6A) or production of infectious virus (Figure 6C); however, in mosquito C6/36 cells, there was a very marked defect in replication for the mutator variants, compared to wildtype virus or the antimutator 483Y strain (Figure 6B), that correlated with the significant reduction in progeny (P<0.01 for all mutators, one-way ANOVA; Figure 6D). Similarly, no detectable infectious progeny was produced following transfection of U4.4 cells with the mutator variants (Figure 6E), further confirming the replication defect observed during infection of cells with virus stocks.
To exclude the possibility that this replication defect is the result of temperature-sensitivity rather than host-specificity, we performed infections in mammalian BHK-21 cells at 28°C (mosquito cell temperature). We observed no difference in the growth of any variant compared to wildtype (Figure 6F). In addition, we transfected mammalian cells grown at 28°C, and saw no difference in subgenomic RNA synthesis, indicating that the reduced polymerase processivity of mutators in mosquito cells is not due to reduced temperature (Figure 6B and 6G). Finally, we determined whether lower temperature could be responsible for the reduced mutation frequencies we observed in mosquito cells. In mammalian cells at 28°C, mutator 483G makes significantly more mutations than in mosquito cells at 28°C (P<0.05, χ2 test; Figure 6H). In contrast to what we observed in mosquito cells, mutator 483G also made significantly more mutations than WT (P<0.05, χ2 test; Figure 6H). These data indicate that lower temperature is responsible for neither the replication defects nor the reductions in mutation frequencies we observed in mosquito cells.
Since host-specific replication defects were observed in mosquito cell culture, we hypothesized that these variants would be even more attenuated in mosquitoes than in mice. We orally infected both Aedes species CHIKV hosts (Ae. albopictus and Ae. aegypti) with a blood meal containing either wildtype or the 483A, G and W mutators. Seven days after infection, when CHIKV has reached peak titers, we quantified viral loads in bodies (infection), legs and wings (dissemination) and saliva (transmission) of individual mosquitoes (Figure 7). Surprisingly, no significant defect was observed in either Aedes species for any of the variants. To address the possibility that the fitness cost of defective replication, observed in mosquito cell culture, would favor the reversion of these mutant polymerases to wildtype, we deep sequenced virus from the body of an individual mosquito that presented the median titer from each group. Indeed, reversion to wildtype (or other replication competent variants, such as 483T or 483V) occurred in 483A (81%), 482G (93%) and 483W (39%). Whether position 483 changed to wildtype depended on the genetic distance of the mutated codon from wildtype: for example, W (TGG) reverted completely to WT (TGT), while A (GCT) reverted predominantly to a combination of V (66%; GTT) and T (ACT; 13%). Interestingly, when we examined higher passages (passage 3) of mutators in C6/36 cell culture, we also observed varying levels of reversion (ranging from less than 1% to as much as 50%), highlighting the strong selective pressure acting against this replication defect.
After confirming that polymerase position 483 plays an important role in modulating fidelity in CHIKV, we examined if this residue is a universal fidelity determinant among the alphaviruses. Indeed, this region of the nsp4 gene containing a cysteine is conserved across the alphavirus family (Figure 8A). Thus, we generated the analogous fidelity variants (482A, G, W) in the well-studied, distantly related alphavirus Sindbis virus (SINV). Genetically stable mutants (482A and G) were screened for changes in ribavirin sensitivity. Both showed significantly higher sensitivity than wildtype SINV (for 482A, at least P<0.01, for 482G, P<0.05, two-way ANOVA; Figure 8B). Moreover, the mutation frequencies determined by molecular clone sequencing confirmed the mutator phenotypes suggested by ribavirin screening (Figure 8C): in comparison to wildtype that presented 3.3 mutations per 10,000 nucleotides, 482A presented 6.0, and 482G presented 6.9 (P<0.05, χ2 test). This confirms that this conserved residue is a general fidelity determinant for the alphaviruses.
We next addressed whether replication defects also existed for these SINV mutators. In mammalian BHK-21 cells, mutator variants produce near wildtype-like titers of infectious particles (Figure 8D), and the same amounts of extracellular RNA genomes (Figure 8E). Importantly, as was observed for CHIKV strains, the SINV mutators presented more significant drops in virus titers in mosquito C6/36 cells (P<0.001, two-way ANOVA; Figure 8F), that correlated with a significant decrease in extracellular RNA genomes (P<0.001, two-way ANOVA; Figure 8G). Given the similarity of in vitro, host-specific phenotypes of CHIKV and SINV mutators, we hypothesized that SINV mutators would behave as CHIKV mutators in vivo (exhibiting attenuation in a mouse model and reversion in mosquitoes). We inoculated 8-day old mice with wildtype and 482G SINV strains, and observed significantly higher survival in mice infected with the mutator (91% compared to 50% for the wildtype, P = 0.0474; Figure 9A). In addition, only 36% of mice inoculated with 482G exhibited complete hind limb paralysis, compared to 100% of mice infected with wildtype SINV (P<0.0001, χ2 test; Figure 9A). Interestingly, this reduced paralysis correlated with significantly lower titers in the brain at day 7 post-infection (P<0.05, Student's t-test), confirming the attenuation of mutator strains in mammalian models in yet another virus (Figure 9B). We next examined the in vivo phenotype of SINV mutator 482G in Ae. aegypti mosquitoes. As expected, we observed reversion of position 482G to wildtype, and therefore, no differences in titers in the mosquito host (Figure 9C–E).
Since SINV has a broader host range than CHIKV, we examined whether the replication defect was mosquito cell-specific, or more general to insects, by infecting Drosophila S2 cells. Interestingly, the mutator strains were replication competent, generating virus titers (Figure 10A) and RNA genome copies (Figure 10B) at levels comparable to wildtype virus. Finally, we injected Drosophila with SINV wildtype and 482G mutator and followed the kinetics of infection by titering virus in flies for seven days post-infection. In contrast to CHIKV and SINV mutators in mosquitoes, when Drosophila flies were infected with wildtype and mutator strains of SINV, mutator 482G presented significantly lower titers than wildtype on day 3 and 5 (P<0.01, Student's t-test; Figure 10C). Sequencing of virus from 482G-infected flies at day 3 and 5 confirmed that no reversion had occurred. These results indicate that in principle, mutators can be attenuated in insects.
Previous work on antimutator CHIKV 483Y suggested this residue could be important for determining intrinsic RdRp fidelity [25]. Although there is no crystal structure available for an alphavirus RdRp, structural models predict that position 483 is located in the same area of the RdRp that generated Coxsackie virus RdRp fidelity variants (Figure S1) [6]. By substituting all other amino acids at this position, more mutator variants were isolated than antimutators, consistent with variants obtained for Coxsackie virus and with variants identified by characterizing the mutation frequencies of previously published reverse transcriptase variants for HIV [6], [32]. To date, all viable RdRp fidelity variants present error rates that remain within the same order of magnitude as their wildtype counterpart [5], [6], [14], [25], [71], [72]. Interestingly, although biochemical assays using purified RdRp of picornaviruses indicate that altering fidelity beyond an order of magnitude is enzymatically possible, these viruses are not viable [73]. Together, these studies suggest that within this viable range, wildtype fidelity sits closer to higher fidelity than lower fidelity. This may be further reflection of how RNA viruses are considered to exist close to a maximum threshold of error [74]. In support of this, in conditions where this reversion does not occur, mutator CHIKV variants present more significant fitness defects in vivo (Figure 3) than antimutator virus [25].
A review of the antimutator and mutator RdRp variant literature in virology reveals the following trends: antimutator strains tend to generate less RNA in vitro, but have higher specific infectivity, and have only been reported to lose fitness in vivo or in competition assays (Figures 2 and 4, and [16], [19], [25]); while mutators generate more RNA in vitro, but of lower specific infectivity, with more prominent fitness defects in mice [6]. Accordingly, CHIKV mutators showed congruous trends, exhibiting no significant replication defects in BHK-21 cells, but showing marked attenuation in the mouse model. Importantly, we showed that the mutator status of these variants (higher mutation frequencies) was maintained in the mouse model at the primary site of CHIKV replication and was likely responsible for the observed attenuation (Figure 3).
However, when we examined CHIKV mutators in the invertebrate host, the previous trends for how mutators behave was reversed. First, the differences in mutation frequencies between wildtype and mutator strains became virtually indistinguishable in mosquito cells (Figure 5), although it is difficult to draw clear-cut conclusions given the reduced replication rate of mutators. For all viruses, mutation frequency was lower in mosquito cells compared to mammalian cells (Figure 1). The role that these differences may play in arbovirus evolvability and fitness remain contradictory. Our observations corroborate previous observations in alphaviruses that inter-host cycling slows adaptation [33], [35], [75]; while flavivirus studies report that diversity is maintained in the mosquito host [38], [41], [76], [77], [78], [79], [80]. Second, and contrary to expectations, we observed a severe replication defect in three different mosquito cell cultures (Figure 4), which had never been observed for RdRp fidelity variants in mammalian cell culture. The lower titers of infectious progeny were not the result of accumulation of detrimental mutations as was observed for mutators in mammalian hosts; rather, there was a direct defect in genomic RNA synthesis in mosquito cells (Figure 4 and 6). Interestingly, similar host-specific replication defects were observed for RdRp mutants of West Nile virus (although it is unclear if these variants have altered fidelity). While differences in host temperature do not seem to be the cause, the cellular host factors implicated or missing in these host cell lines remain to be elucidated. Finally, we could not address whether mutators were attenuated in vivo in mosquitoes; sequencing of virus populations from mosquitoes revealed partial or total reversion of the fidelity-altering residues at position 483. Although one could expect a variant with severe replication defects to be highly attenuated, it is possible that when coupled to a mutator phenotype, reversion would more quickly and favorably occur when the pressure to increase replication remains, as is the case in mosquitoes that are persistently infected. Whether this defect is general to all mutators in mosquitoes, or whether only amino acids A, G, and W at position 483 bear this curiously coupled mutator/replication effect remains to be seen (since not all variants at this position were defective, as 483Y has no replication defects in mosquito cells [25]). Isolation of additional arbovirus mutators mapping to other residues in the polymerase should resolve this issue.
Since the cysteine at position 483 is conserved in the alphavirus genus, we obtained additional arbovirus mutators in Sindbis virus. SINV mutators also showed severe replication defects in mosquito cells, and SINV mutator 482G exhibited the same phenotypes we previously observed in CHIKV mutators in both mice and mosquitoes. However, the wider host range of SINV allowed us to test whether these replication defects occur across all insects or if they were mosquito-specific [81], [82]. In S2 cells, mutators did not present replication defects, allowing us to test, in principle, whether mutators could be attenuated in an insect model (Drosophila flies). Indeed, in the absence of any in vitro replication defect and resulting pressure to revert, the mutator strain was attenuated in fruit flies. Thus, our results confirm that arbovirus mutators can, in principle, be attenuated in insects.
Since the isolation of the first antimutator variant of a RNA virus, the growing body of literature shows that either increasing or decreasing replication fidelity has detrimental effects to virus fitness [6], [7], [16], [19], [24], [25]. However, how mutation rates and replication capacity are coupled will require more study, and the degree of attenuation resulting from altering these biochemical properties needs to be more carefully evaluated. A future challenge will be to quantitatively link the measurements of mutation frequencies (average mutations per nucleotide sequenced) performed in this work to actual mutation rates (average mutations per nucleotide site per replication) [2], [83] and to in vitro biochemical fidelity (rates of incorporation of correct and incorrect nucleotides in absence of selection)[6], [8], [73], [84], [85], [86], [87]. It is possible that the higher mutation frequencies measured for these alphavirus mutator strains are partly skewed by their producing more RNA genomes in shorter replication cycles and thus accumulating mutations more rapidly, rather than incorporating more errors per genome during each replication cycle. Indeed, biochemical studies of single-nucleotide incorporation by other mutator polymerases confirm that mutators are both faster enzymes and have higher frequency of mis-incorporation events per replication. In absence of a biochemical assay for alphaviruses, new technologies using microfluidic single-cell analysis of virus strains during single replication cycles should help correlate mutation frequencies, mutation rates, and enzyme fidelity with more confidence.
Recent studies have proposed both antimutator and mutator strains as candidates for rationally designed live attenuated vaccines [6], [7], [19], [25], [71]. Overall, fidelity variants present attenuated titers in vivo that range from one to several orders of magnitude lower than wildtype virus. Whether this degree of attenuation is sufficient to elicit protective immunity without causing disease will require more careful evaluation in more relevant animal models, as virtually all work has been performed in mice using viruses that are often not natural mouse pathogens. In vitro systems and artificial hosts may alter many of the selective pressures to which a virus would be subjected in a natural host [88], [89], [90], [91]. The present study and other work highlight that intrinsic fidelity and the mutant spectrum are labile and subject to stringent and disparate selective pressures in different hosts [34], [35], [75], [76], [79], [82], [92], [93]. A more comprehensive understanding of the selective pressures in natural hosts is crucial to predicting how viruses will behave in vivo, and essential to evaluating the feasibility of using fidelity variants as vaccines, whether stand-alone or coupled with other, conventional attenuating mutations. Despite the necessity for further research, from a vaccine development perspective these data support that in principle, mutators can be attenuated in a wider range of hosts and may be viable candidates for live-attenuated vaccines.
|
10.1371/journal.pcbi.1003246 | dPeak: High Resolution Identification of Transcription Factor Binding Sites from PET and SET ChIP-Seq Data | Chromatin immunoprecipitation followed by high throughput sequencing (ChIP-Seq) has been successfully used for genome-wide profiling of transcription factor binding sites, histone modifications, and nucleosome occupancy in many model organisms and humans. Because the compact genomes of prokaryotes harbor many binding sites separated by only few base pairs, applications of ChIP-Seq in this domain have not reached their full potential. Applications in prokaryotic genomes are further hampered by the fact that well studied data analysis methods for ChIP-Seq do not result in a resolution required for deciphering the locations of nearby binding events. We generated single-end tag (SET) and paired-end tag (PET) ChIP-Seq data for factor in Escherichia coli (E. coli). Direct comparison of these datasets revealed that although PET assay enables higher resolution identification of binding events, standard ChIP-Seq analysis methods are not equipped to utilize PET-specific features of the data. To address this problem, we developed dPeak as a high resolution binding site identification (deconvolution) algorithm. dPeak implements a probabilistic model that accurately describes ChIP-Seq data generation process for both the SET and PET assays. For SET data, dPeak outperforms or performs comparably to the state-of-the-art high-resolution ChIP-Seq peak deconvolution algorithms such as PICS, GPS, and GEM. When coupled with PET data, dPeak significantly outperforms SET-based analysis with any of the current state-of-the-art methods. Experimental validations of a subset of dPeak predictions from PET ChIP-Seq data indicate that dPeak can estimate locations of binding events with as high as to resolution. Applications of dPeak to ChIP-Seq data in E. coli under aerobic and anaerobic conditions reveal closely located promoters that are differentially occupied and further illustrate the importance of high resolution analysis of ChIP-Seq data.
| Chromatin immunoprecipitation followed by high throughput sequencing (ChIP-Seq) is widely used for studying in vivo protein-DNA interactions genome-wide. Current state-of-the-art ChIP-Seq protocols utilize single-end tag (SET) assay which only sequences ends of DNA fragments in the library. Although paired-end tag (PET) sequencing is routinely used in other applications of next generation sequencing, it has not been much adapted to ChIP-Seq. We illustrate both experimentally and computationally that PET sequencing significantly improves the resolution of ChIP-Seq experiments and enables ChIP-Seq applications in compact genomes like Escherichia coli (E. coli). To enable efficient identification using PET ChIP-Seq data, we develop dPeak as a high resolution binding site identification algorithm. dPeak implements probabilistic models for both SET and PET data and facilitates efficient analysis of both data types. Applications of dPeak to deeply sequenced E. coli PET and SET ChIP-Seq data establish significantly better resolution of PET compared to SET sequencing.
| Since its introduction, chromatin immunoprecipitation followed by high throughput sequencing (ChIP-Seq) has revolutionized the study of gene regulation. ChIP-Seq is currently the state-of-the-art method for studying protein-DNA interactions genome-wide and is widely used [1]–[5]. ChIP-Seq experiments capture millions of DNA fragments ( in length) that the protein under study interacts with using random fragmentation of DNA and a protein-specific antibody. Then, high throughput sequencing of a small region () at the end or both ends of each fragment generates millions of reads or tags. Sequencing one end and both ends are referred to as single-end tag (SET) and paired-end tag (PET) technologies, respectively (Figure 1A). Standard preprocessing of these data involves mapping reads to a reference genome and retaining the uniquely mapping ones [6], [7]. In PET data, start and end positions of each DNA fragment can be obtained by connecting positions of paired reads [8]. In contrast, the location of only the end of each DNA fragment is known in SET data. The usual practice for SET data is to either extend each read to its direction by the average library size which is a parameter set in the experimental procedure [7] or shift the end position of each read by an estimate of the library size [9]. Then, genomic regions with large numbers of clustered aligned reads are identified as binding sites using one or more of the many available statistical approaches [6], [7], [9]–[11] (the first step in Figure 1C).
Currently, the SET assay dominates all the ChIP-Seq experiments despite the fact that PET has several obvious, albeit less studied, advantages over SET. In PET data, paired reads from both ends of each DNA fragment can reduce the alignment ambiguity, increase precision in assigning the fragment locations, and improve mapping rates. This is especially advantageous for studying regulatory roles of repetitive regions of genomes [12], [13]. Although many eukaryotic genomes are rich in repetitive elements, PET technology has not been extensively used with eukaryotic genomes [8], [14]. One of the main reasons for this is that ChIP-Seq data is information rich even when the repetitive regions are not profiled [15] and that the PET assay costs times more than the SET assay. Put differently, given a fixed cost, PET sequencing results in a lower sequencing depth compared to SET sequencing.
In contrast to eukaryotic genomes, prokaryotic genomes are highly mappable, e.g., of the Escherichia coli (E. coli) genome is mappable with reads. This decreases the higher mapping rate appeal of the PET assay for these genomes. In this paper, we systematically investigate advantages of the PET assay from a new perspective and demonstrate both experimentally and computationally that it significantly improves the resolution of protein binding site identification. Improving resolution in identifying protein-DNA interaction sites is a critical issue in the study of prokaryotic genomes because prokaryotic transcription factors have closely spaced binding sites, some of which are only to apart from each other [16]–[19]. These closely spaced binding sites are considered to be multiple “switches” that differentially regulate gene expression under diverse growth conditions [17]. Therefore, identification and differentiation of closely spaced binding sites are invaluable for elucidating the transcriptional networks of prokaryotic genomes.
Although many methods have been proposed to identify peaks from ChIP-Seq data (reviewed in [20]), such as MACS [9], CisGenome [6], and MOSAiCS [10], these approaches reveal protein binding sites only in low resolution, i.e., at an interval of hundreds to thousands of base pairs. Furthermore, they report only one “mode” or “predicted binding location” per peak. More recently, deconvolution algorithms such as CSDeconv [21], GPS [22] (recently improved as GEM [23]), and PICS [11] have been proposed to identify binding sites in higher resolution. However, these methods are specific to SET ChIP-Seq data and are not equipped to utilize the main features of PET ChIP-Seq data. Although a relatively recent method named SIPeS [24] is specifically designed for PET data and is shown to perform better than MACS paired-end mode [9], our extensive computational and experimental analysis indicated that this approach is not suited for identifying closely located binding events. To address these limitations, we developed dPeak, a high resolution binding site identification (deconvolution) algorithm that can utilize both PET and SET ChIP-Seq data. The dPeak algorithm implements a probabilistic model that accurately describes the ChIP-Seq data generation process and analytically quantifies the differences in resolution between the PET and SET ChIP-Seq assays. We demonstrate that dPeak outperforms or performs competitively with the available SET-specific methods such as PICS, GPS, and GEM. More importantly, dPeak coupled with PET ChIP-Seq data improves the resolution of binding site identification significantly compared to SET-based analysis with any of the available methods. Generation and analysis of factor PET and SET ChIP-Seq data from E. coli grown under aerobic and anaerobic conditions reveal the power of the dPeak algorithm in identifying closely located binding sites. Our study demonstrates the importance of high resolution binding site identification when studying the same factor under diverse biological conditions. We further support our findings by validating a small subset of our closely located binding site predictions with primer extension experiments.
The factor is responsible for transcription initiation at over 80% of the known promoters in E. coli [25]. combines with RNA polymerase to bind promoter sequences typically containing two consensus elements located at and upstream of the transcription start site [18]; thus a binding site spans about upstream from the transcription start site. Many E. coli genes contain multiple promoters, and much transcriptional regulation by oxygen as well as by other stimuli occurs by selection of one or a subset of the possible promoters in concert with binding of activators and repressors (e.g., ArcA and FNR for regulation by oxygen [17], [19]). Understanding such regulation requires knowledge of precisely which promoters are used in a given condition. Therefore, the highest possible accuracy of ChIP-signal mapping will allow the best determination of promoter binding by -RNA polymerase holoenzyme.
We generated both PET and SET ChIP-Seq data for factor from E. coli grown under aerobic () and anaerobic () conditions in glucose minimal media on the HiSeq2000 and Illumina GA IIx platforms. We used these experimental data for comparisons of PET and SET assays and evaluation of our high resolution binding site detection method dPeak throughout the paper. Figure 1B displays PET and SET ChIP-Seq coverage plots for the promoter region of the cydA gene under the aerobic condition. The height at each position indicates the number of DNA fragments overlapping that position. The cydA promoter contains five known binding sites separated by to [25]. As evidenced in Figure 1B, coverage plots for PET and SET appear almost indistinguishable visually. To further understand the appearance of peaks that multiple binding events in this region would result in, we simulated PET and SET data with parameters matching to those of this region. Figures S1A, B, C in Text S1 display SET and PET coverage plots of this region when it harbors one and three binding events. These plots support that when binding events are in close proximity with distances less than the average library size, they appear as uni-modal peaks regardless of the library preparation protocol (Figure S1C in Text S1). We next evaluated two peak callers, MACS [9] and MOSAiCS [10], both of which are specifically developed for SET data, on our SET and PET experimental datasets (Table S1 in Text S1). Both methods identified broad regions and the median widths of MACS peaks were to times larger than those of the MOSAiCS peaks. Detailed comparison of the MACS and MOSAiCS peaks revealed that each MACS peak on average has to MOSAiCS peaks (Table S2 in Text S1). Next, we evaluated the number of annotated binding events from RegulonDB [25] (http://regulondb.ccg.unam.mx/) in each of the MACS and MOSAiCS peaks and found that MACS peaks, on average, had to annotated binding events whereas MOSAiCS peaks had to . Overall, we did not observe any differences in the peak widths of the PET and SET assays with MOSAiCS whereas MACS peaks from PET data tended to be wider than those of the SET data. These findings indicate that the potential advantages of the PET assay for elucidating closely located binding sites are not simply revealed from visual inspection and by analysis with methods developed specifically for SET data. Hence, deciphering the advantages of PET over SET for high resolution binding site identification warrants a statistical assessment. Next, we developed a generative probabilistic model and an accompanying algorithm, dPeak, that can specifically utilize local read distributions from SET and PET assays. This algorithm enabled unbiased evaluation of the SET and PET assays using our E. coli SET and PET ChIP-Seq data.
dPeak requires data in the form of genomic coordinates of paired reads (for PET) or genomic coordinates of reads and their strands (for SET) obtained from mapping to a reference genome. For computational efficiency, dPeak first identifies candidate regions (i.e., peaks) that contain at least one binding event and considers each candidate region separately for the prediction of number and locations of binding events (the first step of Figure 1C). Either two-sample (using both ChIP and control input samples) or one-sample (only using ChIP sample when a control sample is lacking) analysis can be used to identify candidate regions. For this purpose, we utilize the MOSAiCS algorithm [10] which produced narrower peaks than the MACS algorithm [9] in our ChIP-Seq datasets (Table S1 in Text S1).
In each candidate region, we model read positions as originating from a mixture of multiple binding events and a background component (the third step of Figure 1C). dPeak infers the number of binding events and the read sets corresponding to each binding event within each region. It iterates the following two steps for each candidate region. First, it assigns each read to a binding event or background, based on the positions and strengths of the binding events. Then, the position and strength of each binding event are updated using its assigned reads. In practice, the number of binding events in each candidate region is unknown a priori. Hence, we consider models with different numbers of binding events and choose the optimal number using Bayesian information criterion (BIC) [26]. We constructed generative probabilistic models for binding event components and a background component for each of the PET and SET data by careful exploratory analyses of multiple experimental ChIP-Seq datasets. Diagnostic plots of the fitted models (Figure S3 in Text S1) indicate that the dPeak model fits ChIP-Seq data well.
dPeak has two unique features compared to other peak deconvolution algorithms (Table S3 in Text S1). First, it accommodates both SET and PET data and explicitly utilizes specific features of both types. Second, it incorporates a background component that accommodates reads due to non-specific binding. Consideration of non-specific binding is critical because the degree of non-specific binding becomes more significant as the sequencing depths get larger. An additional unique feature of dPeak is the treatment of unknown library size for SET data. As discussed earlier, to account for unknown library size, each read is either extended to or shifted by an estimate of the library size in most peak calling algorithms [20]. This estimate is often specified by users [7], [10] or estimated from ChIP-Seq data [9], [11]. Currently available algorithms with the exception of PICS use only one extension/shift estimate for all the regions in the genome. However, our exploratory analysis of real ChIP-Seq data and the empirical distribution of the library size from PET data (Figure S2A in Text S1) indicate that using single extension/shift length might be suboptimal for peak calling (data not shown). In order to address this issue, dPeak estimates optimal extension/shift length for each candidate region. Comparison of empirical distribution of the library size from PET data with the estimates of the region-specific extension/shift lengths indicates that dPeak estimation procedure handles the heterogeneity of the peak-specific library sizes well (Figures S2B, C, D in Text S1). This advancement ensures that dPeak is well tuned for deconvolving SET peaks, which then enables an unbiased computational comparison between the SET and PET assays.
We compared dPeak with two competing algorithms, GPS [22] and PICS [11], for analysis of SET ChIP-Seq data. We did not include the CSDeconv algorithm [21] in this comparison because it is computationally several orders of magnitude slower than the algorithms considered here. We utilized the synthetic ChIP-Seq data which was previously used to evaluate deconvolution algorithms [22]. In this synthetic data, binding events were generated by spiking in reads from predicted CTCF binding events at predefined intervals [22] without explicitly implanting binding sequence motifs. Therefore, we also excluded GEM [23], which capitalizes on motif discovery to infer positions of binding events, from this comparison and used additional computational experiments below to perform comparisons with GEM. The synthetic data from [22] consisted of 1,000 joint (i.e., close proximity) binding events, each with two events, and 20,000 single binding events. We assessed performances of algorithms on these two sets separately.
Figure 2A shows the sensitivity of each algorithm at different distances between the joint binding events. Here, sensitivity is the proportion of regions for which both of the two true binding events are correctly identified. dPeak outperforms other methods across all considered distances between the joint binding events and especially for closely located binding events separated by less than the average library size of . When the distance between the joint binding events is about , dPeak is able to identify both binding events in of the regions whereas neither PICS nor GPS can detect both binding events in more than . Further investigation indicates that PICS merges closely spaced binding events into one event too often (Figure S4 in Text S1). We also found that GPS estimates the peak shape incorrectly when ChIP-Seq data harbors many closely located binding events (Figure S5 in Text S1). Furthermore, the sensitivity of GPS also decreases significantly when the distance between joint binding events increases. A closer look at the results reveals that GPS filters out too many predictions for joint binding events.
To ensure that increased sensitivity of dPeak is not a result of increased number of false predictions, we evaluated positive predictive value (fraction of predictions that are correct) of each method. Specifically, we plotted the number of binding events predicted by each algorithm at different distances between the joint binding events in Figure 2B. Since there are two true binding events in each region, two predictions at every distance correspond to perfect positive predictive value. dPeak on average generates more than one prediction and does not over-estimate the number of binding events when the distance between joint events is less than the average library size. This result confirms that the higher sensitivity of dPeak in Figure 2A is not due to increased number of predictions. In contrast, PICS and GPS on average generate only one prediction for closely located binding events, which recapitulates the conclusions from Figure 2A. In summary, dPeak outperforms state-of-the-art deconvolution methods across different distances between joint binding events, especially when the distance between the binding events is less than the average library size.
Next, we evaluated the sensitivity and positive predictive value of the three methods on 20,000 candidate regions with a single binding event using the additional synthetic data from [22] (Table S4 in Text S1). Average number of predictions per region with at least one predicted binding event and the corresponding standard errors are as follows: dPeak (), PICS (), GPS (). Overall, dPeak slightly over-estimates the number of binding events for regions with a single binding event, and hence PICS is slightly better than dPeak in positive predictive value for these regions. However, as revealed by our joint event analysis, this conservative approach of PICS severely under-estimates the number of binding events when multiple events reside closely. In contrast, GPS significantly under-estimates the number of binding events for the regions with a single binding event since it filters out too many predictions and does not result in a prediction for of the regions. In addition, it over-estimates the number of binding events across regions for which it produces at least one prediction. Comparisons in these two scenarios with and without joint binding events indicate that dPeak strikes a good balance between sensitivity and positive predictive value for both cases.
Once we developed dPeak as a high resolution peak detection method for both SET and PET data, we implemented simulation studies to evaluate the PET and SET assays for resolving closely spaced binding events in an unbiased manner. Although SIPeS [24] supports PET ChIP-Seq data, we excluded it from the comparison of PET and SET ChIP-Seq datasets due to its poor performance (Section 16 of Text S1). We generated simulated PET and SET ChIP-Seq data with two closely spaced binding events and evaluated the predictions of these two data types with dPeak (Section 11 of Text S1; Figure S7 in Text S1).
Figure 2C plots the sensitivity of dPeak as a function of distance between the joint binding events and number of reads for both the PET and SET settings. Note that we evaluated sensitivity up to the distance of because we used windows to determine whether a binding event is correctly identified and as a result, results for the distance less than could be misleading. When the distance between the events is at least as large as the average library size (), the sensitivity using PET and SET data are comparable. However, as the distance between joint binding events decreases, the sensitivity using SET data decreases significantly. In contrast, PET ChIP-Seq retains its high sensitivity even for binding events that are located as close as . As the number of reads decreases, sensitivity for both PET and SET data decreases. When there are only DNA fragments (i.e., reads) per binding event, sensitivity for PET data also decreases as the distance between joint binding events decreases. However, even in this case, sensitivity of PET data is still significantly higher than that of SET data with much higher number of reads. Figure 2D displays the number of binding events predicted by dPeak at different distances between joint binding events when reads correspond to each binding event for both PET and SET data and evaluates positive predictive value. Results are similar for higher number of reads (data not shown). With PET ChIP-Seq, dPeak accurately chooses the number of binding events by BIC out of a maximum of five binding events at any distance between the joint binding events. In contrast, SET ChIP-Seq predicts less than two binding events when the distance between the events is less than .
We present additional simulation results in Section 10 of (Figure S6 in Text S1). These simulations reveal that even for cases with single binding events, PET has a slight advantage over SET because it predicts the location of the binding event more accurately. Specifically, PET data always provides higher resolution compared to SET data regardless of the strength of the binding event, which we measure by the number of DNA fragments associated with the event. For example, for a binding event with DNA fragments, the average distance between the predicted and true binding events is with a standard deviation of in the PET data whereas it is with a standard deviation of in the SET data. Note that although this simulation procedure is based on the assumptions of dPeak model for PET data, our exploratory analysis and goodness of fit (Figure S3A in Text S1) show that these assumptions hold well in the real PET ChIP-Seq data and therefore, these results have significant practical implications for real ChIP-Seq data.
Lower sensitivity of the SET compared to PET data is mainly driven by the loss of information due to unknown library size. We describe this information loss by two concepts named invasion and truncation (Figure 3A). Top diagram of Figure 3A depicts two closely spaced binding events and a DNA fragment that is informative for the first binding event (in red) in the PET data. Invasion refers to over-estimation of the library size and extension of the read to a length longer than the true one. Equivalently, in the shifting procedure, this corresponds to shifting the read more than necessary. As a result, the read extended to the estimated library size covers both of the closely spaced binding events in the SET data and becomes uninformative or less informative for the binding event it corresponds to. Bottom diagram of Figure 3A also depicts two closely spaced binding events and illustrates truncation which we define as under-estimation of the library size. In this case, the displayed DNA fragment is long and spans both binding events (in red). Therefore, it contributes to estimation of both binding events in the PET data. In contrast, the read extended to estimated library size only covers the first binding event in the SET data and, as a result, its contribution to the first binding event is overestimated whereas its contribution to the second binding event is underestimated. We evaluated the frequency by which fragments with invasion and truncation arise in SET data with a simulation study. Our results (Table S5 in Text S1) indicate that as high as and of the fragments for a typical peak region can be subject to invasion and truncation with the SET assay.
Figures 3B, C display the probabilities of invasion and truncation, respectively, of a DNA fragment as a function of the distance between binding events and the variance of the library size. The analytical calculations are based on the dPeak generative model (Section 12 of Text S1). Probabilities of invasion and truncation are higher for closely spaced binding events, especially when the library size is shorter than the estimated library size ( in this case). In Figure 3B, the probability of invasion decreases for very closely spaced binding events, i.e., when the distance between two binding events is less than . As the distance between the binding events decreases, most DNA fragments cover both binding events and the configuration in the first diagram of Figure 3A is unlikely to occur. Hence, there is already insufficient information to predict two binding events even in PET data and relative loss of information (i.e., invasion) in SET data is insignificant. These concepts describe how information on binding events can be lost or distorted by the incorrect estimation of the library size in the SET data. Analytical calculations based on the dPeak generative model show that invasion and truncation influence closely located binding events the most, especially when the library size is not tightly controlled, i.e., exhibit large variation (Figures 3B, C).
We compared the performance of PET and SET sequencing for factor under the aerobic condition by generating a ‘quasi-SET data’ by randomly sampling one of the two ends of each paired reads in PET data and comparing binding events identified from both sets. In order to match number of reads with SET data for fair comparison, only the half number of paired reads was used to construct PET data. Comparison with the quasi-SET data controlled for the differences in the sequencing depths of the original PET and SET samples in addition to the biological variation of the replicates. We then evaluated the dPeak predictions from the PET and SET analyses using the factor binding site annotations in the RegulonDB database as a gold standard. Because a significant number of promoter regions lack RegulonDB annotations, we evaluated the sensitivity based on the regions that contain at least one annotated binding site. This corresponds to binding sites in candidate regions that MOSAiCS identified. Of these regions, harbor only a single annotated binding event. For the regions with more than one annotated binding event, the average distance between binding events is . dPeak analysis of the SET data identifies only of the annotated binding events. In contrast, analysis of PET data with dPeak detects of the annotated binding sites. Figure 4A displays average sensitivity as a function of the average distance between annotated binding events for the regions with at least two RegulonDB annotations. A linear line is superimposed to capture the trend for both data types. Notably, the lower sensitivity of SET compared to PET is mainly due to closely located binding events.
We also compared prediction accuracies of the PET and SET assays for the regions that harbor a single annotated binding event. Figure 4B displays resolutions, which we define as the minimum of distances between predicted and annotated positions of binding events, achieved by the PET and SET assays. Median resolutions are (IQR = ) and (IQR = ) for PET and SET, respectively. This result indicates that positions of binding events can be more accurately predicted with the PET assay compared to SET even for regions with a single binding event.
To further examine the accuracy of the dPeak predictions, primer extension analysis was performed to map the transcription start site for eight genes (Figures S10–S13 in Text S1; Table S7 in Text S1). dPeak analysis of the PET ChIP-Seq data predicts two closely spaced binding sites in the upstream of each of these eight genes with the distance between predictions ranging to . Seven of these predictions are not annotated in RegulonDB and thus represent potential novel transcription start sites. A transcription start site was detected within of () of these binding site predictions (Figure 5A and Table 1), further supporting the accuracy of the dPeak PET predictions.
We treated these validated sites as a gold standard and evaluated the performance of each deconvolution algorithm for these regions. Figure 5B depicts that dPeak with PET ChIP-Seq data attains significantly higher resolution compared to SET-based analysis regardless of the deconvolution algorithm used (p-values of paired t-tests between dPeak using PET data and each of the other methods using SET data are ). dPeak with SET ChIP-Seq data has a resolution comparable to or better than those of the competing algorithms. GPS is not included in this plot because it provides significantly worse resolution compared to other methods (Figure S9C in Text S1). Genome-wide comparisons using the RegulonDB transcription start site annotations as a gold standard also lead to a similar conclusion, supporting the notion that PET-analysis with dPeak provides the best resolution (Figures S9A, B in Text S1).
Figures 4C and 4D display two representative peak regions from these analyses. Figure 4C illustrates two binding events in the promoter regions of sibD and sibE genes separated by . In this case, two peaks are easily distinguishable just by visual inspection and the predictions using both PET and SET data are comparably accurate. Note that although these two binding events are visually distinguishable, standard applications of MACS and MOSAiCS identify this region as a single peak. Widths of MOSAiCS and MACS peaks for this region are and , respectively. MACS identifies the position of the right binding event as the “summit” of this region (position ). Figure 4D displays the promoter region of yejG gene, where the distance between the two experimentally validated binding events is only . In this case, dPeak application to PET data correctly predicts the number of binding events as two and identifies the locations of these events within of the validated sites. In contrast, all of the SET-based analyses with the deconvolution algorithms (PICS, GPS, GEM) incorrectly predict one binding event located in the middle of the two experimentally validated binding sites.
High resolution identification of binding sites is especially important for differential occupancy analysis where a protein of interest is profiled under different conditions. Given the high agreement between the dPeak algorithm and experimentally validated transcription start sites at a subset of promoter regions, we set out to identify differential promoter usage between the aerobic and anaerobic growth conditions by profiling the E. coli factor. Results from the dPeak analysis of the aerobic and anaerobic PET data are summarized in Figure 5C both in the region (i.e., peak) and binding event levels. We identified peaks and dPeak binding events that were common between the and conditions. Interestingly, only peaks were unique to the condition but dPeak analysis identified -specific binding events. Similarly, we identified peaks unique to the condition while dPeak analysis resulted in -specific binding events. We used the SET ChIP-Seq data from additional biological replicates under both conditions as independent validation of the results. This independent validation using SET data identified of the binding events identified by dPeak using PET ChIP-Seq data ( of the common events, of the -specific binding events and of the -specific binding events). Table S8 in Text S1 further summarizes these results by cross-tabulating the number of predicted binding events in each peak across the two conditions. It illustrates that there are indeed many peaks with at least one binding event in each condition and different number of binding events across the two conditions. Figure S14 in displays an example of closely located binding sites that are differentially occupied between aerobic and anaerobic conditions in PET ChIP-Seq data. These results suggest that dPeak analysis identified many unique binding events that could not be differentiated in the peak-level analysis.
High resolution identification of binding sites with ChIP-Seq has profound effects for studying protein-DNA interactions in prokaryotic genomes and differential occupancy. We evaluated PET and SET ChIP-Seq assays and illustrated that PET has considerably more power for deciphering locations of closely spaced binding events. Our data-driven computational experiments indicate that when the distance between binding events gets smaller than the average library size, SET analysis have notably less power than the PET analysis. Furthermore, PET provides better resolution than SET even when a region harbors a single binding event. We developed and evaluated the dPeak algorithm, a model-based approach to identify protein binding sites in high resolution, with data-driven computational experiments and experimental validation. dPeak is currently the only algorithm that can utilize both PET and SET ChIP-Seq data and can accommodate high levels of non-specific binding apparent in deeply sequenced ChIP samples (Table S3 in Text S1). Our data-driven computational experiments and computational analysis of experimentally validated binding sites indicate that it significantly outperforms the currently available PET ChIP-Seq peak finder SIPeS [24]. Application of dPeak to E. coli ChIP-Seq data under aerobic and anaerobic conditions revealed that although many peaks identified by standard application of popular peak finders might appear as common between the two conditions, a considerable percentage of these may harbor condition-specific binding events. The high-resolution binding sites identified by dPeak could be combined with start-site mapping or consensus-sequence identification to assign transcriptional orientation to the binding sites.
The advantages of using the dPeak algorithm are not limited to the study of prokaryotic genomes. Applications in eukaryotic genomes include identification of the exact locations of binding motifs when multiple closely located consensus sequences reside in a peak region, studies of cis regulatory modules (CRM), and refining consensus sequences. Figure S16 in Text S1 displays an example application of dPeak for differentiating among multiple closely located GATA1 binding sites with consensus WGATAR within a ChIP-Seq peak region critical for erythroid differentiation in mouse embryonic stem cells (data from [27]). CRM studies investigate relationships between spatial configurations of binding sites of multiple transcription factors and gene expression. Relative orders, positions, and distances of binding sites of multiple factors and their relative strengths are key factors in CRM studies [28]. Because dPeak facilitates identification of binding sites of transcription factors in high resolution from ChIP-Seq data, it can enable construction of complex interaction networks among diverse factors across multiple growth conditions.
We evaluated the performance of dPeak on eukaryotic genome ChIP-Seq data that GPS and PICS were optimized for. Figure S17 in Text S1 shows the performance comparison results for transcription factor GABPA profiled in GM12878 cell line from the ENCODE database. It indicates that dPeak performs comparable to or outperforms GPS and PICS. In the case of sequence-specific factors with well-conserved motifs such as the GABPA factor, we observed that dPeak prediction can be further improved in a straightforward way by incorporating sequence information. Figure S17 in Text S1 illustrates that dPeak with incorporated sequence information performs comparable to GEM and identifies the GABPA binding sites with high accuracy.
Recently, ChIP-exo assay [29], a modified ChIP-Seq protocol using exonuclease, has been proposed as a way of experimentally attaining higher resolution in protein binding site identification. Because the ChIP-exo protocol is new and relatively laborious, there are not yet many publicly available ChIP-exo datasets. We utilized ChIP-exo of CTCF factor in human HeLa-S3 cell line [29] and compared their binding event predictions with dPeak predictions on SET ChIP-Seq data of CTCF in the same cell line. Figure S18 in Text S1 illustrates that dPeak using SET ChIP-Seq data provides higher resolution than ChIP-exo data and that dPeak can be readily utilized for ChIP-exo data analysis. Furthermore, it also indicates that dPeak performs comparable to or outperforms currently available methods such as GPS and GEM for both ChIP-exo and SET ChIP-Seq data. Although the real power of the ChIP-exo technique will be revealed as more ChIP-exo datasets are produced and compared with ChIP-Seq datasets, our results with the currently available data suggest that analyzing ChIP-Seq data with powerful deconvolution methods such as dPeak might perform as well as ChIP-exo.
We implemented dPeak as an R package named dPeak. dPeak utilizes the fast estimation algorithm we developed and parallel computing. Analysis of the data (∼1,000 candidate regions, each with ∼2,300 reads on average) using our current sub-optimal implementation of dPeak takes about minutes using CPUs () when up to binding events are allowed in each candidate region, while it takes about minutes to run PICS and GPS (also using 20 CPUs). Similarly, analysis of human ENCODE POL2-H1ESC data (∼14,000 candidate regions, each with reads on average) takes about minutes for dPeak, while it takes and minutes for GPS and PICS, respectively. dPeak is currently available at http://www.stat.wisc.edu/ ~chungdon/dpeak/ and will be contributed to public repositories such as Bioconductor [30] and Galaxy Tool Shed [31] upon publication.
All strains were grown in MOPS minimal medium supplemented with glucose [32] at and sparged with a gas mix of and (anaerobic) or , , and (aerobic). Cells were harvested during mid-log growth ( of using a Perkin Elmer Lambda Spectrophotometer). WT E. coli K-12 MG1655 (, , ) was used for the experiments (Kiley lab stock).
ChIP assays were performed as previously described [33], except that the glycine, the formaldehyde, and the sodium phosphate mix were sparged with argon gas for minutes before use to maintain anaerobic conditions when required. Samples were immunoprecipitated using of RNA Polymerase antibody from NeoClone (W0004).
For ChIP-Seq experiments, of immunoprecipitated and purified DNA fragments from the aerobic and anaerobic samples (one biological sample for both aerobic and anaerobic growth conditions), along with of input control (two biological replicates for anaerobic Input and one biological sample for aerobic Input), were submitted to the University of Wisconsin-Madison DNA Sequencing Facility for ChIP-Seq library preparation. Samples were sheared to during the IP process to facilitate library preparation. All libraries were generated using reagents from the Illumina Paired End Sample Preparation Kit (Illumina) and the Illumina protocol “Preparing Samples for ChIP Sequencing of DNA” (Illumina part # 11257047 RevA) as per the manufacturer's instructions, except products of the ligation reaction were purified by gel electrophoresis using SizeSelect agarose gels (Invitrogen) targeting fragments. After library construction and amplification, quality and quantity were assessed using an Agilent DNA 1000 series chip assay (Agilent) and QuantIT PicoGreen dsDNA Kit (Invitrogen), respectively, and libraries were standardized to . For PET ChIP-Seq data, cluster generation was performed using an Illumina cBot Paired End Cluster Generation Kit (v3). Paired reads, run was performed for each end, using v3 SBS reagents and CASAVA (the Illumina pipeline) v 1.8.2, on the HiSeq2000. For SET ChIP-Seq data, cluster generation was performed using an Illumina cBot Single Read Cluster Generation Kit (v4) and placed on the Illumina cBot. A single read, run was performed, using standard SBS kits (v4) and SCS 2.6 on an Illumina Genome Analyzer IIx. Base calling was performed using the standard Illumina Pipeline version 1.6. Sequence reads were aligned to the published E. coli K-12 MG1655 genome (U00096.2) using the software packages SOAP [34] and ELAND (within the Illumina Genome Analyzer Pipeline Software), allowing at most two mismatches. PET experiments yielded million (M) and mappable paired 36mer reads and SET yielded and mappable 32mer reads for aerobic and anaerobic conditions, respectively. Control input experiments, generated with SET sequencing, resulted in and mappable 32mer reads for the aerobic and anaerobic conditions, respectively. Raw and aligned data files are available at ftp://ftp.cs.wisc.edu/pub/users/keles/dPeak and are being processed by GEO for accession number assignment.
For PET data, if a DNA fragment (paired reads) belongs to -th binding event, we model its leftmost position conditional on its length as Uniform distribution between and , where is the position of -th binding event. Lengths of DNA fragments, , are modeled using the empirical distribution obtained from actual PET data. For SET data, if a read belongs to -th binding event, we model its end position conditional on its strand as Normal distribution. Specifically, if a read is in the forward strand, its end position is modeled as Normal distribution with mean and variance . end positions for reverse strand reads are modeled similarly with Normal distribution with mean and variance . Parameters and are common to all binding event components in each candidate region. Strands of reads are modeled as Bernoulli distribution. Background reads are assumed to be uniformly distributed over the candidate region that they belong to. Parameters are estimated via the Expectation-Maximization (EM) algorithm [35]. Additional details on the dPeak model and the estimation algorithm for the PET and SET settings are available in Sections 2 and 3 of Text S1.
We compared the sensitivity and the number of predictions of dPeak with those of PICS [11], GPS [22], and GEM [23]. Sensitivity is the proportion of regions for which both of the two true binding events are correctly identified. A binding event is considered as ‘identified’ if the distance between the actual binding event and the predicted position is less than . Note that we chose a more stringent criteria than the used by GPS for defining true positives because is not high enough resolution for prokaryotic genomes. For the PICS algorithm, we used the R package PICS version 1.10, which is available from Bioconductor (http://www.bioconductor.org/packages/2.10/bioc/html/PICS.html). For the GPS algorithm, we used its Java implementation version 1.1 from http://cgs.csail.mit.edu/gps/. In the performance comparisons using ChIP-Seq data, we also incorporated GEM, a recently modified and extended version of GPS, which incorporates genome sequence of the peaks to improve binding event identification. For the GEM algorithm, we used its Java implementation version 0.9 from http://cgs.csail.mit.edu/gem/. We downloaded the synthetic data used for the method comparisons from http://cgs.csail.mit.edu/gps/ and its description is provided in Supplementary information of the GPS paper [22]. This synthetic data consists of “chrA” with 1,000 regions that harbor two closely spaced binding events and “chrB” to “chrK” with a total of 20,000 regions with a single binding event. We evaluated performances of the methods on joint and single binding event regions separately so that we could assess sensitivity and specificity for each of these cases. Candidate regions for dPeak were identified using the conditional binomial test [6] with a false discovery rate of by applying the Benjamini-Hochberg correction [36]. These regions were also explicitly provided to the GPS and GEM algorithms as candidate regions. Candidate regions for PICS were identified using the function segmentReads() in the PICS R package (default parameters). Default tuning parameters were used during model fitting for all the methods.
We considered distances between binding sites ranging from to which characterize the typical binding event spacing in E. coli. We generated and assigned DNA fragments to each of two binding events as follows. For each DNA fragment, we drew the length () from the distribution of library size, , estimated empirically from the actual PET ChIP-Seq data and group index () from multinomial distribution with parameters (, ). Then, for given a library size and group index (), leftmost position of the paired reads () was generated from Uniform distribution between and , where is the position of -th binding event. Rightmost position was assigned as . SET data was generated by randomly sampling one of two ends from each of these paired reads. For the SET analysis, average library size was assumed to be . Then, only half of the total number of paired reads was used to construct PET data, in order to match number of reads with SET data for fair comparison. In addition, we randomly assigned DNA fragments to arbitrary positions within the candidate region to generate non-specific binding (background) reads. The sensitivity and the number of predictions were summarized over simulated datasets generated by this procedure. A binding event was considered as ‘identified’ if the distance between the binding event and the predicted position is less than . We repeated these PET versus SET analyses by comparing all the PET data with SET data constructed from selecting one of two ends of each read pair and obtained little or no change in the results (data not shown).
We identified candidate regions, i.e., peaks with at least one binding event, using the MOSAiCS algorithm [10] (two-sample analysis with a false discovery rate of ). In each candidate region, we fitted the dPeak model, which is a mixture of binding event components and one background component (Figure 1C). In the current analysis, up to five binding event components () were considered. The optimal number of binding events was chosen with BIC for each candidate region. We utilized top of the predicted binding events from each condition for the comparison between the aerobic and anaerobic conditions. Overall conclusions remained the same when the full set of predicted binding events are considered.
Total RNA was isolated as previously described [37]. Oligonucleotide primers (Table S7 in Text S1) were labeled at the end using []ATP () and T4 polynucleotide kinase (Promega) followed by purification with a G25 Sephadex Quick Spin Column (GE). Labeled primer () was annealed with total RNA in and extended with avian myeloblastosis virus reverse transcriptase (Promega) as described by the manufacturer, except that actinomycin D was present at [38]. Primer extension experiments were implemented for spr ( RNA), dcuA ( RNA), serC ( RNA), aroL ( and RNA for and , respectively), yejG ( RNA), hybO ( RNA), ybgI ( RNA), and ptsG ( RNA). A dideoxy sequencing ladder was electrophoresed in parallel with the primer extension products on a 8% () polyacrylamide gel containing urea. In cases where the transcription start site could be assigned to one of two nucleotides, preference was given to the purine nucleotide.
The dPeak algorithm is implemented as an R package named dpeak and is freely available from http://www.stat.wisc.edu/~chungdon/dpeak/. We will commit dpeak to Bioconductor (http://www.bioconductor.org) and Galaxy Tool Shed (http://toolshed.g2.bx.psu.edu) upon publication.
|
10.1371/journal.pbio.1000288 | Role of Plastid Protein Phosphatase TAP38 in LHCII Dephosphorylation and Thylakoid Electron Flow | Short-term changes in illumination elicit alterations in thylakoid protein phosphorylation and reorganization of the photosynthetic machinery. Phosphorylation of LHCII, the light-harvesting complex of photosystem II, facilitates its relocation to photosystem I and permits excitation energy redistribution between the photosystems (state transitions). The protein kinase STN7 is required for LHCII phosphorylation and state transitions in the flowering plant Arabidopsis thaliana. LHCII phosphorylation is reversible, but extensive efforts to identify the protein phosphatase(s) that dephosphorylate LHCII have been unsuccessful. Here, we show that the thylakoid-associated phosphatase TAP38 is required for LHCII dephosphorylation and for the transition from state 2 to state 1 in A. thaliana. In tap38 mutants, thylakoid electron flow is enhanced, resulting in more rapid growth under constant low-light regimes. TAP38 gene overexpression markedly decreases LHCII phosphorylation and inhibits state 1→2 transition, thus mimicking the stn7 phenotype. Furthermore, the recombinant TAP38 protein is able, in an in vitro assay, to directly dephosphorylate LHCII. The dependence of LHCII dephosphorylation upon TAP38 dosage, together with the in vitro TAP38-mediated dephosphorylation of LHCII, suggests that TAP38 directly acts on LHCII. Although reversible phosphorylation of LHCII and state transitions are crucial for plant fitness under natural light conditions, LHCII hyperphosphorylation associated with an arrest of photosynthesis in state 2 due to inactivation of TAP38 improves photosynthetic performance and plant growth under state 2-favoring light conditions.
| Plants are able to adapt photosynthesis to changes in light levels by adjusting the activities of their two photosystems, the structures responsible for light energy capture. During a process called state transitions, a part of the photosynthetic complex responsible for light harvesting (the photosynthetic antennae) becomes reversibly phosphorylated and migrates between the photosystems to redistribute light-derived energy. The protein kinase responsible for phosphorylating photosynthetic antenna proteins was identified recently. However, despite extensive biochemical efforts to isolate the enzyme that catalyzes the corresponding dephosphorylation reaction, the identity of this protein phosphatase has remained unknown. In this study, we identified and characterized the thylakoid-associated phosphatase TAP38. We first demonstrate by spectroscopic measurements that the redistribution of excitation energy between photosystems that are characteristic of state transitions do not take place in plants without a functional TAP38 protein. We then show that the phosphorylation of photosynthetic antenna proteins is markedly increased in plants without TAP38, but decreased in plants that express more TAP38 protein than wild-type plants. This, together with the observation that addition of recombinant TAP38 decreases the level of antenna protein phosphorylation in an in vitro assay, suggests that TAP38 directly acts on the photosynthetic antenna proteins as the critical phosphatase regulating state transitions. Moreover, in plants without TAP38, photosynthetic electron flow is enhanced, resulting in more rapid growth under constant low-light regimes, thus providing the first instance of a mutant plant with improved photosynthesis.
| Owing to their sessile life style, plants have to cope with environmental changes in their habitats, such as fluctuations in the incident light. Changes in light quantity or quality (i.e., spectral composition) result in imbalanced excitation of the two photosystems and decrease the efficiency of the photosynthetic light reactions. Plants can counteract such excitation imbalances within minutes by a mechanism called state transitions, which depends on the reversible association of the mobile pool of major light-harvesting (LHCII) proteins with photosystem II (state 1) or photosystem I (PSI) (state 2) (reviewed in [1]–[5]). In detail, the accumulation of phosphorylated LHCII (pLHCII), stimulated in low white light, or by light of wavelengths specifically exciting PSII (red light), causes association of pLHCII with PSI (state 2), thus directing additional excitation energy to PSI. Conditions like darkness or light of wavelengths specifically exciting PSI (far-red light), as well as high intensities of white light, stimulate pLHCII dephosphorylation and its migration to PSII (state 1), thus redirecting excitation energy to PSII.
LHCII phosphorylation and state transitions have been extensively studied in the green alga Chlamydomonas reinhardtii and the flowering plant Arabidopsis thaliana [2],[4]–[6]. In C. reinhardtii, the impact of state transitions on interphotosystem energy balancing and on promoting cyclic electron flow is well established [2],[5]. In flowering plants, however, the physiological significance of state transitions is less clear, because their mobile LHCII pools are significantly smaller than those in green algae [7],[8]. Thus, A. thaliana mutant plants impaired in state transitions are only marginally affected in their development and fitness [9]–[11], even under fluctuating light or field conditions [12],[13]. However, when Arabidopsis state transition mutants are perturbed in linear electron flow, effects on plant performance and growth rate become evident [14], indicating that also in flowering plants, state transitions are physiologically relevant.
The protein kinase responsible for phosphorylating LHCII is membrane bound and activated upon reduction of the cytochrome b6/f (Cyt b6/f) complex via the plastoquinone (PQ) pool under state 2-promoting light conditions (low white light or red light) [15],[16]. PQ oxidizing conditions induced by state 1-promoting light conditions (dark or far-red light) inactivate the LHCII kinase and result in association of pLHCII with PSII (state 1, reviewed in [4],[5]). The LHCII kinase activity, however, is also inactivated under high white light conditions, when the stromal reduction state is very high. In vitro and, more recently, in vivo studies suggest that suppression of LHCII kinase activity might be mediated by reduced thioredoxin [17],[18]. In C. reinhardtii and A. thaliana, the orthologous thylakoid protein kinases Stt7 and STN7, respectively, are required for LHCII phosphorylation and state transitions [12],[19]. Coimmunoprecipitation assays showed that the Stt7 kinase interacts with Cyt b6/f, PSI, and LHCII [17], suggesting that Stt7 (and STN7 in Arabidopsis) directly phosphorylates LHCII, rather than being part of a Stt7/STN7-dependent phosphorylation cascade.
Under PQ oxidizing conditions when the LHCII kinase becomes inactivated, pLHCII is dephosphorylated by the action of an as-yet unknown protein phosphatase, thus allowing the association of the mobile fraction of LHCII with PSII (state 1) [5],[7]. For many years, attempts were undertaken to elucidate the characteristics and to identify the LHCII protein phosphatase(s). By means of biochemical approaches [20]–[22], it was shown that protein phosphatases of different families must be involved in the reversible phosphorylation of thylakoid phosphoproteins. A PP2A-like phosphatase was postulated to be responsible for the desphosphorylation of the PSII core proteins [23], whereas the LHCII phosphatase activity was shown to be dependent on the presence of divalent cations and not to be inhibited by microcystin and okadaic acid [21],[22]. These findings strongly suggested an involvement of a PP2C-type phosphatase in pLHCII dephosphorylation [24].
Here, we show that the thylakoid protein phosphatase TAP38 is required for pLHCII dephosphorylation and state transitions. In plants with markedly reduced TAP38 levels, hyperphosphorylation of LHCII is associated with enhanced thylakoid electron flow, resulting in more rapid growth under constant low-light regimes. Together with the results of an in vitro dephosphorylation assay, our data indicate that TAP38 dephosphorylates pLHCII directly.
To identify the LHCII phosphatase, we systematically isolated loss-of-function mutants for known chloroplast protein phosphatases and assessed their capacity to dephosphorylate pLHCII (see below for details). However, none of the nine protein phosphatases At3g52180 (DSP4/SEX4), At4g21210 (AtRP1), At1g07160, At3g30020, At4g33500, At1g67820, At2g30170, At3g10940, or At4g03415, demonstrated to reside in the chloroplast [25]–[29], qualified as the LHCII phosphatase (unpublished data). Next, we extended our search to protein phosphatases tentatively identified as chloroplast proteins by proteomic analyses in A. thaliana [30],[31]. Of those, the serine/threonine protein phosphatase At4g27800 turned out to be the most promising candidate.
Proteins with high homology to At4g27800 exist in mosses and higher plants, but not in algae or prokaryotes. Furthermore, At4g27800 and its homologs share a predicted N-terminal chloroplast transit peptide (cTP), a putative transmembrane domain at their very C-terminus and a protein phosphatase 2C signature (Figure 1). For A. thaliana, three At4g27800 mRNAs are predicted (Figure S1A). To verify their existence and to distinguish between the different splice forms, reverse-transcriptase PCR analyses were performed. Only At4g27800.1, and much less At4g27800.2, were detectable in leaves, whereas for the At4g27800.3 splice variant, no signal could be obtained (Figure S1B).
In protoplasts transfected with At4g27800.1 fused to the coding sequence for the red fluorescent protein (RFP) [32], the fusion protein localized to chloroplasts (Figure 2A). Chloroplast import assays with the radioactively labeled At4g27800.1 protein confirmed the uptake into the chloroplast with concomitant removal of its cTP. Mature At4g27800.1 has a molecular weight of ∼38 kDa (Figure 2B). Immunoblot analysis using a specific antibody raised against the mature At4g27800.1 protein (Figure 2C) detected the protein in thylakoid preparations but not in stromal fractions. It is noteworthy, that the putative translation products At4g27800.2 and At4g27800.3 (∼32 kDa) were undetectable (Figure 2C). At4g27800.1 is therefore the major isoform in leaves, and was renamed TAP38 (Thylakoid-Associated Phosphatase of 38 kDa).
Two tap38 insertion mutants, tap38-1 (SAIL_514_C03) [33] and tap38-2 (SALK_025713) [34], were obtained from T-DNA insertion collections (Figure S1A). In tap38-1 and tap38-2 plants, amounts of TAP38 transcripts were severely reduced, to 10% and 13% of WT levels, respectively (Figure 3A). Conversely, in transgenic lines carrying the TAP38 coding sequence under control of the 35S promoter of Cauliflower Mosaic Virus (oeTAP38), levels of TAP38 mRNA were much higher than in wild type (WT) (Figure 3A). TAP38 protein concentrations reflected the abundance of TAP38 transcripts: tap38-1 and tap38-2 thylakoids had <5% and ∼10% of WT levels, respectively, whereas the oeTAP38 plants displayed >20-fold overexpression on the protein level (Figure 3B). TAP38 protein levels were also determined under light conditions relevant for state transitions (see Materials and Methods). In WT plants, TAP38 was constitutively expressed at similar levels under all light conditions applied (Figure 3C).
To determine whether TAP38 is involved in state transitions, chlorophyll fluorescence was measured in WT, tap38, and oeTAP38 leaves (Figure 4A). Plants were exposed to light conditions that stimulate either state 2 (red light) or state 1 (far-red light) [35],[36], and the corresponding maximum fluorescence in state 2 (FM2) and in state 1 (FM1) values were determined. Because the light intensity chosen to induce state transitions did not elicit photoinhibition (as monitored by measurements of the maximum quantum yield [FV/FM]), changes in FM, the maximum fluorescence, could be attributed to state transitions alone. This allowed us to calculate the degree of quenching of chlorophyll fluorescence due to state transitions (qT) [36]. In the tap38 mutants, qT was markedly decreased (tap38-1, 0.01±0.003; tap38-2, 0.03±0.001; WT, 0.10±0.001). In tap38-1 plants complemented with the TAP38 genomic sequence (including its native promoter), qT values were normal, confirming that state transitions require TAP38. Interestingly, oeTAP38 plants exhibited qT values of about 0.01±0.001, indicating that both absence and excess of TAP38 interfere with the ability to undergo reversible state transitions.
To determine the antenna sizes of PSII and PSI, 77K fluorescence emission spectra were measured under state 1 (exposure to far-red light) and state 2 (low light) conditions as described [11],[12],[37] (Figure 4B). The spectra were normalized at 685 nm, the peak of PSII fluorescence. In WT, the transition from state 1 to state 2 was accompanied by a marked increase in relative PSI fluorescence at 730 nm, reflecting the redistribution of excitation energy from PSII to PSI. In contrast, in tap38 leaves, the PSI fluorescence peak was relatively high even under state 1-promoting conditions, implying that the mutants were blocked in state 2—i.e., pLHCII should be predominantly attached to PSI. Additionally, under state 2-promoting light conditions, the PSI antenna size (expressed as F730/F685) was larger in tap38 mutants than in WT (tap38-1, 1.47; tap38-2, 1.45; WT, 1.38; see also Table S1), arguing in favor of the idea that in tap38 plants, a larger fraction of the mobile pool of LHCII can attach to PSI. On the contrary, in oeTAP38 plants, the relative fluorescence of PSI hardly increased at all under conditions expected to induce the state 1→state 2 shift (Figure 4B; Table S1). This behavior resembles that of stn7 mutants, which are blocked in state 1, i.e., with LHCII permanently attached to PSII [12].
It is generally accepted that state transitions require reversible phosphorylation of LHCII [2],[4],[5]. Therefore, the phosphorylation state of LHCII was monitored under light conditions that favor state 1 (dark or far-red light treatment) or state 2 (low light). Plants with abnormal levels of TAP38, and WT plants were dark adapted for 16 h (state 1), then exposed to low light (80 µmol m−2 s−1, 8 h) (state 2), and then to far-red light (4.5 µmol m−2 s−1, 740 nm) for up to 120 min to induce a return to state 1. Thylakoid proteins were isolated after each treatment, fractionated by sodium dodecyl sulfate (SDS)-PAGE, and analyzed with a phosphothreonine-specific antibody (Figure 5, left panels). WT plants showed the expected increase in pLHCII during the transition from state 1 (dark [D]) to state 2 (low light [LL]), followed by a progressive decrease in pLHCII upon exposure to far-red light (FR). In tap38 mutants, levels of pLHCII were aberrantly high at all time points, whereas the oeTAP38 plants again mimicked the stn7 phenotype [9],[12], displaying constitutively reduced levels of pLHCII.
To directly visualize how alterations in LHCII phosphorylation in lines lacking or overexpressing TAP38 affect the distribution of the mobile LHCII fraction between the two photosystems, we subjected thylakoid protein complexes of plants adapted to state 1 (dark and far-red light treatments) or state 2 (low-light treatment) to nondenaturing Blue-native (BN) PAGE [38] (Figure 5, right panels). In this assay, a pigment–protein complex of about 670 kDa, which represents pLHCII associated with the PSI-LHCI complex [14],[38],[39], can be visualized. Whereas in WT thylakoids, the 670-kDa complex was only observable under state 2 conditions (Figure 5A, right panel), as previously reported [14],[39]; the constitutive phosphorylation of LHCII in the tap38 mutants was associated with the presence of a prominent band for the 670-kDa complex under all light conditions (Figure 5B, right panel). The formation of the 670-kDa complex was totally prevented in oeTAP38 plants with a block in state 1 and highly reduced levels of pLHCII (Figure 5C, right panel). Two-dimensional (2D) PA gel fractionation confirmed that the pigment–protein complex consists of PSI and LHCI subunits, together with a portion of pLHCII that associates with PSI upon state 1→state 2 transition in WT plants (Figure 6; [14]). Additionally, quantification of the different PSI complexes on 2D PA gels showed that the number of PSI complexes associated with LHCII was increased in the tap38 mutants (Figures 6B and 6C), supporting the findings obtained from the 77K fluorescence analyses.
An in vitro dephosphorylation assay was established to assess the capability of TAP38 to directly dephosphorylate pLHCII. To this purpose, an N-terminal His-tag fusion of the TAP38 phosphatase was expressed in Escherichia coli and purified (see Materials and Methods). Solubilized thylakoids from tap38-1 mutant plants were then fractionated by sucrose gradient ultracentrifugation, and the protein fraction enriched in pLHCII was isolated. Subsequently, the pLHCII pigment–protein complex was incubated at 30°C for 2 h either in the presence or absence of the recombinant TAP38 phosphatase. At the end of the incubation period, the reaction mixture was fractionated by SDS-PAGE and subjected to immunoblotting using a phosphothreonine-specific antibody (Figure 7). Clearly, the addition of the recombinant TAP38 decreased the level of LHCII phosphorylation by about 50% (relative to the untreated pLHCII sample). In the presence of the phosphatase inhibitor NaF, TAP38 addition did not markedly alter the phosphorylation level of LHCII. Taken together, these findings suggest that TAP38 is able to directly dephosphorylate pLHCII.
When kept under low-light intensities (80 µmol m−2 s−1) that favor state 2, tap38 mutants grew larger than WT plants (Figure 8A), whereas oeTAP38 plants behaved like WT (unpublished data). Detailed growth measurements revealed that the tap38 mutants exhibited a constant growth advantage over WT plants, starting at the cotyledon stage (Figure 8B). Because this difference might be attributable to altered photosynthetic performance, parameters of thylakoid electron flow were measured. The fraction of QA (the primary electron acceptor of PSII) present in the reduced state (1-qP) was lower in tap38-1 (0.06±0.01) and tap38-2 plants (0.07±0.01) than in WT (0.10±0.01), when both genotypes were grown as in Figure 8A and chlorophyll fluorescence was excited with 22 µmol m−2 s−1 actinic red light. Comparable differences in the redox state of the primary electron acceptor persisted up to 95 µmol m−2 s−1 actinic red light (Figure 8C), indicating that the tap38 mutants can redistribute a larger fraction of energy to PSI, in accordance with the increase in its antenna size under state 2 light conditions (see Figure 4B; Table S1 and Figure 6). This idea was supported by measurements of the maximum (FV/FM) and effective (ΦII) quantum yields of PSII. FV/FM remained unaltered in mutant plants (see Figure 8D, dark-adapted plants, photosynthetically active radiation [PAR] = 0), indicating WT-like efficiency of mutant PSII complexes. However, ΦII was increased in tap38-1 (0.75±0.01) and tap38-2 (0.73±0.02) relative to WT (0.72±0.01), suggesting that electron flow through the thylakoids was more efficient in tap38 mutants (Figure 8D). The improvement in photosynthetic performance of the tap38 mutants was most pronounced under low and moderate illumination (Figures 8C and 8D), as expected from their growth phenotype.
How does TAP38 control LHCII dephosphorylation? Three possibilities appear plausible: TAP38 (1) negatively regulates the activity of STN7 (e.g., by dephosphorylating it [40]), (2) dephosphorylates LHCII directly, or (3) forms part of a phosphorylation/dephosphorylation cascade that controls the activity of the LHCII kinase or phosphatase. The observation that oeTAP38 plants, although showing a >20-fold increase in TAP38 levels, still exhibit residual LHCII phosphorylation (see Figure 5C), argues against the idea that TAP38 does inhibit STN7 by dephosphorylation. Differences in TAP38 levels resulted in a clear change in pLHCII levels: although in tap38 mutants a strong reduction in TAP38 led to a constantly high level of pLHCII and an increase in the amount of the PSI-LHCI-LHCII complex, strong overexpression of TAP38 (oeTAP38) caused the complete disappearance of pLHCII attached to PSI, although pLHCII was still present.
Taking these observations together, it appears that the TAP38 phosphatase acts specifically on pLHCII associated to PSI-LHCI complexes. Indeed, the dephosphorylation of pLHCII still attached to PSII under state 2-inducing light conditions seems unfavorable in terms of energy efficiency.
Interestingly, in WT where pLHCII levels can vary dramatically depending on the light conditions [9],[12] (see also Figure 5A), TAP38 seems to be constitutively expressed under the different light conditions applied (see Figure 3C). A plausible explanation for this is that TAP38 is constitutively active and directly responsible for the dephosphorylation of pLHCII. For that, TAP38 would need to be present in a certain concentration range (as it is the case for WT) to constantly dephosphorylate pLHCII. In agreement with that, thylakoid protein phosphatase reactions have been described as redox independent, leading to the conclusion that the redox dependency of LHCII phosphorylation is a property of the kinase reaction [41]. This, together with the observation that Stt7 levels increase under prolonged state 2 conditions (favoring LHCII phosphorylation) and decrease under state 1 conditions (favoring dephosphorylation of LHCII) [17], argues in favor of the hypothesis that the LHCII kinase is the decisive factor in controlling the phosphorylation state of LHCII. Despite the obvious TAP38 dosage dependence of pLHCII dephosphorylation (see Figures 5 and 6), TAP38 activity could be regulated on other levels than only its abundance. However, the strong decrease or increase of TAP38 levels in tap38 mutant and oeTAP38 plants might interfere with other types of regulation in these genotypes.
As outlined above, the dependence of LHCII dephosphorylation upon TAP38 dosage—when comparing tap38 mutants, WT, and TAP38 overexpressors—strongly suggests that TAP38 dephosphorylates pLHCII directly, particularly when it is associated with the PSI-LHCI complex. Alternatively, TAP38 could act in a phosphorylation/dephosphorylation cascade that controls the activity of the LHCII phosphatase. Although the latter hypothesis cannot be totally excluded, a set of evidences point to a direct role of TAP38 on LHCII phosphorylation. Indeed, our in vitro dephosphorylation assay clearly indicated that TAP38 can dephosphorylate pLHCII directly (see Figure 7). Moreover, as in the case of STN kinases, extensive efforts searching to identify other LHCII phosphatase candidates failed: knockout lines for all the protein phosphatases demonstrated to be located in the chloroplast [25]–[29] did not show any alteration in LHCII phosphorylation. Additionally, extensive biochemical studies did not reveal the existence of a complex network of phosphatases involved in LHCII dephosphorylation, but postulated the involvement of only two distinct chloroplast protein phosphatases from different families in the dephosphorylation of thylakoid phosphoproteins [20]–[23],[42]. Our data support this notion, as shown by the absence of major alterations in the phosphorylation pattern of CP43, D1, and D2 subunits in tap38 mutant plants (see Figure 5). Moreover, pLHCII dephosphorylation was suggested to be catalyzed by only two independent protein phosphatases, a membrane-bound one and a stromal protein phosphatase [42]. In contrast to this, our results clearly show that TAP38, a thylakoid-associated phosphatase, alone is responsible for LHCII dephosphorylation. Thus, although slightly leaky, the tap38-1 mutants show a large fraction of LHCII in the phosphorylated state under all investigated conditions (see Figure 5). If a second LHCII phosphatase with redundant function would operate in chloroplasts, one would expect some residual dephosphorylation of pLHCII. A plausible explanation for the previously shown stromal pLHCII dephosphorylation activity [22] might be that during the preparation of stromal extracts, a significant portion of TAP38 was released from the thylakoid membrane into the stroma. Interestingly, TAP38 appears to influence also the phosphorylation levels of other thylakoid proteins, as shown by the higher phosphorylation of the CAS protein in tap38-1 thylakoids (see Figure 5). Taking these observations together, it appears that, as in the case of the STN kinases, two distinct phosphatases are needed to dephosphorylate LHCII and PSII core proteins. TAP38, similar to the STN7 kinase, seems to have a high specificity for pLHCII associated with PSI-LHCI complexes as substrate. The counterpart of STN8 [9],[43], the PSII core–specific phosphatase, remains to be identified. However, as in the case of the STN7 and STN8 kinases, some degree of substrate overlap seems to exist also between the phosphatases, as shown by the more rapid dephosphorylation of PSII-D1/D2 subunits in the TAP38 overexpressor lines exposed to far-red light conditions (see Figure 5C). Additionally, it is noteworthy that the activity of TAP38 does not seem to be restricted to STN7 substrates, as shown by its influence on CAS protein phosphorylation, previously reported to be a substrate of the STN8 kinase [44].
It is known that an increase in the relative size of the reduced fraction of the plastoquinone pool (PQH2) enhances phosphorylation of LHCII [1],[5],[39],[45]. Depletion of TAP38 in tap38 mutants, however, increases both LHCII phosphorylation (see Figure 5B) and PQ oxidation (see 1-qP values in Figure 8C). This discrepancy can be resolved by assuming that the enhanced oxidation of PQ caused by the increase in PSI antenna size (and LHCII phosphorylation) in tap38 plants is not sufficient to down-regulate the LHCII kinase to such an extent that it can compensate for the decline in LHCII dephosphorylation.
The enhanced photosynthetic performance indicated by an increase in ΦII and a decrease of 1-qP (see Figure 8C and 8D), as well as the growth advantage of the tap38 mutants under constant moderate-light intensities that stimulate LHCII phosphorylation and state 2, can be attributed to the redistribution of a larger fraction of energy to PSI. This is in accordance with the increase in PSI antenna size in tap38 mutants when compared to WT plants (see Figure 4B, Table S1, and Figure 6). Therefore, it is straightforward to speculate that the enhanced PSI antenna size provides the tap38 mutants with a more robust photosynthetic electron flow under conditions that preferentially excite PSII and induce state 2. As a consequence of the more balanced light reaction, the photosynthetic efficiency is improved resulting in an increased growth rate. However, the fitness advantage will revert under conditions that induce state 1, or under more natural conditions with fluctuating light; here, it can be expected that tap38 mutants will perform less efficiently than the WT with respect to photosynthesis and growth, very similar to what has been observed for the stn7 mutant [12],[13].
Taken together, future analyses should clarify which protein phosphatase is involved in the dephosphorylation of PSII core proteins and which are the counterparts of higher plant phosphatases, including TAP38, in Chlamydomonas (which apparently lacks a TAP38 ortholog). Additionally, further biochemical evidences that TAP38 (and STN7) uses pLHCII as a substrate will be very important for the complete molecular dissection of state transitions.
Procedures for plant propagation and growth measurements have been described elsewhere [46]. The tap38-2 insertion line (SALK_025713) was identified in the SALK collection [34] (http://signal.salk.edu/), whereas insertion line tap38-1 (SAIL_514_C03) originated from the Sail collection [33]. Both lines were identified by searching the insertion flanking database SIGNAL (http://signal.salk.edu/cgi-bin/tdnaexpress). To generate oeTAP38 lines, the coding sequence of TAP38 was cloned into the plant expression vector pH2GW7 (Invitrogen). For complementation of the tap38-1 mutant, the TAP38 genomic DNA, together with 1 kb of its natural promoter, was ligated into the plant expression vector pP001-VS. The constructs were used to transform flowers of Col-0 or tap38-1 mutant plants by the floral dipping technique as described [47]. Transgenic plants, after selection for resistance to hygromycin (oeTAP38) or Basta herbicide (complemented tap38-1), were grown on soil in a climate chamber under controlled conditions (PAR: 80 µmol m−2 s−1, 12/12 h dark/light cycles). The T2 generation of the oeTAP38 plants was used for the experiments reported. Successful complementation of tap38-1 mutants was confirmed by measurements of chlorophyll fluorescence and LHCII phosphorylation levels under light regimes promoting state transitions.
The full-length coding region of the TAP38 gene was cloned into the vector pGJ1425, in frame with, and immediately upstream of the sequence encoding dsRED [32]. Isolation, transfection, and fluorescence microscopy of A. thaliana protoplasts were performed as described [48].
The coding region of TAP38 was cloned into the pGEM-Teasy vector (Promega) downstream of its SP6 promoter region, and mRNA was produced in vitro using SP6 RNA polymerase (MBI Fermentas). The TAP38 precursor protein was synthesized in a Reticulocyte Extract System (Flexi; Promega) in the presence of [35S]methionine. Aliquots of the translation reaction were incubated with intact chloroplasts, and protein uptake was analyzed after treatment of isolated chloroplasts with thermolysin (Calbiochem) as described previously [49]. Labeled proteins were subjected to SDS-PAGE and detected by phosphorimaging (Typhoon; Amersham Biosciences).
Total RNA was extracted with the RNeasy Plant Mini Kit (QIAGEN) according to the manufacturer's instructions. cDNA was prepared from 1 µg of total RNA using the iScript cDNA Synthesis Kit (Bio-Rad) according to the manufacturer's instructions. For semiquantitative reverse-transcriptase PCR, cDNA was diluted 10-fold, and 3 µl of the dilution was used in a 20-µl reaction. Thermal cycling consisted of an initial step at 95°C for 3 min, followed by 30 cycles of 10 s at 95°C, 30 s at 55°C, and 10 s at 72°C. For real-time PCR analysis, 3 µl of the diluted cDNA was mixed with iQ SYBR Green Supermix (Bio-Rad). Thermal cycling consisted of an initial step at 95°C for 3 min, followed by 40 cycles of 10 s at 95°C, 30 s at 55°C, and 10 s at 72°C, after which a melting curve was performed. Real-time PCR was monitored using the iQ5Multi-Color Real-Time PCR Detection System (Bio-Rad). All reactions were performed in triplicate with at least two biological replicates.
Total protein extracts and proteins from total chloroplasts, thylakoids, and the stroma fraction were prepared from 4-wk-old leaves in the presence of 10 mM NaF as described [48],[50]. Immunoblot analyses with phosphothreonine-specific antibodies (Cell Signaling) or polyclonal antibodies raised against the mature TAP38 protein were performed as described [45].
For BN-PAGE, thylakoid membranes were prepared as described above. Aliquots corresponding to 100 µg of chlorophyll were solubilized in solubilization buffer (750 mM 6-aminocaproic acid; 5 mM EDTA [pH 7]; 50 mM NaCl; 1.5% digitonin) for 1 h at 4°C. After centrifugation for 1 h at 21,000g, the solubilized material was fractionated by nondenaturing BN-PAGE at 4°C as described [38].
For 2D-PAGE, samples were fractionated in the first dimension by BN-PAGE as described above and subsequently by denaturing SDS-PAGE as described previously [51]. Densitometric analysis of the stained gels was performed using the Lumi Analyst 3.0 (Boehringer).
State transitions were measured by pulse amplitude modulation fluorometry (PAM) [35],[36] and 77 K fluorescence emission analysis [12],[37]. Plants adapted to state 1 conditions were obtained by incubation either in darkness or far-red light, whereas state 2 was induced by either red- or low-light illumination. Both state 1 and state 2 light-inducing conditions were used in different combinations, since they resulted in identical effects on state transitions. Additionally, there was no major reason to prefer one light setting to the other, except for the fact that the PAM fluorometer is equipped with red and far-red lights. For state transition measurements, five plants of each genotype were analyzed, and mean values and standard deviations were calculated. In vivo chlorophyll a fluorescence of single leaves was measured using the Dual-PAM 100 (Walz). Pulses (0.5 s) of red light (5,000 µmol m−2 s−1) were used to determine the maximum fluorescence and the ratio (FM−F0)/FM = FV/FM. Quenching of chlorophyll fluorescence due to state transitions (qT) was determined by illuminating dark-adapted leaves with red light (35 µmol m−2 s−1, 15 min) and then measuring the maximum fluorescence in state 2 (FM2). Next, state 1 was induced by adding far-red light (maximal light intensity corresponding to level 20 in the Dual-PAM setting, 15 min), and recording FM1. qT was calculated as (FM1−FM2)/FM1 [36].
For 77 K fluorescence emission spectroscopy, the fluorescence spectra of thylakoids were recorded after irradiating plants with light that favored excitation of PSII (80 µmol m−2 s−1, 8 h) or PSI (LED light of 740 nm wavelength, 4.6 µmol m−2 s−1, 2 h). Thylakoids were isolated in the presence of 10 mM NaF as described [11], and 77 K fluorescence spectra were obtained by excitation at 475 nm using a Spex Fluorolog mod.1 fluorometer (Spex Industries). The emission between 600 and 800 nm was recorded, and spectra were normalized relative to peak height at 685 nm. Data frequency was of 0.5 nm with an integration time of 0.1 s.
pLHCII was obtained from fractionation of tap38-1 thylakoids by sucrose gradient ultracentrifugation as previously described [45]. The cDNA sequence of mature TAP38 was cloned into pET151 (Invitrogen), and recombinant TAP38 (recTAP38) was expressed in the E. coli strain BL21 with a N-terminal-6x His-tag. recTAP38 was purified under denaturing conditions following a Ni-NTA batch purification procedure according to the manufacturer's instructions (Qiagen). After protein precipitation in 10% trichloroacetic acid (TCA) followed by three washing steps with absolute ethanol, around 500 µg of TAP38 protein were resuspended in 500 µl of 1% (w/v) lithium dodecyl sulfate (LDS), 12.5% (w/v) sucrose, 5 mM ε-aminocaproic acid, 1 mM benzamidine, and 50 mM HEPES KOH (pH 7.8), as previously described [52]. Subsequently, TAP38 protein was boiled for 2 min at 100°C and then transferred for 15 min at 25°C. Then, dithiothreitol (DTT; 75 mM final concentration) was added, and the solution was subjected to three freezing-thawing cycles (20 min at −20°C, 20 min at −80°C, 20 min at −20°C, thawing in a ice-water bath, and 5 min at 25°C). After completion of the three freezing-thawing cycles, octyl-glucopyranoside (OGP; 1% [w/v] final concentration) was added, and the solution was kept on ice for 15 min. Afterwards, KCl (75 mM, final concentration) was added to precipitate the LDS detergent. After centrifugation at 16,000g at 4°C for 10 min, the supernatant containing the refolded TAP38 in the presence of 1% (w/v) OGP was collected. Subsequently, 1 µl of phosphatase was incubated together with pLHCII corresponding to 2 µg of total chlorophyll. The dephosphorylation reaction was performed in 50 µl containing 0.06% (w/v) dodecyl-ß-D-maltoside, 5 mM Mg-acetate, 5 mM DTT, 100 mM HEPES (pH 7.8), at 37°C for 2 h as previously described [22]. The reaction mixture was loaded on a SDS-PAGE and immunodecorated with a phosphothreonine-specific antibody, as described above.
|
10.1371/journal.pntd.0005682 | Novel inference models for estimation of abundance, survivorship and recruitment in mosquito populations using mark-release-recapture data | Experiments involving mosquito mark-release-recapture (MRR) design are helpful to determine abundance, survival and even recruitment of mosquito populations in the field. Obstacles in mosquito MRR protocols include marking limitations due to small individual size, short lifespan, low efficiency in capturing devices such as traps, and individual removal upon capture. These limitations usually make MRR analysis restricted to only abundance estimation or a combination of abundance and survivorship, and often generate a great degree of uncertainty about the estimations.
We present a set of Bayesian biodemographic models designed to fit data from most common mosquito recapture experiments. Using both field data and simulations, we consider model features such as capture efficiency, survival rates, removal of individuals due to capturing, and collection of pupae. These models permit estimation of abundance, survivorship of both marked and unmarked mosquitoes, if different, and recruitment rate. We analyze the accuracy of estimates by varying the number of released individuals, abundance, survivorship, and capture efficiency in multiple simulations. These methods can stand capture efficiencies as low as usually reported but their accuracy depends on the number of released mosquitoes, abundance and survivorship. We also show that gathering pupal counts allows estimating differences in survivorship between released mosquitoes and the unmarked population.
These models are important both to reduce uncertainty in evaluating MMR experiments and also to help planning future MRR studies.
| Mosquito-borne diseases such as dengue and malaria impose a global burden with recurrent outbreaks. Recently, emergence of arboviral diseases caused by Zika and chikungunya viruses has also become a global concern. Knowledge about the ecology of mosquito populations under natural conditions may provide significant aid to help designing more effective vector control strategies. Quantitative metrics such as the abundance of mosquito populations are difficult to be measured in the field without resorting to experiments with markers. There are, however, limitations to these kinds of experiments such as short mosquito lifespan, marking limitations due to small body size, low efficiency in capturing devices such as traps, and once-only individual capture. Due to these limitations most methods estimate either only abundance or a combination of abundance and survivorship. In this work, we present statistical methods designed to estimate abundance, survivorship and recruitment using inference models and information such as counts of pupae. Results indicate that having low capture efficiencies as often observed in field assays still permits good estimation. Also, low number of released mosquitoes compromise density and survival estimations. We expect these methods to be helpful to people collecting mosquito field data and for health analysts to evaluate possible outcomes of control interventions.
| Mark-release-recapture (MRR) methods applied to study mosquito populations permit analysis of vector survival, dispersal, and abundance in natural environment. Various mosquito species, in particular of the Aedes, Culex and Anopheles genera, are vectors associated with persistent diseases such as dengue, filariasis and malaria and also emergent infections by chikungunya and Zika viruses. Given such medical importance, early mathematical models for malaria transmission [1,2] established the vectorial capacity as an important metric to assess epidemic risk by a mosquito population. Reliable vectorial capacity assessment requires accurate estimations of mosquito density (mosquitoes/human) and survivorship (daily survival probability). These estimates typically help to improve vector control policies and practices in endemic regions and might lead to mitigation of disease transmission [3].
By their nature, mosquito MRR experiments have important design restrictions that hinder the application of more sophisticated capture-recapture models such as the commonly known Jolly-Seber method [4]. For example: (a) individual mosquitoes are released and typically not recaptured multiple times because once collected at traps they do not survive for new releases, (b) recapture rates are low, often ranging from 5–10% [5], (c) most of the experimental designs, with notable exceptions [6], consider groups of marked individuals as cohorts due to small mosquito body size and consequent difficulty of individual marking methods and because a high number of mosquitoes are released from a few selected points, and (d) average lifespan under natural conditions is short. These limitations restrict models which consider individual markers and multiple recaptures. In early designs of capture-recapture experiments involving mosquitoes, most works used deterministic estimators such as Lincoln-Petersen and Fisher-Ford indexes to evaluate vector abundance [3,7]. Currently, deterministic models are still used mainly due to lower mathematical complexity, when compared to stochastic/Bayesian models. In the case of the Lincoln-Petersen index, the ratio between the number of marked individuals recaptured and the total insects released allows estimation of the total abundance from the count of captures of unmarked individuals. For an MRR experiment spanning at most a dozen days, we have observations over multiple days, but only a low number of recaptures due to low capture efficiencies at traps. In the case of mosquito populations, Lincoln-Petersen abundance estimation is expectedly inaccurate, since the number of marked mosquitoes alive for trapping after a few days is significantly smaller than the number released due to a sharpened mortality across the released cohort, plus trapping on previous days [8–10]. In fact, daily captures of mosquitoes at traps typically exhibit an exponential decay largely due to mortality of marked individuals. The Fisher-Ford model [11] is another deterministic method that requires the probability of daily survival to adjust the capture ratio for the multiple estimations over time. Estimates of survival probability are possible using MRR data from the exponential decay of capture counts of marked individuals. In order to estimate abundance, Buonaccorsi et al. [12] consider not only the survival probability but also removal of individuals captured at traps. Recruitment in mosquito MRR experiment areas occurs either through birth or immigration. Recruitment rate estimation is possible under stable abundance, even though still challenging due to mosquito MRR limitations.
Here we build Bayesian models that leverage the concepts behind the Fisher-Ford model [11] and Buonaccorsi et al.’s model [12]. Moreover, we propose another two novel Bayesian approaches to estimate relevant parameters of mosquito population biology such as adult population size, survival rates and recruitment. Recruitment estimation is possible if assuming equal adult survival rates or including a component into the model that uses counts of immature individuals, typically pupae. For various mosquito species such as Ae. aegypti pupae are known to present low mortality and thus are likely to emerge as adult individuals [13]. Analyses using these models permit us to infer abundance, survivorship and recruitment rate using both field data and datasets obtained from simulations, when taking into account counts of immature individuals. Furthermore, our results reveal the degree of tolerance of these methods to both capture efficiency at traps and number of released mosquitoes.
We used the capture counts of adult females obtained from trap collections in the Z-10 neighborhood located at the city of Rio de Janeiro, Brazil, during an MRR experiment described by Villela et al. [10]. We used data from experiment ST2, in which a single release point (map available in Villela et al. [10]—supplementary files) was considered. We summed the number of trapped individuals over all traps for each day in the study. Before releases started, pupal surveys were carried out over all of the breeding sites found in the 66 premises containing an adult trap, observed in the same occasion when the trap was installed. A total of 212 larvae and 47 pupae were collected in 7 containers from 7 (11%) different dwellings. All immatures were collected in man-made containers such as plastic plant dishes and uncovered water tanks and were brought to the entomology laboratory at Fiocruz for further classification using taxonomic keys. The choice about using number of adult females is due to the use of adult traps specifically designed to attract female mosquitoes [10].
Several designs are used for mosquito MRR trials. Guerra et al. [15] assembled data from publicly reported mosquito MRR trials and provided a quantitative synthesis. In most of the experimental designs mosquitoes typically are not recaptured multiple times and are marked as cohorts using markers such as fluorescent dust. In MRR experiments, such as reported by Maciel-de-Freitas et al. [8,9] and Ritchie et al. [16], typically counts of recaptured mosquitoes from all cohorts (signaled by color of fluorescent dust) and counts of unmarked, captured mosquitoes are taken at each of multiple traps across an area over around a 10-day period.
We simulated multiple scenarios numerically (for instance, varying number of releases, capture probabilities and survival probabilities). Each simulation requires initial conditions and parameters such as abundance U at the beginning of the experiment, daily probabilities φ and φu of survival for both marked and unmarked individuals, the daily recruitment rate b. An MRR study requires a number D of days of mosquito collection at traps. Mosquito capture occurs with a given capture efficiency β0, the number of released mosquitoes N, and the number of traps J. A pupal search in the experiment area that typically would occur in the field a day or two earlier than releasing time collects a number of pupae npupae. If the pupal search is imperfect, the number of pupae collected is given by μnpupae, where μ describes the efficiency of pupal search. The simulation returns the daily numbers mi and ui of individuals captured at traps, both marked and unmarked ones, respectively. Table 1 shows the variables, parameters used in the simulation model and a short description.
The inference models describe relationships between the known values, such as number N of released mosquitoes, the number of days post-release D, and observed data, such as numbers mi and ui of marked and unmarked mosquitos collected at traps at day i, 1 ≤ i ≤ D, using parameters to be estimated.
Let p be the vector of probabilities of capture along the observation periods i, 1 ≤ i ≤ D. First, we consider the number of individuals captured over the MRR experiment time as mc ~ Binomial(Σi=1Dpi, N). Then, we take a multinomial distribution for the observations mi captured at each day i: m ~ Multinomial(p, mc).
For a first naïve model M0, we consider the probability pi of capture to be only dependent on the trap capture efficiency β0. For a second model MS, we describe the capturing probability by a product of capture efficiency β0 and time effects, to be estimated. Therefore, log (pi) = θ0 + i θ1, where the estimated capture efficiency β0 = exp(θ0) and estimated survival probability φ = exp(θ1). Such a model is more general than model M0, since the basic assumption in model M0 is equivalent to assume simply that θ1 = 0, which corresponds to no mortality effects at any time i. Multiple daily estimations applying models M0 and MS are Bayesian counterparts to multiple values of abundance obtained by Lincoln-Petersen and Fisher-Ford estimators, respectively. For a third model MB, we allow for removal of individuals given the daily captures at traps in a Bayesian counterpart to the model proposed by Buonaccorsi et al. [12]. In this case, the probability of capture is pi = β0(1 − β0)i−1φi, for marked individuals and pi = β0(1 − β0)i−1, for unmarked individuals. For unmarked individuals, this model does not permit estimation of probability of daily survival φu. In this case, the underlying assumption is that over a short period of time, typically few days, recruitment is equal to mortality.
The observed number ui of unmarked individuals collected at traps is modeled as ui ~ Poisson(U pi), for models M0, MS, and MB, where the abundance number U is to be estimated. We use a prior distribution for abundance U ~ Gamma(0.001, 0.001). We also have prior distribution for capture efficiency β0 ~ Beta(2,4) and for probability of daily survival of marked individuals φ = Beta(4,2), which are lightly informative distributions, concentrating most mass at values close to 0 in the case of capture efficiency and close to 1 in the case of probability of daily survival.
We build two other models that include a recruitment component, including one considering the number of pupae collected from experiments before releasing mosquitoes. We build these models using relationships also described for model MB, i.e., having survivorship equal along with the experiment days and also accounting for removal of individuals. We also consider for both models the number of individuals captured over the MRR experiment time as mc ~ Binomial(Σi=1Dpi, N). Then, we take a multinomial distribution for the observations mi captured at each day i: m ~ Multinomial(p, mc).
We define model MRSU for which we assume survival of unmarked individuals equal to the one of marked individuals, i.e., essentially φu = φ. Therefore, over a short period of time such as an MRR experiment duration, recruitment should occur at rate that maintains population at a constant level. The number of unmarked individuals at each time period, i.e. at risk of being trapped, is the sum of a number Ui of surviving individuals from start of the experiment and the total number Vi of recruited individuals. In the model a number of mosquitoes given by a recruitment rate b enter the experiment at each time interval. Therefore, from time i-1 to time i the number of mosquitoes should increase by rate bi = bβ0(1 − β0)i−1φi. We have the same vector of probability capture described for model MB, pi = β0(1 − β0)i−1φi. In model MRSU, the sum of remaining individuals is a latent variable given by Ui ~ Poisson(U(1 − β0)i−1φi) and the number of recruited individuals is another latent variable given by Vi ~ Poisson(bΣj=1i(1−β0)j−1φj).
We define model MRP distinguishing the probabilities of daily survival of unmarked and marked mosquitoes, in order to estimate parameter φu. We describe the number of immature collected before the experiment to be npupae ~ Binomial(fa(1 − φu),τ U/s), where fa is a factor that describes how extensive is the immature search, τ is the pupal maturation time and s is the fraction of the targeted group in the mosquito population. Very commonly, the purpose is to estimate the abundance of female mosquitoes. Here, pupal maturation time is τ = 2 days and the fraction of female mosquitoes is s = 0.5 [3]. Factor fa represents an adjustment since the immature search typically covers only a fraction of the area surveyed, or alternatively a fraction of the number of premises. We have the remaining and recruited individuals assessed in the same way, but survivorship for unmarked individuals is given by φu: capture counts of surviving individuals Ui ~ Poisson(U (1−α)i−1φui) and recruitment quantities Vi ~ Poisson(bΣj=1i(1−β0)j−1φuj).
For both models MRSU and MRP, the observed number of individuals is given by ui ~ Binomial(β0, Ui + Vi). We use a prior distribution for abundance U ~ Gamma(0.001, 0.001), for capture efficiency β0 ~ Beta(2,4) and for probability of daily survival of individuals φ = Beta(4,2) and φu = Beta(4,2), where appropriate, and for basic recruitment rate b ~ Lognormal(10, 0.25).
As a reference, Table 2 describes the assumptions behind each of these models and which estimators can be extracted from them.
We implemented the simulation tool using the R platform [17]. We also wrote description models using the WinBUGS language [18] for the statistical models (M0, MS, MB, MRSU, MRP). We analyze the simulation data via Monte-Carlo Markov chain simulations (MCMC), by running 3 separate chains, 360,000 iterations during each of the chains, with a 320,000 burn-in period. These numbers sufficed for good convergence except otherwise noted within our results. We use R to load the simulation data and streamline pre-processed data via package R2JAGS [19] into JAGS [20], the selected tool for MCMC analysis. Output from MCMC analysis permits us to obtain samples of the posterior distribution, and as a result, mean and median values, as well as credibility intervals (CI). In S1 Text we present boxes that contain the description of our models prepared for JAGS tool. Our scripts for simulation and analysis are publicly available at https://github.com/DVMath/MosqCapRecap.
We estimated abundance of Aedes aegypti mosquito population in the Z-10 neighborhood in Rio de Janeiro from inference analysis using models M0, MS, MB, MRSU, and MRP. Results from using model M0 reveal a much larger abundance (Fig 1A). Indeed, an overestimation is expected, since this model does not consider either survival estimation or removal of individuals. Estimation from the posterior distribution results average values of abundance in the releasing day that are 3,326 (95% CI: 2,794–3,944), 2,875 (95% CI: 2,171–3,676), 1,636 (95% CI: 1,152–2,345), and 2,143 (95% CI: 1,574–2,890) female mosquitoes, from analysis using models MS, MB, MRSU, MRP, respectively. Probability of daily survival from posterior distributions obtained from analyses of models MB, MRSU, and MRP were very similar (Fig 1B). In the case of model MRP the mean probability of daily survival was 0.77 (95% CI: 0.72–0.83). The mean recruiting rate was estimated at 530 mosquitoes per day (95% CI: 383–701) for model MRP. Since the method is sensitive to the number of pupae collected in the field, we estimate abundance using model MRP considering various alternative possibilities such as a twofold, half and a quarter of the collected number of pupae (Fig 1C). The last two possibilities (half and quarter of the collected number) result in smaller abundance estimations, when considering the collected number to be the closest to the number of pupae in the area. By contrast, a larger recruiting rate is expected if the real number of pupae to be collected is twofold.
Table 3 contains results obtained from multiple simulation experiments using models M0, MS, MB, MRSU, MRP. Model MRP included the assumed input values of abundance within credibility intervals in 16 simulation studies (indicated by the number of asterisks in the MRP column). For assumed values of abundance of 8,000 mosquitoes and above, model MRP underestimated the abundance. For values of probability of daily survival of unmarked individuals less than 0.8, model MRP resulted in either overestimation (study 19) or underestimation (studies 23 and 25).
Comparing assumed input values for study 1 and its estimations in Table 4, all parameters were estimated close to the assumed values and the 95% credibility intervals indeed contain these assumed values. Analysis by model MRP results in abundance of 4,220 mosquitoes (95% CI: 3,572–5,067) for an assumed abundance value of 4,000 mosquitoes. We also estimated probability of daily survival (PDS) for unmarked at 0.86 and marked individuals at 0.77 and recruitment rate 624 individuals/day. Analysis from model MRSU reveals an estimation of a 95% credibility interval also containing the abundance value for simulation study 1. The probability of daily survival, however, is wrongly estimated due to the assumption of equal survival rates for all individuals whether marked or not. Model MB permits estimation of abundance, probability of daily survival (marked individuals) and trap capture efficiency. Estimates given by model MB are also close to the assumed values, which are well within the 95% credibility intervals. Model MS does not consider removal of individuals, an assumption that proves costly since it underestimated both the abundance and probability of daily survival. Model M0 results greatly overestimate abundance due to not considering the daily survival.
For simulations with at least 1000 marked mosquitoes, mean estimated abundance values are close to the assumed values, which are within the 95% credibility interval. Values below 1,000 marked mosquitoes were not quite as close to the estimation value. Also, the 95% credibility interval in these cases gets much larger as size of the released cohort decreases. Inspection of results from very low values indicates high uncertainty, as expected (Fig 2A). Fig 2B indicates that the low levels of capture counts due to relatively few release numbers prove costly to the capture efficiency estimation resulting in overestimation. As a consequence, abundance is underestimated.
Fig 3 shows results for survival probabilities under MRP model considering only simulation experiments with same abundance values and release numbers, but varying probability of daily survival of marked individuals and unmarked population. Estimation of PDS for both marked cohort and unmarked cohorts are close to the input values assumed in the simulations, although in some cases for marked population the assumed values are closer to the extremes of the 95% credibility intervals.
If efficiency at collecting pupae is low, results from using model MRP indicate estimations deviating from the assumed input values for abundance, recruitment and probability of daily survival. Fig 4 shows this pupal search efficiency at different levels equal to 25%, 50%, 75% and 100%. As expected, the ideal case (100%) is the best scenario, since estimations are close to the assumed values. For low number of released mosquitoes the estimation also gets worse due to the low capture counts.
Fig 5 shows the impact of distinct probabilities of daily survival between marked individuals and unmarked individuals. The cases in the middle column correspond to the assumption in the model, and assumed input values in the simulations lie within the 95% credibility intervals. In the cases where there is difference (left and right columns) between the survival of the two populations, results for recruitment rate (Chart B) in model MRSU are not as close to the expected values. Since estimation of all parameters is intertwined, abundance estimates (Chart A) also get worse.
In Fig 6 capture efficiency varies from 0.03 to 0.1, as we consider only simulation studies that assumed all other parameters (abundance, survival, recruitment) equal. As expected, as the capture efficiency lowers, uncertainty increases, since capture counts are low. As a consequence, 95% credibility intervals are large for capture efficiency smaller than 0.05 (5%). As a surprising effect, the estimations for capture efficiencies at 0.05, 0.08 and 0.1 do not reveal significant difference in their 95% credibility intervals.
We defined Bayesian models to estimate abundance, recruitment and probability of daily survival of mosquito populations in the field from MRR experiment data. Analyses using these models result in posterior distributions for these parameters, hence mean and 95% credibility intervals can be obtained. Moreover, counts from pupal surveys were instrumental to obtain estimated recruitment rates of the wild population. These estimates are particularly interesting since immature counting in breeding sites is one of the most common vector control approaches in countries endemic for arboviruses infections.
Our first set of simple Bayesian inference models is based on estimating the capture efficiency and the probability of daily survival with close relationship to existing methods used for MRR analysis. Since mosquitoes are not often individually captured multiple times (once captured they are effectively removed from the study), a Bayesian model should better describe removal of individuals not only due to mortality but also from the capture process itself. This model has close association to the method proposed by Buonaccorsi et al.[12]. Simpler models are defined by neglecting removal of individuals, but still assuming limited survivorship and also neglecting mortality in order to establish Bayesian counterpart models to commonly used Fisher-Ford and Lincoln-Petersen estimators [3,11]. In the case of Lincoln, such approach is not unprecedented since Gaskell and George [21] presented a Bayesian estimation for the Lincoln index. The Bayesian method enabled by our model MB permits inference about the capture efficiency and the probability of daily survival. However, it may not achieve accurate estimations, depending on conditions of large difference between probability survival of marked and unmarked mosquitoes, large abundance or low capture efficiency.
Estimation of recruitment becomes challenging due to the usual mosquito MRR limitations. The concept of using pupal counts for assessment of abundance has been proposed by Focks et al. [22] and also advised in other works [15,23]. The estimation implicitly assumes, based on strong sampling efficiency, that pupae numbers should balance with mortality rates, for constant population sizes, therefore the pupae count is the product of the abundance and the mortality rate, but also accounting for sex ratio and the average pupating time. Since our Bayesian framework assumes priors for probability of daily survival and abundance, a description in the model for the number of pupae relating to both survival and abundance is natural, accounting for a factor that the pupae collection might not cover the whole study area. Models MRP and MRSU also permit to estimate recruitment, either assuming collection of pupae or not, respectively. Depending on this information, we can evaluate any potential difference between daily survival of marked and unmarked mosquitoes.
We estimated abundance, survivorship and recruitment rate of an Aedes aegypti population in an area in Rio de Janeiro, Brazil, from an experiment conducted in March 2013, described by Villela et al.[10]. The mean number per premise varied from 2.1 mosquitoes per premise (MRSU model) to 4.2 mosquitoes per premise (MS). Such twofold increase shows the importance of choosing the appropriate model to describe parameters of Aedes aegypti biology. As shown when using simulated datasets, analysis using model MRP achieves intervals that include the simulation input value in most of the studied scenarios. Daily recruitment rate in the field was about 0.67 mosquitoes per premise in the analysis from model MRP. In this case, the recruited number would be about a quarter of the total abundance. The effectiveness of vector control approaches such as targeting the most productive container or using chemical compounds (insecticides) might be evaluated based on potential changes on mosquito recruitment rates. For more effective the vector control intervention, greater decrease in recruitment rate would be expected.
Our results from analyses of simulated datasets show that these models can tolerate capture efficiencies as low as the ones observed for mosquito MRR. We also varied the abundance levels, as opposed to the released numbers, and differences in the survivorship between marked, released mosquitoes and the unmarked population. In the case of immature counts (pupae), recruitment rate can also be estimated, but we find it to be highly dependent on extensive pupal collection, which can require extensive resources in the field.
Limitations in the design of mosquito MRR studies expectedly impact estimation of abundance, survivorship and recruitment rates. First, when abundance is large, the number of released mosquitoes is critical, regardless of the method used. Also capture efficiency in regular MRR experiments is usually small, varying in the range of 5–10% [5,10]. We have shown that such rates are still acceptable, but capture efficiencies below this range lead to higher degree of uncertainty in the estimation. By contrast, to reduce credibility intervals most likely we would need a combination of higher efficiencies and multiple individual recaptures, which is very difficult to implement in the field due to trap conditions. Also, if adapting these methods to have spatial estimations, we expect effects from low capture counts, as opposed to aggregate counts. Otherwise, methods such as proposed by Villela et al. [10] that also involve a likelihood component are required due to distance from mosquito concentration areas to traps.
Collecting pupae in the field can be difficult due to limited accessibility to breeding sites, but we think that results from model MRP should motivate getting such samples to have better estimates. Our results indicate sensitivity of recruiting rate estimates when assuming different number of pupae. Because pupal surveys may have difficult feasibility to be conducted on the routine of vector control programs, our results demonstrate that surveys with varying degrees of imperfection lead to biased estimations of abundance, recruitment and survival rates. Conversely, public health decision makers might adopt models such as MRP and MRSU with attention to these issues. For example, the Brazilian dengue national control program recommends a survey 4–6 times yearly in around 10% of cities of each district of important cities to determine infestation and Breteau Indexes, plus the most productive container type across the country [24]. If at least one of these surveys, e.g. the one immediately before dengue transmission starts, has high-quality pupal surveys being conducted in blocks representatives of disease transmission over the city, estimates on vector abundance, survivorship and recruitment rate might be helpful to improve vector control efficiency by directing existing strategies towards areas in which Ae. aegypti population has greater vectorial capacity.
Our models estimating recruitment assume that the population stays constant during the short period of experiment time. If such assumption does not hold due to abundance fluctuations occurring as a result of changing environmental conditions, use of insecticides, or any other, we expect difficulties to get accurate estimations applying this modeling, unless the exogenous conditions can be modeled.
Simulated datasets and analyses consider typical designs used for MRR experiments involving mosquito populations of Aedes aegypti, a known vector of Zika, dengue and chikungunya viruses. However, these models can possibly be applied to other mosquito populations. Laboratory-reared individuals of Aedes aegypti used in previous field studies [8–10] had the genetic background of field mosquitoes. In this case, such designs would not necessarily imply different survival of the released mosquitoes compared to the field mosquitoes. However, daily survival estimates are essential for use of modified mosquitoes such as Wolbachia-carrying mosquitoes as described by Garcia et al. [5].
There is vast literature on MRR experiments to study ecology of wild animal species [25] (and references therein), instead of mosquito populations. Studies with mosquito MRR may benefit from more advanced techniques, including possibility of using covariates such as environmental variables, individual tagging, positional and distance effect, if overcoming important design limitations. For instance, individual marking, multiple sightings and geoposition recording has been done for estimating abundance of mammals [26]. For a few insect populations, individual marking is possible through code systems, such as applying distinct dots to the body [27], for instance by elytra puncture in beetles [28]. Krebs et al. describe density estimation of rodent population in a Canadian area, by live-trapping individuals [29]. More refined models departing from other study designs and including other variables can take elements from capture-recapture designs for populations other than mosquito ones.
Bayesian models permit us to include all parameters instead of serial parameter estimation and to use prior beliefs, if any, or vague priors in order to obtain not only mean estimations but also credibility intervals. Traditional methods require a sequence of estimations for survival and abundance, and if possible recruitment, from observed field data. Smith and McKenzie [30] demonstrated the impact on vector control strategies relying on each of the parameters of the basic reproduction number for the Ross-Macdonald model [1] for malaria. More recently the classical models for malaria have been revisited to study sensitivity in applying strategies for disease control [31,32]. Models for transmissibility of other vector-borne diseases such as Zika, dengue, and chikungunya viruses can also benefit from sensitivity analysis, if using estimated parameters describing the interaction of these pathogens and their vectors. Bayesian models reveal uncertainties that coupled with sensitivities model greatly enhances estimation of vectorial capacities. Advancing towards Bayesian models that encompass a whole set of parameters greatly enhances understanding not only of the underlying dynamics but also on sensitivity to each of the biological aspects. Such models are useful to advance on strategies of vector control that aim at reducing the vectorial capacity of mosquito populations.
|
10.1371/journal.pntd.0004887 | Development of 2, 7-Diamino-1, 8-Naphthyridine (DANP) Anchored Hairpin Primers for RT-PCR Detection of Chikungunya Virus Infection | A molecular diagnostic platform with DANP-anchored hairpin primer was developed and evaluated for the rapid and cost-effective detection of Chikungunya virus (CHIKV) with high sensitivity and specificity. The molecule 2, 7-diamino-1, 8-naphthyridine (DANP) binds to a cytosine-bulge and emits fluorescence at 450 nm when it is excited by 400 nm light. Thus, by measuring the decline in fluorescence emitted from DANP—primer complexes after PCR reaction, we could monitor the PCR progress. By adapting this property of DANP, we have previously developed the first generation DANP-coupled hairpin RT-PCR assay. In the current study, we improved the assay performance by conjugating the DANP molecule covalently onto the hairpin primer to fix the DANP/primer ratio at 1:1; and adjusting the excitation emission wavelength to 365/430 nm to minimize the background signal and a ‘turn-on’ system is achieved. After optimizing the PCR cycle number to 30, we not only shortened the total assay turnaround time to 60 minutes, but also further reduced the background fluorescence. The detection limit of our assay was 0.001 PFU per reaction. The DANP-anchored hairpin primer, targeting nsP2 gene of CHIKV genome, is highly specific to CHIKV, having no cross-reactivity to a panel of other RNA viruses tested. In conclusion, we report here a molecular diagnostic assay that is sensitive, specific, rapid and cost effective for CHIKV detection and can be performed where no real time PCR instrumentation is required. Our results from patient samples indicated 93.62% sensitivity and 100% specificity of this method, ensuring that it can be a useful tool for rapid detection of CHIKV for outbreaks in many parts of the world.
| Chikungunya has reemerged as an important mosquito-borne infection with global health significance. Rapid diagnosis plays an important role in early clinical management of patients due to lack of a vaccine and effective treatment. Laboratory diagnosis is generally accomplished by blood tests to detect virus-specific antibodies but these antibodies are usually developed one week after infection, which misses the window of effective clinical management. On the other hand, although detecting the viral genome can be done in early stage of infection by real-time polymerase chain reaction (PCR) but it is costly to the patients. Here we utilized a fluorescent compound to improve the cost-efficiency of the molecular assay for diagnosis of Chikungunya virus infection. By testing on 77 serum samples, this improved assay has proven to be highly sensitive and specific towards Chikungunya virus. We believe that this research could benefit both clinicians and patients by providing early and accurate diagnosis.
| Chikungunya virus (CHIKV) is an arthropod-borne virus transmitted to humans primarily via the bite of an infected [1] Aedes agypti and Aedes albopictus mosquito. [2, 3] Currently, there are more than 40 countries including Africa, United States, European countries and Southeast Asian countries affected by chikungunya fever. [2] CHIKV is an enveloped positive-sense single stranded RNA virus belonging to Alphavirus genus of Togaviridae family. [4] The genome is approximately 11.8 Kb long, encoding four non-structural proteins (nsP1, nsP2, nsP3, nsP4) and five structural proteins (C, E3, E2, 6K and E1). [5] The clinical symptoms of chikungunya fever are similar to that of dengue fever which is caused by Dengue virus (DENV), an arthropod virus belonging to Flaviviriridae family transmitted by same vectors as CHIKV. [6] This may result in cases of misdiagnosis in places where both viruses co-exist.
As there is no vaccine or specific therapeutic agent available for CHIKV infection, early diagnosis of CHIKV is crucial in preventing the collapse of health care system due to unprecedented number of cases usually encountered during CHIKV epidemics. [7] Virus isolation is classified as the gold standard in detection of CHIKV despite being a time-consuming process requiring 1–2 weeks to determine the presence of virus. The limitations associated with virus isolation resulted in the development of serological and molecular diagnostic methods that are rapid and less labour intensive. Enzyme-linked-immunosorbent assay (ELISA) and Immunochromatographic test (ICT) are examples of serological diagnostic assays which detect IgM and/or IgG antibodies that are specific to CHIKV present in patient sera. ELISA and ICT tests are inexpensive and easy to perform as they do not require handling live viruses. A four-fold increase in antibodies by comparing acute phase and convalescent phase serum samples is usually required to confirm CHIKV infection. IgM is detected on an average of two days after infection and persists for several weeks to three months, while IgG is detected in convalescent samples and may persist for years. [8] The outcome of having antibodies present in serum samples after recovery phase may deduce as false-positive detection. Blacksell and co-workers reported that commercially available antibody-based assays are not suitable for acute diagnosis of CHIKV as the results obtained showed ICT and ELISA kits having sensitivity of 1.9–3.9% and 3.9% respectively. [9] An alternative serological method of anti-CHIKV antibody detection has been reported to be used in commercial ELISA kits, but has shown cross-reactivity with other alphaviruses such as Ross River and O’ nyong-yong viruses as they are closely related serologically. [9] Thus, serological methods for CHIKV detection have been inefficient for acute phase diagnosis. [9–11]
Recently, molecular diagnosis has been well established for rapid, highly sensitive and specific detection of CHIKV infection during the acute phase. Viral RNA is extracted from serum samples collected 1–7 days post-infection [8] were detected by primers targeting the conserved regions of Chikungunya genome specifically. In comparison, conventional RT-PCR appears to be a less sensitive and relatively more time-consuming process than TaqMan and SYBR Green I-based real-time RT-PCR assays. However, real-time RT-PCR assays require highly sophisticated instruments with yearly maintenance and calibration, restricting the utilization of such assays in places with poor financial and technical resources. [12]
Previously, we have reported a novel diagnostic assay for CHIKV detection by adapting hairpin primers and fluorescent molecule, 2, 7-Diamino-1,8-naphthyridine (DANP), into a conventional PCR procedure. [13] In brief, DANP molecule contains a naphthyridine ring which enables it to bind specifically to a cytosine-bulge in a hairpin structure of the PCR primer by hydrogen bonds. [13] The binding of DANP molecule to DNA gives rise to a 400 nm excitation and 450 nm emission property to the bound DANP molecule. As PCR proceeds, the primer is incorporated into double stranded DNA and the hairpin is opened, causing the release of DNAP molecule and thereby decreasing the fluorescence intensity. [13] The utilization of DANP coupled hairpin PCR has also been demonstrated in a single-nucleotide polymorphism study of the cytochrome P450 gene 2C9*3 by Takei and colleagues. [14] However, the binding of DANP molecule to the hairpin-primer is in an equilibrium manner, so that excess DANP molecules must be added to ensure a detectable fluorescence intensity. Therefore, the background signal given off by unbound DANP molecules limits the sensitivity and consistency of the assay.
In the present study, DANP molecule was covalently immobilized on the hairpin PCR primers containing C-G base-pairs directly after the C-bulge to quench the fluorescence emission, as shown in Fig 1A. As PCR progresses, the hairpin structure is opened up and the DANP molecule is moved to the outer surface of the double-stranded DNA molecule, away from cytosine-bulge, resulting in an increase in fluorescence emission at 430 nm when it is subjected to UV-light at 365 nm. Increments in fluorescence intensity can be picked up only if the viral RNA template is present in the reaction with negligible background signal as no excess DANP molecules were added to the reaction. The method is highly effective as it uses a conventional RT-PCR protocol followed by measurement of fluorescence signal intensity using a spectrophotometer. The assay is more rapid and cost-effective as compared to real-time PCR methods. The assay was also validated with CHIKV infected patient serum samples and healthy individual serum samples for its sensitivity and specificity.
CHIKV (GenBank accession No. FJ445502) was isolated from an infected patient during the CHIKV outbreak in Singapore in 2008. The virus was propagated in Aedes albopictus C6/36 cells. Briefly, cells were grown to about 80% confluency in T75 tissue culture flasks. Following removal of the growth media, virus inoculum was added to give a multiplicity of infection (MOI) of 0.1 PFU/cell. Flasks were incubated at 28°C for 1 hours with constant agitation at every 15 min interval. After the incubation, Rosewell Park Memorial Institute (RPMI) 1640 growth medium (Sigma-Aldrich Corp) supplemented with 2% FBS (Hyclone) was added and flasks were maintained at 28°C for about 3–5 days or until cells showed 80% cytophatic effects (CPE). The viral titers were determined by plaque forming assay. [13] Ross River virus (RRV), Sindbis virus (SINV), Kunjin virus (KUNV, MRM 61C strain), West Nile virus (WNV, Sarafend strain), Zika virus (ZIKV, MR 766 strain), DENV-1 (S144 strain), DENV-2 (New Guinea C strain), DENV-3 (Eden 130/05 strain), DENV-4 (S8976 strain), Influenza A virus subtype H1N1, H3N2, Poliovirus type 1 (PV1, Sabin strain), type 2 (PV2, Sabin strain), type 3 (PV3, Sabin strain), Human enterovirus 71 (HEV71, AF316321 strain), Coxsackie B2 virus (CB2), Coxsackie A16 virus (CA16, WHO strain) and Enteric cytopathic human orphan virus 7 (Echo7) were also used to examine the cross-reactivity of this assay. The ZIKV, DENV1-4, PV1-3, HEV71, CB2, CA16 and Echo7 viruses were maintained in the laboratory. The RRV, KUNV and WNV were kindly provided by Professor Mary Mah-Lee Ng, Department of Microbiology, National University of Singapore. The Influenza A viruses were kindly provided by Associate Professor Tan Yee Joo, Department of Microbiology, National University of Singapore.
A set of 22 serum samples from CHIKV-infected patients, and 30 from uninfected individuals were collected at the National University Hospital, Singapore, with informed consent, to evaluate the clinical sensitivity and specificity of the DANP-anchored assay. All of the sera were confirmed as febrile illness associated with a positive result from the real-time RT-PCR. [15] This part of the study was performed in accordance with the National University of Singapore Institutional Review Board approved protocol (No. 10–234). Environmental Health Institute (EHI), National Environmental Agency of Singapore kindly provided a set of 25 serum samples from clinically-suspected patients in which the presence of CHIKV was confirmed by a real-time RT-PCR assay. [16] Written informed consent was given for all samples involved in this study.
CHIKV RNA was extracted from 140 μL of infected cell culture supernatants (3.6 X 10^7 PFU/mL) and serum samples using the QIAamp viral RNA mini kit (Qiagen, Hilden, Germany) according to the manufacturer’s instructions. The RNA was eluted in a final volume of 50 μL of nuclease-free water and stored at −80°C until use.
The full genomes of multiple geographically different strains of CHIKV from recent outbreaks were retrieved from GenBank and aligned using the ClustalX (version 2.1) [17] sequence alignment software. Primers were designed to target the highly-conserved nsP2 regions of CHIKV genome, as shown in Table 1. The primers were designed with hairpin (underlined sequences in Table 1) at the 5’ end to accommodate DANP molecule which is covalently conjugated to the thymine nucleotide (bolded sequences in Table 1) of the primers.
RT-PCR reactions were performed in C1000 thermal cycler (Bio-Rad, Hercules, CA). Reactions were optimized with a One Step RT-PCR kit (Biotech Rabbit, Hannover, Germany). Each reaction was performed in 25 μL total reaction mixture containing 12.5 μL of 2x reaction buffer, 0.2 μmol/L of each forward and reverse DANP hairpin primers, 1.25 μL of 20x RT-RI blend (reverse transcriptase and RNAse Inhibitor) and 1 μL of viral RNA. 10 μL from each total reaction volume was set aside while the remaining 15 uL of reaction was subjected to RT-PCR. The thermal profile was optimized as follows; reverse transcription step at 45°C for 20 minutes, activation of Taq polymerase at 95°C for 2 minutes, followed by 30 cycles of PCR cycling steps consisting of 95°C for 10 seconds, 60°C for 10 seconds and 72°C for 15 seconds.
In order to determine the fluorescence intensity, 10 μL of each reaction was diluted with 90 μL of nuclease-free water in each well of a white opaque flat-bottom 96-well plate. The fluorescence intensity from each well was scanned by Infinite® 200 PRO microplate reader (Tecan Trading AG, Switzerland) with an excitation filter at 365-nm and an emission filter at 430-nm. A sample positive for CHIKV infection was determined as the increment in fluorescence intensity after PCR was more than 100 arbitrary units (AU) as compared to background fluorescence in pre-PCR reaction mixture. For assay validation, all PCR products were analysed using 8% native polyacrylamide gel electrophoresis (PAGE), followed by ethidium bromide staining for two minutes. Gel images were captured using the GeneSnap software version 7.02 (Syngene, Cambridge, UK).
In order to determine the number of PCR cycles that gives off the most significant increment in fluorescence intensity after PCR, the fluorescence intensity level was measured and compared after every five PCR cycles using CHIKV genomic RNA as positive control and nuclease-free water as negative control (NTC).
CHIKV RNA was extracted from 140 μL of infected cell culture supernatants with viral titre of 3.6 X 10^7 PFU/mL, using the QIAamp viral RNA mini kit (Qiagen, Hilden, Germany) according to the manufacturer’s instructions. The viral RNA was eluted in a final volume of 50 μL of nuclease-free water and then serial diluted logarithmically until a final concentration of 1 X 10−5 PFU/μL. 1 μL of each of the serial diluted viral RNA samples was subjected to the DANP-anchored RT-PCR assay to determine the limit of detection of the assay.
The RNA concentration ranges tested were 1.0 X 101 to 1.0 X 10−5 PFU/reaction (3.6 X 103 to 3.6 X 10−3 PFU/mL).
RRV, SINV, KUNV, WNV, ZIKV, DENV1-4, Influenza H1N1, H3N2, PV1-3, HEV71, CB2, CA16 and Echo7 were used to examine the cross-reactivity of this assay. Viral RNA was extracted from 140 μL of each of the viruses and eluted in 50 μL of nuclease-free water. The RNA concentration were measured using NanoDrop ND2000 Spectrophotometer (NanoDrop Technologies, Inc., Wilmington, DE, USA) and only samples with at least 50 ng/μL were proceeded to cross reactivity study of the assay.
In order to determine the number of PCR cycles that gives off the most significant increment in fluorescence intensity after PCR, the fluorescence intensity level was measured and compared after every five PCR cycles using CHIKV genomic RNA as positive control and nuclease-free water as negative control (NTC). As shown in Fig 2, the difference in fluorescence intensity between before PCR and after PCR samples reached the maximum at 30 cycles. Due to the formation of primer dimer and non-specific PCR products in the NTC, the difference in fluorescence intensity began to narrow down after 30 cycles. Therefore, 30 cycles of PCR reaction was used in the rest of the study.
To validate the suitability of hairpin primers and to verify the initial fluorescence intensity level, DANP-anchored hairpin RT-PCR procedure was carried out with and without CHIKV genomic RNA. All PCR products were analyzed by PAGE to determine the assay specificity. As indicated in Fig 3A, the specific PCR product of 296 bps can only be seen when CHIKV RNA is present. There was no significant change in fluorescence intensity between before/after PCR in NTC reactions (Fig 3B). In contrast, an increment of more than 2000 AU of fluorescence intensity was observed after 30 cycles of PCR in the presence of CHIKV RNA.
The detection limit of the DANP-anchored RT-PCR assay was determined through replicates of reactions, using serial logarithmic dilutions of the control CHIKV genomic RNA. Fig 4 shows the change in fluorescence intensity before and after 30 cycles of PCR reaction. A statistically significant increase of 120 AU was observed in 0.001 PFU per reaction, the lowest level of detection by the assay.
The cross-reactivity of the assay was determined by using a panel of other RNA viruses. RRV and SINV were used as representative members of Alphavirus; KUNV, WNV, ZIKV and DENV-1, DENV-2, DENV-3, and DENV-4 were used as representative members of Flavivirus in the cross-reactivity study. The remaining RNA viruses included H1N1, H3N2, Polio 1, Polio 2, Polio 3, HEV71, CB2, CA16 and ECHO7. Only CHIKV RNA samples demonstrated positive results, indicating the lack of cross-reactivity with other viruses tested (Fig 5).
In order to evaluate the sensitivity and specificity of the present assay for clinical diagnosis, 47 serum samples obtained from patients with confirmed CHIKV infection during the acute phase and 30 serum samples from uninfected individuals were tested. The present assay demonstrated high sensitivity by picking up 44 of the 47 CHIKV cases (93.62% sensitivity; 95% CI, 81.44% to 98.37%). None of the 30 serum samples from uninfected individuals was false diagnosed as positive (100% detection specificity; 95% CI, 85.87% to 100%) (Table 2).
CHIKV has been relatively understudied as it was restricted to Africa and Asia countries. [18] Since 2005, CHIKV started to spread to countries in Indian Ocean and then globally. [19] It is estimated that > 1.5 million people were infected in India during the 2006 outbreak alone, [20] and, currently, Chikungunya fever has been documented in more than 40 countries. [21] The unprecedented worldwide spread of CHIKV was driven by international travel and the A226V mutation on the envelope protein 1, which better adapts the virus to Aedes albopictus. [3] Due to increased globalization and mosquito vectors expand to new areas, early diagnosis of CHIKV is critical in the absence of any licensed antiviral therapy and prophylaxis, especially in developing countries.
Currently, the diagnosis of CHIKV largely relies on virus isolation, detection of specific antibody and nucleic acid. Virus isolation in tissue culture is time-consuming and technically complex that is limited in developing countries. Because of extensive cross-reaction between alphaviruses due to common antigens, serological assays often face the difficulty in differentiating commonly occurring alphaviruses. These drawbacks have made molecular assays the method of choice for diagnosis during acute phase of chikungunya fever. Molecular techniques based on the detection of genomic sequences by RT-PCR, nested RT-PCR, and real-time RT-PCR are rapid and sensitive and have replaced virus isolation as the new standard method for the detection of CHIKV in acute-phase serum samples, but the reagents and equipment are too costly for widespread use. In this regard, this DANP-anchored RT-PCR assay reported in this study is advantageous, because of its simplicity, rapidity, and cost-effectiveness. Only a standard conventional PCR procedure, with DANP hairpin primer used, and a fluorescence reading procedure is required without involving of sophisticated instrument or costly reagent.
In comparison, previously, we have reported a novel DANP-coupled hairpin RT-PCR for rapid detection of CHIKV in the acute phase serum samples. PCR primers were designed specifically to target nsP2 gene of CHIKV with hairpin tag containing a cytosine-bulge. The DANP molecule binds to the C-bulge in its protonated form (DANPH+) before PCR reaction starts, resulting in fluorescence emission. During PCR amplification, the hairpin primer opens up and releases the DANP molecule resulting in a drop in fluorescence emission giving rise to a ‘turn-off’ system. CHIKV positive samples are determined by comparing the fluorescence intensity recorded before and after PCR process by subjecting the PCR products to UV-light and detecting the emitted fluorescence at 430 nm. Despite it was a rapid, sensitive, specific and cost effective assay; the optimization of DANP concentration due to background signal restricted its usage. In order to overcome the issue, in the present study, we covalently conjugated the DANP molecule onto the hairpin structure of the PCR primer. Therefore, the ratio between DANP molecule and primer is fixed at 1:1 and this standardization simplifies the optimization of the assay. In addition, by changing the reading spectrum from 400 nm excitation and 450 nm emission to 365 nm excitation and 430 nm emission, we managed not only to minimize the background signal but also give rise to a ‘turn-on’ system. In addition, we have also shortened the assay turnaround time from 90 minutes to 60 minutes including fluorescence reading, by cutting down the reverse transcription step duration and optimizing the PCR cycle number from 40 to 30.
The detection limit of the assay was 0.001 PFU per reaction that is lower than that of the previous DANP coupled assay [13] and is comparable to real time RT-PCR assays developed by other groups [22, 23]. A side-by-side comparison of our assay with the abTES DEN 5 qPCR I Kit (Cat: 300152) from AIT biotech, a Taqman probe-based multiplex real time RT-PCR for DENV/CHIKV detection. Comparable limit of detection was noted. Despite the fact that the viremia load is usually above 4 log10 during the acute phase of CHIKV infection, [24] low detection limit of the assay enables us to detect CHIKV RNA even during late acute phase when the viral titers start to decline rapidly. More importantly, the present assay is not cross-reactive with a panel of RNA viruses that are co-circulating in endemic regions. Given that CHIKV is commonly misdiagnosed as DENV and vice versa, the outstanding specificity of our assay could benefit both clinicians and patients at the point-of-care by providing accurate diagnosis.
|
10.1371/journal.pntd.0002583 | Heterogeneities in Leishmania infantum Infection: Using Skin Parasite Burdens to Identify Highly Infectious Dogs | The relationships between heterogeneities in host infection and infectiousness (transmission to arthropod vectors) can provide important insights for disease management. Here, we quantify heterogeneities in Leishmania infantum parasite numbers in reservoir and non-reservoir host populations, and relate this to their infectiousness during natural infection. Tissue parasite number was evaluated as a potential surrogate marker of host transmission potential.
Parasite numbers were measured by qPCR in bone marrow and ear skin biopsies of 82 dogs and 34 crab-eating foxes collected during a longitudinal study in Amazon Brazil, for which previous data was available on infectiousness (by xenodiagnosis) and severity of infection.
Parasite numbers were highly aggregated both between samples and between individuals. In dogs, total parasite abundance and relative numbers in ear skin compared to bone marrow increased with the duration and severity of infection. Infectiousness to the sandfly vector was associated with high parasite numbers; parasite number in skin was the best predictor of being infectious. Crab-eating foxes, which typically present asymptomatic infection and are non-infectious, had parasite numbers comparable to those of non-infectious dogs.
Skin parasite number provides an indirect marker of infectiousness, and could allow targeted control particularly of highly infectious dogs.
| Zoonotic visceral leishmaniasis is a sandfly-borne disease of humans and dogs caused by the intracellular parasite Leishmania infantum. Dogs are the proven reservoir. The disease is usually fatal unless treated, and is of global health significance. Diagnosis of canine infections relies on serum antibody-based tests that measure infection. In some endemic regions, a test-and-slaughter policy of seropositive dogs forms part of the national control policy to reduce human infection. However, this strategy is not considered effective. Since not all infected dogs are infectious to sandfly vectors, one option is to target control at infectious dogs, as only these dogs maintain transmission. We quantify Leishmania numbers in individual host tissues from time of infection using molecular methods. Comparing these results with their infectiousness to sandflies, we also evaluate the performance of molecular and immunological assays to identify infectious animals. Parasite numbers varied substantially between individuals, increasing with duration and severity of disease. Infectiousness to the sandfly vector was associated with high parasite numbers, and parasite loads in the skin was the best predictor of being infectious. The results suggest that molecular quantitation is useful in identifying individuals and populations responsible for maintaining transmission, with potential application in operational control programmes.
| Studies of microparasites usually consider hosts as homogeneous infection units (infected or uninfected), despite knowledge that infections progress through states of clinical severity, that clinical severity is often associated with the number of infecting microorganisms (load), and that individual transmission potential may be related to infection load. The significance of “super-spreaders” responsible for spreading infection to a disproportionate number of secondary cases has long been recognised [1], [2], however the relationships between parasite load and transmission are rarely measured; even in well-studied macroparasites (e.g. helminths) infectiousness is assumed to correspond to worm burden and egg count [3]–[6].
Variations in individual infection loads tend to be characterised by right-skewed (over-dispersed or aggregated) frequency distributions. Over-dispersion translates into diminishing proportions of the host population harbouring disproportionately higher infection loads. Where transmission potential is directly related to infection load, over-dispersed distributions may be interpreted as a small fraction of the population being responsible for most transmission, giving rise to the “20/80 rule” (whereby 20% of cases cause 80% of transmission), proposed for a number of parasitic agents (e.g. [7]–[10]). Heterogeneity in transmission can increase the basic case reproduction number R0 of a pathogen compared to that under assumptions of homogeneous mixing or density-dependent contact networks [9], [11], and affect the effort required, and choice of strategy (mass or targeted), to interrupt transmission [7]–[9], [12]. Molecular techniques, such as real-time quantitative PCR (qPCR), have been used recently to differentiate between infected individuals and to help understand the spread and treatment of emerging infectious diseases e.g. [2], [9], [13]–[15], nevertheless few empirical studies relate individual infection loads to transmission.
Zoonotic visceral leishmaniasis (ZVL) is a fatal disease of humans and canids caused by the protozoan parasite Leishmania infantum, and transmitted between hosts by Phlebotomine sandflies. The domestic dog is the only proven reservoir [16], though severity of infection and infectiousness varies greatly between individuals; in humans and wild mammals the majority of infections are asymptomatic and non-infectious [16]. Control of ZVL focuses on the detection and elimination of infected dogs (particularly in South America), indoor residual spraying of insecticide, and human case treatment [17]. Positivity to serum anti-Leishmania antibodies is the principal criterion for mandatory slaughter of dogs [17]. Analyses indicate that this policy has little impact on reducing ZVL incidence, though robust data are lacking [16], and there have been calls to re-evaluate the ZVL control program in Brazil [16], [18]–[21]. Contributing factors to the lack of effectiveness include delays between testing and slaughter, low test sensitivity [22], and significant dog-owner non-compliance [21]. An alternative strategy could be to target infectious rather than infected dogs, providing infectious hosts can be identified. Direct measurement of infectiousness by xenodiagnosis requires blood-feeding of colony-reared sandflies on hosts followed by screening for parasite infections in the vector. Rearing large quantities of vectors for community surveillance however is not practical. Tissue parasite loads have the potential to provide a reliable indirect marker of infectiousness [23]–[30], though no studies have tested these relationships through the time course of infection.
Here we measure L. infantum loads in cohorts of naturally infected domestic dogs Canis familiaris and crab-eating foxes Cerdocyon thous in Amazon Brazil. This study is unique in being able to relate host tissue parasite loads to serial xenodiagnosis from time of natural infection. The aims were (i) to characterize the heterogeneities in L. infantum loads between sampled tissues and between individual hosts with different severity of infection, (ii) to investigate whether tissue parasite loads can predict infectiousness to the sandfly vector; (iii) to compare parasite loads between dogs and crab-eating foxes, and (iv) to evaluate the performance of qPCR and ELISA diagnostic assays to identify infectious animals in mixed populations.
Canine samples were collected with informed consent from dog owners. Sampling was performed in accordance with UK Home Office guidelines.
Dog samples were available from −80°C archived material generated in a cohort study of naturally exposed dogs between April 1993 and July 1995 in the municipality of Salvaterra, Marajó Island, Pará State, Brazil, in which bone marrows aspirated from the iliac crest and 3 mm skin biopsy punches of the ear pinnae outer edge were sampled repeatedly at approximately 2 month intervals for up to 27 months post initial exposure [31]. Ear skin was the preferred skin sample since it is reported to be more infectious to sandflies than abdomen skin [23], [30]. Both skin and bone marrow are reported to be more sensitive than blood for parasitological and molecular detection of L. infantum, and higher qPCR counts are recorded in bone marrow than in blood [32]–[34]. For the present study, 265 bone marrow samples were available from 82 infected dogs (1–10 samples per dog), and 185 ear skin biopsy samples were available from 64 infected dogs (1–6 samples per dog), of which 173 samples from 63 dogs had paired bone marrow samples. Fox samples were collected during a concurrent longitudinal study of sympatric marked-recaptured free-ranging foxes [35]. Here, 67 bone marrow samples from 34 infected foxes, and 51 ear biopsy samples from 30 infected foxes, were available; all ear biopsy samples had paired bone marrow samples. Dog samples were collected with informed consent from dog owners.
Dog and fox samples were assayed at all, or at the majority, of time-points, for (i) anti-Leishmania IgG by ELISA using crude leishmanial antigen (CLA), with antibody concentrations expressed as arbitrary units/mL relative to a positive control serum [31] (n = 277 samples); (ii) PCR on bone marrow biopsies using primers specific for kinetoplast DNA (kDNA) and ribosomal RNA [36] (n = 277 samples); (iii) rK39 Kalazar Detect Rapid Diagnostic Test (RDT), Inbios International Inc., WA., USA [37], (iv) qPCR primers for kDNA (described below), and (v) clinical score, defined as the sum of the score of six typical clinical signs (alopecia, dermatitis, chancres, conjunctivitis, onychogryphosis, and lymphadenopathy), each scored on a semi-quantitative scale from 0 (absent) to 3 (intense) [36] (n = 266 samples). Animals were assessed for infectiousness to the sandfly vector by xenodiagnosis, using uninfected colony-reared Lutzomyia longipalpis, and following dissection 4–5 days post full engorgement [22], [35]. Here, matching xenodiagnosis data were available for 103 dog bone marrow samples (36 infected dogs, 3,751 fed flies dissected), 58 dogs ear samples (26 infected dogs, 1,702 flies), 39 fox bone marrow samples (22 infected individuals, 1,309 flies), and 30 fox ear samples (18 foxes, 1,187 flies).
DNA was extracted from 100 µL aliquots of bone marrow, using phenol-chloroform [38]. DNA from 3 mm ear skin punch biopsies (average: 0.029 grams, range: 0.0144–0.0837) was extracted using a commercial kit (DNeasy: Qiagen, UK). qPCR was performed using primers specific for a conserved region of Leishmania kDNA [27]. Quantification of Leishmania DNA was performed by comparison of Ct values with those from a standard curve constructed from 10-fold dilutions of L. infantum DNA extracted from cultured parasites, from 1×105 to 0.001 parasite equivalents/mL (strain MHOM/MA/67/ITMAP-263). Samples were tested in duplicate and standards in triplicate on every plate. The occasional duplicates giving one positive and one negative result were re-tested: none remained unresolved after re-testing. A non-template control (NTC) was run in triplicate on every plate. A plate of negative controls including DNA extracted from blood samples of 30 UK dogs with no history of foreign travel, and 40 endemic control dogs from São Paulo, Brazil was run every 5 plates. A standardised Ct threshold value of 0.01 was selected as cut-off value to define infection based on the NTC signal. The endogenous control was a eukaryotic 18S rRNA gene as a reference of total canine DNA quantified in a separate qPCR reaction to the Leishmania assay using pre-developed TaqMan Assay reagents (Applied Biosystems, UK) following the manufacturer's recommendations. Parasite loads were normalized (d) between animals to the eukaryotic 18S rRNA gene per reaction, where d = absolute Leishmania kDNA equivalents/(copy number of 18S rRNA gene/2)/ng tissue DNA extracted measured spectrophotometrically. Normalized log10 parasite numbers and absolute log10 parasite numbers per ml (bone marrow) or per gram (ear skin) were strongly correlated (r2 = 0.93 and r2 = 0.98 respectively). Consequently, for ease of interpretation, we report the per unit absolute log10 parasite numbers.
The date of patent infection for dogs and foxes was estimated as the first date at which animals were positive by any serological or parasitological assay; all samples thereafter were considered as infected based on previous analyses demonstrating a very low incidence of serological reversal [31], [35], [36]. At each bimonthly examination, dogs were classified according to their total clinical score as asymptomatic (scores 0–2), oligosymptomatic (3–6) and symptomatic (>6). Dogs with >8 months post infection follow-up and all bimonthly clinical scores <3 were considered long-term asymptomatic. Infectiousness was assessed as either positive (≥1 sandfly infected) or negative, or as the proportion of sandflies infected at any single time point (point xenodiagnosis). Dogs were also classified previously [22], [35] as “highly infectious” (>20% of total flies infected), “mildly infectious” (>0% and <20% flies infected), and “non-infectious” (no flies infected) by serial xenodiagnoses (n = 6,002 flies dissected from 173 independent trials): the highly infectious group were shown to be responsible for >80% of all transmission events [22]. All foxes were non-infectious (n = 1,469 flies from 44 trials) [35].
Parasite aggregation was characterised by the dispersion coefficient k of the fitted negative binomial distribution. Negative binomial models were used to test for differences in parasite loads between groups. Analysis of parasite loads against independent variables were conducted using negative binomial mixed models, with animal identity included as the random effect. The relationship between infectiousness and markers of infection was analysed by logistic regression.
Receiver Operating Curves (ROC) were used to identify parasite load (qPCR) and anti-Leishmania antibody (ELISA) threshold values that maximised test sensitivity and specificity to differentiate currently infectiousness and non-infectious dogs. Areas under the ROCs were similar: 0.937 (ear biopsies, n = 58), 0.837 (bone marrows, n = 103) and 0.846 (ELISA, n = 173) (χ2 = 72.0, df = 2, P = 0.699, n = 52), providing test threshold values of 4.64 log10 parasites/gram (ear biopsies), 3.51 log10 parasites/mL (bone marrows), and 4.59 log10 antibody units/mL, respectively. These values were then used to evaluate the performance of threshold-based qPCR and ELISA assays to detect dogs classified by longitudinal infectious status in the mixed population. The average times of detection by the threshold-based assays relative to infection were calculated using Kaplan-Meier survival analysis. Differences in Kaplan-Meier curves were compared by log rank test, and confidence limits calculated following [39]. All analyses were carried out in Stata v.11.1 (Stata Corporation, College Station, Texas, USA).
Parasite loads were quantified by qPCR in 265 post-infection bone marrow samples from 82 dogs, and 185 post-infection ear skin biopsies from 64 dogs (Table 1). The median parasite loads were 142 parasites/mL in bone marrow and 119 parasites/gram in ear skin (Table 1) but the correlation was not strong (Spearman's ρ = 0.56, P<0.001). Note that since the unit of measurement of these two samples differ, the magnitude of the parasite loads in skin and bone marrow were not directly compared. The frequency distributions of parasite loads in both tissues was highly skewed, with maximum burdens of 2.4×106 parasites/mL and 1.3×108 parasites/gram in bone marrow and ear skin, respectively (Figure 1). The degree of parasite aggregation, measured by the negative binomial parameter k, was very high, with loads in ear skin (k = 0.066) showing greater aggregation than those in bone marrow (k = 0.104). Comparable degree of aggregation was observed for mean parasite loads in individual dogs (Table 1). Of the total L. infantum loads recorded in bone marrows biopsies, 90% of parasites were found in 8% (21/265) of samples and 16% (13/82) of dogs; for skin biopsies, the equivalent figures were 8% (14/185) of samples and 9% (6/64) of dogs.
Parasite loads in both tissues increased on average with time since infection (Table 2; Figure 2). Ear skin loads increased at a faster average rate than bone marrow loads, reflected in the ear skin to bone marrow parasite load ratios being significantly greater in later infection (Table 2). However, the relationship between parasite load and time varied between individual dogs, showing positive to negative slopes for both tissues (Figure 3). Both bone marrow and ear skin loads were significantly higher in sick dogs, in infectious dogs and in dogs with higher anti-Leishmania antibody levels (Table 2). Severity of infection was also associated with greater ear skin to bone marrow parasite ratios (Table 2). However, in symptomatic dogs this ratio did not vary according to the type of symptom: dogs with skin symptoms had comparable ratios to those with only non-skin symptoms (IRR = 0.67 (95% CL 0.28–1.62), χ2 = 0.79, P = 0.37).
The probability of a dog being infectious to sandflies at point xenodiagnosis was positively associated with parasite load, PCR status, IgG antibody titer, total clinical score, and time since infection; the strongest predictor of being infectious was ear skin parasite load (Table 3); similar results were seen when analysis was restricted to only paired bone marrow and ear skin samples (data not shown). Infectivity to sandflies was associated with high parasite loads in ear skin (Figure 4): the majority of dogs had loads <106 parasites per gram and were very rarely infectious. Highly infectious dogs had higher mean parasite loads than mildly infectious dogs (ears: Wald χ2 = 7.36, P = 0.0073; marrow: χ2 = 7.21, P = 0.0067), the latter showing greater average loads than non-infectious dogs (ears: χ2 = 13.35, P = 0.0003; marrows: χ2 = 14.56, P = 0.0001) (Figure 5).
L. infantum was detected in bone marrow of 50% (17/34) and skin of 67% (20/30) of infected foxes. Parasite loads showed similar over-dispersion as for dogs (Table 1). Of the total L. infantum loads recorded in bone marrows, 90% was attributed to 8% (5/67) of samples and to 12% (4/34) of the foxes. The equivalent figures for skin biopsies were 8% (4/53) of samples and 10% (3/30) of foxes.
Bone marrow loads varied significantly with fox age, rising rapidly in the first 6 months of life (age (months): IRR = 1.25 (95% CL 1.11–1.42), P = 0.0004) and declining thereafter (months2: IRR = 0.997 (0.995–0.999), P = 0.0015); a similar, though not significant, pattern was seen for ear skin samples (P = 0.25) (Supplementary Figure S1). No parasites were detected in 15 bone marrow samples from 6 foxes over 6 years old, whereas 4/6 of these foxes (4/12 samples) showed residual parasites in ear skin. In contrast, anti-Leishmania IgG titres did not decline in older age classes (Supplementary Figure S1). There were significant positive relationships between fox tissue parasite numbers and anti-Leishmania IgG titres (marrow Wald χ2 = 16.0, df = 1, P = 0.0001; skin Wald χ2 = 5.68, df = 1, P = 0.017), and ear skin to bone marrow parasite ratios were moderately higher in foxes with high titres (Wald χ2 = 3.81, df = 1, P = 0.051). Only one fox showed any clinical signs of disease (alopecia) but which was mild and transitory.
Skin and bone marrow parasite loads of foxes were similar to those in non-infectious dogs (P>0.10) (Figure 5). Seven long-term “truly” asymptomatic infected dogs were identified: they transmitted infection to 1/678 sandflies exposed in 24 xenodiagnosis trials on 4 dogs. Their parasite loads were similar to those in foxes (P>0.18), which were all asymptomatic by the same definition (Figure 5). None of the 22 infected foxes tested were infectious in 39 xenodiagnosis trials. Applying the model coefficients from analysis of dog infectivity (Table 2) to fox ear skin parasite data (n = 53), foxes were predicted to have been infectious with ≥15% probability (≥104.64 parasites/gram in skin) on 6 of 53 occasions for 4 foxes, equivalent to a total predicted number of infectious samples of 2.9 of 53, compared to the observed 0/39 xenopositive trials of infected foxes.
The performances of qPCR and ELISA to differentiate dogs of different infectious status in the mixed population were tested using positivity threshold values calculated by ROC analysis of the point xenodiagnosis data (see Methods). PCR-based diagnostic tests showed a high sensitivity (94–100%) to detect highly infectious dogs, though the sensitivities of serology-based tests were somewhat lower (78–100%) (Table 4). The sensitivities of most tests to detect mildly infectious dogs were lower, but these dogs contributed <20% of transmission. Only tests based on qPCR thresholds showed high specificities for infectious dogs (i.e. low sensitivities to detect non-infectious dogs) (Table 4). Highly infectious dogs were detected by qPCR significantly earlier after patent infection (152 days [95% CI: 117–186]) than either mildly infectious dogs (442 days [302–582]) or non-infectious dogs (435 days [317–553]) (log rank tests: qPCR: χ2>17.3, P<0.0003); estimates for the latter two groups were statistically indistinguishable (P = 0.70). Detection time of highly infectious dogs approximated their observed time to becoming infectious (134 days [68–201]).
We demonstrate pronounced heterogeneity in L. infantum loads between dogs, assessed by qPCR in bone marrow and ear skin. Loads were highly over-dispersed with evidence of greater aggregation in ear skin relative to bone marrow (9% vs 16% of dogs harboured 90% of total parasites). Parasite loads in the two tissues showed different dynamics: bone marrow loads increased rapidly reaching a peak 100–200 days after infection, while ear skin loads continued to increase over a 600 day period, resulting in increased skin to bone marrow load ratios in late infection. Dissemination to the skin varied between dogs, being greater in sick and infectious dogs. Evidence of L. infantum parasite over-dispersion has been reported in different dog tissues [27], [40]–[42] and in human blood [15], [26], [43], and greater variation in parasite loads in ear skin compared to paired bone marrows, lymph nodes, blood, and liver and spleen samples has been reported for Brazilian dogs [40], [42]. However these studies did not evaluate parasite loads through time. One cohort study of Italian dogs noted a decrease in ear to lymph node parasite ratios during clinical development, in apparent contrast to results here. In that study, the time of infection was not established, so dogs may have been at a different stage and severity of infection [34].
Tissue parasite load, particularly in ear skin, was the best predictor of being currently infectious to vectors. L. infantum amastigotes in skin tissue or skin capillaries are directly accessible to sandflies, which are known to feed abundantly on ear pinnae; and ear skin appears to be more infective than abdomen skin [23], [30]. Some of the variation in parasite loads between ear tissue samples may also reflect small scale spatial variation in parasite density within the ear. We did not restrict sandflies to feed only on ears, unlike other studies [23], [29], [44]. However, the proportion of dogs that were infectious was substantially lower than the proportion with detectable skin parasites, and only dogs with very high skin parasite loads were consistently infectious. Highly infectious dogs showed greater average loads compared to mildly infectious and non-infectious dogs, and also tended to fall within the top 20% parasite loads for each tissue. These data, and the observed high degree of parasite aggregation in ear skin, suggest that the majority of transmission events to vectors result from a small proportion of infectious dogs. Previously we reported for these dogs that 7 of 42 infectious dogs (17%) were responsible for >80% of all sandfly infections [22]. Similar over-dispersion in infectiousness can be calculated from published xenodiagnosis studies, with 15% to 44% of dogs accounting for >80% of transmission events [23], [44]. qPCR studies of canid tissue L. infantum loads relative to xenodiagnoses are not available elsewhere, but parasite estimates by immunohistochemistry of ear skin show moderate correlations with xenodiagnosis positivity [30], [45]. Our current results suggest that high parasite loads in dog ear skin, rather than the simple presence of parasites, is the important metric to identify likely infectious individuals and potential reservoir populations. In the current study, all infections were shown to be L. infantum [36]. To identify super-spreaders in regions of mixed Leishmania co-infections, the specificity of qPCR methods would need to be fully validated.
Current ZVL control strategy in Brazil includes mass test-and-slaughter of Leishmania antibody positive dogs [17], which is criticised on theoretical, logistical and also on ethical grounds [18]–[22]. If the small fraction of dogs that are responsible for the majority of transmission could be identified (e.g. by detection of high parasite loads) and targeted, this would directly address many of these issues, and may be more cost-effective than mass interventions [9], [12]. Canine infectiousness to sandflies is known to increase with the severity of disease and high anti-parasite antibody, but sensitive and specific markers of infectiousness have not been identified [22], [23], [29], [30], [46]. Here, we show that adopting quantitative test threshold values based on skin parasite numbers, highly infectious dogs can be distinguished from non-infectious dogs. These tests were highly sensitive for highly infectious dogs, equivalent to detection of 87–94% of sandfly infections in these samples (data not shown), and importantly also showed high specificities (0.83–0.99) to detect non-infectious dogs, unlike conventional tests for infection. Since up to 50% of seropositive dogs may be asymptomatic in a single community survey, such a targeted approach should also raise dog-owner compliance.
The crab-eating fox occurs widely in South America, and is commonly infected with L. infantum [16], [35], [47], and thus often assumed to be a sylvatic reservoir. However, few infected foxes have been shown to infect sandflies [48], [49], and in our cohort study none of the foxes were infectious [35]. Here, we show that fox parasite loads, though heterogeneous, were significantly lower than those of infectious dogs, and similar to non-infectious dogs, providing further evidence that foxes are not likely to be important for maintaining transmission [22], [35]. The results also provide a parasitological explanation for why the foxes here, and probably wild canids more generally, tend to present asymptomatic infections [16], [50], [51]. Relatively low parasite loads were also noted in the truly asymptomatic cohort dogs, as also reported in asymptomatic human infection [26], [28], [52]. Whether asymptomatic human infections with L. donovani is associated with low parasite loads and thus low transmission potential remains speculative, and further studies are needed [16]. Variation in parasite load between individuals of other potential reservoir hosts (e.g. hares in Iberia [53]), and variation in parasite load in skin between different parts of the host, would also be informative.
In conclusion, this study highlights the importance of quantifying heterogeneities in infection loads in relation to transmission potential through prospective studies, underpinning development of novel tools for parasitic disease management. Studies are now needed to confirm the efficacy of diagnostic threshold-based driven actions against transmission, and to develop diagnostic kits, based on the detection of parasite DNA (e.g. isothermal amplification) or parasite antigens, for practical field use.
|
10.1371/journal.ppat.1000606 | Non-Human Primate Model of Kaposi's Sarcoma-Associated Herpesvirus Infection | Since Kaposi's sarcoma-associated herpesvirus (KSHV or human herpesvirus 8) was first identified in Kaposi's sarcoma (KS) lesions of HIV-infected individuals with AIDS, the basic biological understanding of KSHV has progressed remarkably. However, the absence of a proper animal model for KSHV continues to impede direct in vivo studies of viral replication, persistence, and pathogenesis. In response to this need for an animal model of KSHV infection, we have explored whether common marmosets can be experimentally infected with human KSHV. Here, we report the successful zoonotic transmission of KSHV into common marmosets (Callithrix jacchus, Cj), a New World primate. Marmosets infected with recombinant KSHV rapidly seroconverted and maintained a vigorous anti-KSHV antibody response. KSHV DNA and latent nuclear antigen (LANA) were readily detected in the peripheral blood mononuclear cells (PBMCs) and various tissues of infected marmosets. Remarkably, one orally infected marmoset developed a KS-like skin lesion with the characteristic infiltration of leukocytes by spindle cells positive for KSHV DNA and proteins. These results demonstrate that human KSHV infects common marmosets, establishes an efficient persistent infection, and occasionally leads to a KS-like skin lesion. This is the first animal model to significantly elaborate the important aspects of KSHV infection in humans and will aid in the future design of vaccines against KSHV and anti-viral therapies targeting KSHV coinfected tumor cells.
| Kaposi's sarcoma-associated herpesvirus (KSHV or human herpesvirus 8), the most recently identified human tumor-inducing virus, has been linked to Kaposi's sarcoma, pleural effusion lymphomas and multicentric Castleman's disease. In fact, KSHV accounts for a large proportion of the cancer deaths in Africa. Further, the incidence of KSHV in the US and Europe has greatly increased due to the AIDS pandemic. Despite these pressing human health problems, studies of KSHV infection are greatly hampered by the lack of cell culture and animal models. To address this serious need, we set out to develop an animal model for KSHV infection. In this manuscript, we report the successful zoonotic transmission of KSHV into common marmosets (Callithrix jacchus, Cj), a New World primate. Our study demonstrates that experimental KSHV infection of the common marmoset is highly analogous to its infection of humans, including the means of infection, sustained serological responses, latent infection of PBMCs, virus persistence, and KS-like skin lesion development, although the latter was infrequent in experimental KSHV infections. This model thus provides a unique opportunity to dissect the molecular mechanisms of KSHV infection, persistence, and pathogenesis directly in primates.
| The most recently described human tumor virus, KSHV is a γ-2 herpesvirus and was first identified in association with KS, the most common neoplasm amongst AIDS patients [1]. KS is clinically separated into four different forms: classical KS, endemic KS, iatrogenic KS, and epidemic HIV-associated KS [2]. In addition to KS, KSHV is linked with two other cancers, Primary effusion lymphoma (PEL) [1],[3],[4] and Multicentric Castleman's disease [5],[6],[7]. Both of these cancers are B cell proliferative disorders, generally have poor outcomes, and have short median survival times complicated by their association with AIDS.
One experimental barrier to working with KSHV has been the lack of an in vitro system for examining lytic replication. While KSHV can infect a wide variety of primary cells and cell lines, none support the growth of KSHV to a high titer [8]. Typically, viruses can be stimulated toward replication only through the addition of agents like phorbol esters [9], this limitation extending to the in vivo setting. These problems had previously been addressed in two ways: through manipulation of the virus for increased titer or cell infectivity and the use of highly related viruses. By inserting a gene conferring resistance to an antibiotic, one can select cell populations that are essentially 100% infected [10]. Meanwhile, two examples of related viruses used as stand-ins for KSHV are Herpesvirus saimiri (HVS) [11] and Rhesus rhadinovirus (RRV) [12],[13]. These viruses are largely co-linear with KSHV, carry many of the same genes, and are known to infect non-human primates [13]. RRV infection develops abnormal cellular proliferations characterized as extranodal lymphoma and retroperitoneal fibromatosis, a proliferative mesenchymal proliferative lesion, in an experimentally co-infected rhesus macaque with simian immunodeficiency virus, suggesting an excellent primate model to investigate KSHV-like pathogenesis [14],[15]. In the case of HVS, infection of New World primates results in an aggressive, fulminant lymphoma. However, HVS primarily infects T cells, not B cells, as KSHV does. RRV persists upon infection in rhesus macaques, infects B cells, and induces B cell hyperplasia, but no KS-like disease occurs [15]. On the other hand, murine Herpesvirus 68 (MHV-68) provides a small, experimentally accessible mouse model, but its infection does not associate with KS or related diseases [16]. The introduction of KSHV genes into these systems has proven to be useful, albeit limited, for the study of KSHV [17].
Besides these related virus models, in vitro experiments and transgenic animal models have been the main forces in elucidating the potential roles of individual KSHV proteins in cell culture and mouse models, respectively [18],[19],[20],[21]. In a recent study, SCID-hu Thy/Liv mice reconstituted with the liver and thymus of human fetuses were utilized to study viral transcription as well as the susceptibility of the mice to infection with BCBL-1 derived KSHV [19],[22]. In addition, Parsons et. al have shown that NOD/SCID mice infected with purified KSHV provide a system for demonstrating latent and lytic viral gene expression in addition to cell tropism [19],[22]. Furthermore, they have investigated immune responses to KSHV via implanted NOD/SCID mice reconstituted with human fetal bone, thymus, and skin [19],[22]. In spite of these significant improvements, none of these models truly reflect the in vivo setting. To understand the relative contributions of KSHV proteins to the cellular activation of KSHV-associated diseases and host-viral interactions for viral persistent infection, an animal model that provides a complete viral infection in addition to latent and lytic viral gene expression within the context of an intact host immunity still needs to be developed. In this report, we describe the efficient zoonotic transmission of KSHV into common marmosets (Callithrix jacchus, Cj), a New World primate. Common marmosets intravenously inoculated with recombinant KSHV rapidly seroconverted and maintained high antibody responses for over one and a half years. In addition, KSHV DNA and LANA proteins were readily detectable in PBMCs and various tissues of the infected marmosets at a variety of time points. Furthermore, two common marmosets inoculated with rKSHV.219 by the oral route seroconverted and were positive for viral DNA in their PBMCs. Remarkably, a common marmoset infected with rKSHV.219 via the oral route developed a KS-like lesion with the characteristic spindle cells along with small blood vessels and extravasated erythrocytes. These results demonstrate that human KSHV effectively infects common marmosets, establishes persistence, and occasionally associates with the development of KS-like skin lesions. This is the first animal model of KSHV persistent infection to allow for analyses of the molecular mechanisms of the KSHV lifecycle directly in a non-human primate.
Given its magnitude as a human health problem, it is crucial to understand the molecular details of KSHV biology. However, the lack of an animal model of KSHV infection greatly hampers studies of KSHV pathogenesis and persistence. Therefore, we explored whether common marmosets can be experimentally infected with human KSHV, and if so, to what extent experimental infection recapitulates the important aspects of KSHV infection. To facilitate the infection of common marmosets with KSHV, we chose a recombinant KSHV, rKSHV.219, from KSHV-infected JSC-1 cells [23]. rKSHV.219 expresses red fluorescent protein (RFP) from the KSHV lytic PAN promoter, green fluorescent protein (GFP) from the EF-1α promoter, and contains the gene for puromycin resistance as a selectable marker [10]. Two common marmosets (Cj15-05 and Cj16-05) were inoculated intravenously with 5×106 infectious units (IU) of rKSHV.219. Blood samples were obtained at various time points to measure the marmosets' antibody responses and to detect viral DNA and proteins. Both monkeys quickly seroconverted to anti-KSHV-positive status within 20 days after inoculation (Fig 1A) and the animals' anti-KSHV antibodies persisted at very high levels for over 1.5 years. In addition, the sera from both infected monkeys readily reacted with purified KSHV virion proteins on immunoblots, with higher antibody reactivities detected over time (Fig 1B). A positive control, immunoreactive serum from a KSHV-infected patient was included to validate this immunoblot assay (Supplemental Fig S1). PBMCs from KSHV-infected animals were PCR-positive for KSHV LANA and ORF9 10 and 20 days after infection, respectively (Fig 1C and 1D). Viral DNA persisted in the PBMCs of monkey Cj15-05 for the entire 1.5- year span the animal was studied, whereas the level of viral DNA in monkey Cj16-05 decreased after 200 days postinfection (Fig 1C). Real-time PCR analysis indicated that the KSHV DNA copy number/µg of genomic PBMC DNA from the infected monkeys were substantially lower than that of rKSHV.219-infected Vero cells (Fig 1D). The low viral DNA copy number suggests that a minor population of the monkey PBMCs carried the rKSHV.219 genome. This was confirmed by confocal microscopy which showed that the KSHV LANA protein was detected in monkey PBMCs at a frequency of 2–5 cells per 1×106 cells (Fig 1E). It should be noted that this figure depicts a rare positive field and does not reflect the overall incidence of LANA-positive cells. However, the infection frequency is similar to that of other γ-2 Herpesviruses, such as Rhesus lymphocryptovirus, Rhesus rhadinovirus, and Herpesvirus saimiri, which persist asymptomatically in their natural hosts [11],[12],[15],[24]. Due to the lack of an efficient in vitro culture system for KSHV infection and replication, virus recovery from the PBMCs of the experimentally infected marmosets was unsuccessful. Nevertheless, these results unambiguously demonstrate the persistent infection of naïve common marmosets by rKSHV.219.
Analysis of CD20+ B cells in the blood of the two infected marmosets showed increased B cell populations when compared to naïve marmosets (Fig 2). Although the total number of B cells did not change significantly in the initial few months after infection with rKSHV.219, their numbers noticeably increased around seven months postinfection and remained relatively elevated for as long as we followed these animals. Naïve common marmosets had 10–15% CD20+ B cells in their PBMCs while rKSHV.219-infected marmosets Cj15-05 and Cj16-05 had 15–20% CD20+ B cells (Fig 2 and Supplemental Fig S2). Interestingly, a 7 fold increase in HLA-DR− CD20+ B cells at 200 days postinfection was observed in marmosets infected with rKSHV.219 (Fig 1F) and this population was maintained until these animals were euthanized (data not shown), indicating that rKSHV.219 infection leads to elevated levels of CD20+ B cells in marmosets. However, no B cell hyperplasia was observed in either monkey (data not shown).
To detect early KSHV infection, a marmoset (Cj325-04) was sacrificed 21 days after intravenous inoculation with 5×106 IU of rKSHV.219. An enzyme-linked immunosorbent assay (ELISA) showed a strong anti-KSHV antibody response (Fig 3A) while PCR products from a variety of tissues to test the presence of KSHV LANA DNA were yielded positive results only for samples from the jejunum and liver (Fig 3B). PBMCs were collected at the time of sacrifice and cultured in vitro for one week. Approximately 1/104–5 lymphocytes were GFP-positive and RFP-negative, suggesting latent KSHV infection (Fig 3C). In addition, the KSHV LANA latency-associated protein was readily detected in PBMCs of the rKSHV.219-infected marmoset (Fig 3D). Taken together, these results indicate that KSHV rapidly establishes latent infection in common marmoset PBMCs. It should be noted that the GFP-positive signal from PBMCs of infected marmosets disappeared after an additional week of incubation in vitro (data not shown). Due to a low frequency of GFP-positive cells, however, it is unclear whether the GFP-positive cells died or lost the viral genome after a week of incubation in vitro. This suggests that while GFP is a convenient marker for viral infection, its use may be limited to the early phase of KSHV infection in marmosets.
In transplant recipients, immune suppression is thought to disturb the host's surveillance of KSHV, leading to viral reactivation and an increased systemic viral load [2]. In contrast, rapamycin, an immunosuppressive drug, reduces KSHV-infected PEL cell growth in culture [25] and inhibits the progression of dermal KS in kidney transplant recipients while providing effective immunosuppression [26]. To investigate the direct impact in vivo of an immunosuppressive agent on persistent KSHV infection, two common marmosets (Cj333-04 and Cj139-04) were treated with FK506 (Tacrolimus or Fujimycin, 100 µg/kg/day) immunosuppressive drug [27] for 14 days prior to intravenous inoculation with 5×106 IU of rKSHV.219, with treatment continuing for an additional 100 days. Immune-suppressed monkeys inoculated with rKSHV.219 seroconverted to anti-KSHV-positive status within 14 days after inoculation (Fig 4A). Although the anti-KSHV response and KSHV DNA remained present 100 days after infection, they were of a considerably lower magnitude than those of immune-competent monkeys. Compared to untreated animals Cj15-05 and Cj16-05, FK506-treated monkeys Cj333-04 and Cj139-04 showed anti-KSHV antibody responses approximately 2–3 fold lower throughout at 40–80 days postinfection, a shorter duration of time when a positive KSHV LANA DNA signal could be obtained by PCR, and substantially lower copy numbers (Fig 4B). To assess the tissue distribution of rKSHV.219 in infected marmosets, KSHV LANA-specific DNA was amplified from various tissues and organs of immune-competent and immune-suppressed common marmosets. rKSHV.219 DNA was readily detected in numerous tissues of infected monkeys Cj15-05 and Cj16-05, including the tonsils, tongue, lymph nodes, spleen, jejunum, lungs, colon, liver, thymus, submandibular salivary gland, inguinal skin, and bone marrow (Fig 5). In contrast, only the submandibular lymph nodes and bone marrow were positive for rKSHV.219 DNA in FK506-treated animals Cj333-04 and Cj139-04 (Fig 5). These results indicate that immunosuppressive drug treatment leads to a significant reduction of persistent KSHV infection in vivo.
Two common marmosets were orally inoculated with 5×107 IU of rKSHV.219. Both monkeys inoculated with rKSHV.219 quickly seroconverted to anti-KSHV-positive status within 20 days after inoculation (Fig 6A). However, the anti-KSHV response was much lower in orally infected marmosets than in intravenously infected marmosets and persisted for a shorter period of time (Fig 6A). PBMCs from KSHV-infected marmosets Cj10-05 and Cj11-05 collected on day 41 postinfection were positive for the KSHV LANA sequence but we could not detect KSHV LANA DNA in the PBMCs at any other time under the same conditions (Fig 6B). Like the intravenously infected marmosets, the KSHV LANA protein was also detectable in the PBMCs of orally infected marmosets (Fig 6C). However, since the efficiency of oral infection was much lower than that of intravenous infections, the detection frequency and time period of LANA positivity via confocal microscopy was much lower and more limited with oral infection compared to that in intravenous infection. These results indicate that KSHV infects common marmosets through the oral route but as seen with other viruses [24], oral infection of KSHV is not as efficient as intravenous infection in common marmosets.
Remarkably, one (Cj10-05) of two animals orally infected with KSHV developed a skin lesion on its ventral abdomen at 41 days postinfection (Fig 7A). This purple skin lesion was approximately 1.5 cm in diameter and had similar histopathological features to those observed in AIDS-associated KS lesions. Histological examination of a biopsy revealed a nonencapsulated dermal mass with characteristics typical of KS lesions. Pleomorphic spindle cells arranged in short bundles were observed, along with small blood vessels and extravasated erythrocytes (Fig 7B and C). The spindle cells had a high mitotic index and had infiltrated into the surrounding soft tissues (Fig 7B and C). Additionally, this KS-like lesion of Cj10-05 was PCR-positive for the KSHV LANA sequence (Fig 7D). Immunohistochemistry with antibodies against LANA, vIL-6, and K8.1 detected KSHV LANA, vIL-6, and K8.1, respectively, in the skin lesion (Fig 7E and F and Supplemental Fig S3). The anti-KSHV LANA, anti-vIL-6, and anti-K8.1 reactivities were specific, and no staining was observed with these antibodies in control skin tissues from naïve common marmosets or in the biopsy tissue of Cj10-05 when the antibodies were replaced with isotype-matched irrelevant antibodies (data not shown and Supplemental Fig S3). Moreover, immunohistochemistry showed virtually identical phenotypes for the skin lesion on Cj10-05 and human KS lesions (table in Fig 7 and Supplemental Fig S4). These results collectively demonstrate that a common marmoset orally infected with KSHV can develop a skin lesion with similar histopathological features to those seen in human KS lesions and is also positive for KSHV DNA and proteins. However, the intensity and prevalence of LANA positive staining within the KS-like lesion of Cj10-05 were not as strong or widespread as those of human KS lesions. In addition, the expression of vIL-6 and K8.1 was restricted to a few cells in the KS-like lesion of Cj10-05, the question of whether these cells were latently infected or lytically replicated requiring further study.
Previous studies have illustrated the persistent infection of rhesus macaque monkeys infected with RRV, a primate homolog of KSHV [15],[28]. Macaques inoculated with RRV alone display transient viremia followed by a vigorous anti-RRV response with no specific clinical features. In contrast, experimental RRV infection of SIV-infected rhesus macaques induces some of the hyperplastic B cell lymphoproliferative diseases which manifest themselves in AIDS patients coinfected with KSHV [15]. Furthermore, cotton-top tamarins (Saguinus oedipus) inoculated with human Epstein-Barr virus (EBV) develop diffuse malignant lymphomas resembling human reticulum cell or immunoblastic sarcomas [29],[30]. Additionally, Rhesus lymphocryptovirus, which is very similar to EBV and naturally endemic in rhesus monkeys, can efficiently infect naïve animals orally, with the resulting infection closely mimicking key aspects of human EBV infection [31]. Our study demonstrates that experimental KSHV infection of the common marmoset is highly analogous to its infection of humans, including the means of infection, atypical lymphocytosis, sustained serological responses, latent infection of PBMCs, and virus persistence. However, it should be noted while one of two KSHV-infected marmosets developed KS-like lesion, the number of marmosets used for this experiment is too low to reach the specific conclusion of the frequency of KS-like lesion development induced by KSHV infection. Additional extensive experiments with different conditions such as various KSHV strains and titers and co-infection with HIV-1 or EBV may be able to increase the incidence of KS development. Nevertheless, this model thus provides a unique opportunity to dissect the molecular mechanisms of KSHV infection, persistence, and pathogenesis directly in primates.
We have found that FK506 immunosuppressive drug treatment leads to a significant reduction of persistent KSHV infection in vivo. Zoeteweij et al demonstrated that cyclosporine and FK506, specific inhibitors of calcineurin-dependent signal transduction, effectively block the in vitro KSHV reactivation induced by ionomycin and thapsigargin, activators of intracellular calcium mobilization [27]. This indicates that FK506 immunosuppressive drug treatment may directly block KSHV reactivation in vivo or suppress lymphocyte activation at an early stage of infection, indirectly affecting KSHV persistent infection. Additional experiments, e.g. the timing and type of immune suppression, are necessary to determine the role of host immune competence in the establishment of KSHV persistent infection.
The examination of multiple primate species have demonstrated that γ-herpesviruses are nearly ubiquitous, with homologs of each virus found in a number of different species [32]. Infection of a naïve, natural host by these viruses usually results in a persistent infection that rarely progresses to a pathogenic event outside of specific clinical events. In contrast, cross-species transmission of these viruses can result in profound diseases. For example, herpesvirus saimiri (HVS) is a natural virus of squirrel monkeys, found in over 90% of animals in captivity with no observable pathogenesis [11]. However, transmission of HVS to common marmosets results in a fatal, lymphoproliferative disorder with 100% efficiency. In addition, the transmission of RRV into common marmoset was clear in regard to persistent infection, although it was less conclusive with regards to pathogenicity (unpublished results). Furthermore, viruses could be re-isolated from this animal at multiple time points throughout the experiment (unpublished results). Thus, it is intriguing that the common marmoset is highly susceptible to infection by various pathogens. We speculate that the limited major histocompatibility complex class I (MHC I) polymorphism of common marmosets may contribute to their susceptibility to KSHV infection and pathogenesis. Extensive polymorphism of the MHC is thought to confer immune protection on populations. Restriction fragment length polymorphism analysis showed that there were a limited number of common marmoset MHC class I alleles, whereas the MHC class II gene loci were polymorphic. This may play a role in the susceptibility of this New World primate species to a variety of pathogens. In summary, this is the first animal model that significantly recapitulates the important aspects of KSHV infection in humans and will greatly aid the future designing of anti-viral therapies and be of use in the development of vaccines against KSHV.
All common marmosets (Callithrix jacchus, Cj) were housed at the New England Primate Research Center (NEPRC) in accordance with the standards of the American Association for Accreditation of Laboratory Animal Care and Harvard Medical School's Internal Animal Care and Use Committee. Common marmosets experimentally inoculated with rKSHV.219 were individually housed in bio-level 3 containment facilities.
Vero cells carrying rKSHV.219 were stimulated with trichostatin A (TSA), the supernatants were harvested to purify rKSHV.219, and the virus titer was determined in 293A and Vero cells by performing a GFP-positive infection assay as described [10]. In order to produce the virus, Vero cells carrying rKSHV.219 were stimulated with 75 nM of TSA for 24 hrs, the media changed, and grown for an additional 48 hrs in DMEM media without FBS. To harvest the virus, the cells were pelleted at 2000 rpm for 10 min with Sorvall SW40 rotor and the ensuing supernatant passed through a 0.45 µm filter, after which it was centrifuged at 18000 rpm for 3 hrs with Sorvall SA-600 rotor to concentrate rKSHV.219. 5×106 IU of this rKSHV.219 was then used to intravenously inoculate the marmosets, for which a flexible, small-bore orogastric tube was utilized to slowly deliver the viral inoculum to the caudal aspect of the oral cavity, tonsils, and nostril mucosa. The inoculum (5×107 IU of rKSHV.219) was dripped slowly over the oral mucosa while the animal was under light sedation, whereby the swallow reflex was maintained. The animals were given ketamine (10–20 mg/kg body weight) intramuscularly prior to blood sampling, inoculation, and euthanasia. The animals underwent periodic blood sampling for viral isolation attempts and the detection of viral DNA. The animals were maintained until the termination of the study or when any of the conditions for euthanasia were met.
rKSHV.219 was purified from Vero.rKSHV.219 cells and lysed with a 1% Triton X-100 buffer, after which it was put through five cycles of freezing in liquid nitrogen followed by thawing. The virion proteins were coated onto plates and used to detect reactive antibodies by ELISA as described [12],[28].
Following experimental rKSHV.219 inoculation, all animals were examined daily. Body temperature and clinical data were recorded via an implanted microchip and transponder (Bio Medic Data Systems, Maywood, NJ). PBMCs were obtained prior to inoculation and at various time points throughout the study. Tissues were fixed in 10% neutral-buffered formalin and snap-frozen at −70°C. Blood was obtained at various time points to quantify the viral antibody response, viral load, and complete blood count.
Formalin-fixed, paraffin-embedded, and snap-frozen tissues were used in immunohistochemical procedures to define the immunophenotype of cells within the skin lesion tissue as described [28]. Briefly, tissue sections were fixed in 2% paraformaldehyde and immunostained with an avidin-biotin-horseradish peroxidase complex technique with diaminobenzidine chromogen. The primary antibodies used in this study were anti-CD20 (B1), anti-CD8 (DK25), anti-CD3 (Nu-Th/1), anti-HLA-DR (CR3/43), anti-Ki67 (MIB-1), anti-vWF (A00082), anti-vimentin (3B4), and anti-HAM56 (M0632). Primary antibodies for viral markers LANA (clone 4A4) and vIL-6 were obtained from Advanced Biotechnologies Inc. (Columbia, MD). Confocal microscopy was used to define the immunophenotypes of virus-infected cells among the PBMCs.
To test for the presence of rKSHV.219, DNA was extracted from PBMCs or fresh frozen tissues using a QIAmp tissue kit (Qiagen, Valencia, CA) according to the manufacturer's instructions. DNA was eluted in 50 to 100 µl of sterile water treated with diethylpyrocarbonate and PCR was performed as described below. PBMCs were separated from the whole blood of infected common marmosets using standard Ficoll isolation techniques as described by the manufacturer (Organon Teknika, Malvern, PA).
Real-time PCR was performed with genomic DNA isolated from the PBMCs or tissues of the infected animals and specific primers based on previous analyses of the KSHV sequence [33]. LANA-specific primers [forward primer (5′-CCT CCA TCC CAT CCT GTG TC-3′) and backward primer (5′-GGA CGC ATA GGT GTT GAA GAG-3′)] were used to generate a 146-bp product for LANA detection. ORF9-specific primers [forward primer (5′-ATT CAA GGT CAT ATA CGG CG-3′) and backward primer (5′-CTG GAC AAA ACG ACA GGC TG-3′)] were used to generate a 262-bp product for ORF9 detection. Amplification was performed at 95°C for 25 s and 67.5°C for 60 s for 45 cycles in an iCycler thermal cycler system (Bio-Rad, CA). Data was obtained at CT values as per the manufacturer's guidelines (the cycle number at which logarithmic PCR plots cross a calculated threshold line). The PCR products were resolved on a 3% ethidium bromide-stained agarose gel and sequenced to confirm the identities of the KSHV LANA and ORF9 gene fragments.
As described in the section “Antibody responses,” virion particles were prepared by stimulating Vero.rKSHV.219 cells with TSA, then by freezing and thawing rKSHV.219 a total of five times in a 1% Triton X-100 buffer using liquid nitrogen. Purified virion proteins (20 µg) were resolved by SDS-polyacrylamide gel electrophoresis (PAGE) and transferred onto a PVDF membrane (Bio-Rad). Immunodetection was achieved with 1∶500 diluted monkey sera. The proteins were visualized by a chemiluminescence reagent (Pierce) and detected by a Fuji chemiluminometer.
5×105 cells per sample were washed with PBS medium containing 1% fetal calf serum and incubated with either fluorescein isothiocyanate-conjugated (FITC) or phycoerythrin-conjugated (PE) monoclonal antibodies for 30 min at 4°C. After washing, each sample was fixed with a 2% paraformaldehyde solution and flow cytometry analysis was performed with a FACS Scan (Becton Dickinson Co.).
|
10.1371/journal.pcbi.1001035 | Incremental Mutual Information: A New Method for Characterizing the Strength and Dynamics of Connections in Neuronal Circuits | Understanding the computations performed by neuronal circuits requires characterizing the strength and dynamics of the connections between individual neurons. This characterization is typically achieved by measuring the correlation in the activity of two neurons. We have developed a new measure for studying connectivity in neuronal circuits based on information theory, the incremental mutual information (IMI). By conditioning out the temporal dependencies in the responses of individual neurons before measuring the dependency between them, IMI improves on standard correlation-based measures in several important ways: 1) it has the potential to disambiguate statistical dependencies that reflect the connection between neurons from those caused by other sources (e.g. shared inputs or intrinsic cellular or network mechanisms) provided that the dependencies have appropriate timescales, 2) for the study of early sensory systems, it does not require responses to repeated trials of identical stimulation, and 3) it does not assume that the connection between neurons is linear. We describe the theory and implementation of IMI in detail and demonstrate its utility on experimental recordings from the primate visual system.
| The root of our brain's computational power lies in its trillions of connections. With our increasing ability to study these connections experimentally comes the need for analytical tools that can be used to develop meaningful quantitative characterizations. In this manuscript, we present a new such tool, incremental mutual information (IMI), that enables the characterization of the strength and dynamics of the connection between a pair of neurons based on the statistical dependencies in their spiking activity. IMI is an important step forward from existing approaches, as it has the potential to disambiguate dependencies due to the connection between two neurons from those due to other sources, such as shared external inputs, provided that the dependencies have appropriate timescales. We demonstrate the utility of IMI through the analysis of simulated neuronal activity as well as activity recorded in the primate visual system.
| To understand the function of neuronal circuits and systems, it is essential to characterize the connections between individual neurons. The major connections between and within many brain areas have been mapped through anatomical studies, but these maps specify only the existence of connections, not their strength or dynamics (temporal properties). Measuring the strength and dynamics of the connection between two neurons requires physiological experiments in which the activity of both neurons is measured. The most direct of these experiments involves intracellular recordings, which allow the connection between the two neurons to be directly investigated. However, intracellular recordings are difficult to perform in vivo and impossible to obtain from more than a few cells at a time. Instead, most physiological studies of connectivity rely on extracellular recordings from multi-electrode arrays (or, more recently, imaging of calcium activity). In these experiments, it is not usually possible to explicitly verify anatomical connectivity, nor to directly characterize the connections. Instead, the strength and dynamics of ‘functional’ connectivity must be inferred through statistical methods.
The traditional method for characterizing the strength and dynamics of the connection between two neurons is the cross correlation function (CXY), which measures the linear correlation between two signals over a range of specified delays [1]. While CXY and its variants have been used successfully in a number of studies (see, for example, Usrey and Reid [2] for a review of many such studies in the visual system), it has limitations that must be considered when studying the connection between neurons [3]–[5]. The limitations of CXY arise from the fact that it is a measure of the total (linear) dependency between two signals and, thus, implicitly assumes that all dependencies between them are due to their connection. In the case of neurons, there are in fact many potential sources of dependency – shared external stimuli, intrinsic cellular and network properties, etc. – and CXY cannot disambiguate these dependencies from those due to the actual connection. Several modified versions of CXY have been proposed to address these drawbacks. For example, if neuronal activity in response to repeated trials of the same external stimulus is available for analysis, as is often the case in early sensory systems, the ‘shift-predictor’ can be used to remove some of the correlations due to the stimulus [1]. Further modifications to CXY have also been proposed to remove the correlations due to stimulus-driven covariations in activity [6] and background activity [7]. While these modified approaches have certainly improved upon the standard CXY, the confound of dependencies due to the connection and those arising from other sources remains a general problem.
In addition to correlation-based methods, there are several other approaches to characterizing the dependency between two signals that can be used to study the connection between two neurons. These methods can be generally divided into two classes: model-based and model-free. The most common model-based approach to characterizing dependency is Granger causality (GC) [8]. With GC, one signal is predicted in two different ways: 1) using an autoregressive model based on its own past and 2) using a multivariate autoregressive model based on its own past and the past of the second signal. The strength of the dependency is given by the difference in the predictive power of the two models and the dynamics of the dependency are reflected in the regression parameters that correspond to the influence of the second signal. The power of model-based approaches such as GC is dependent on the validity of the underlying model; if the dependency between the two signals is approximately linear, then the characterization provided by GC will be accurate, but in situations where the properties of the dependency are complex or unknown, as is often the case with neurons, a model-free approach may be more appropriate. The most common model-free approach to characterizing dependency is transfer entropy (TE), the information-theoretic analog of GC [9]. TE measures the reduction in the entropy of one signal that is achieved by conditioning on its own past and the past of the second signal relative to the reduction in entropy achieved by conditioning on its own past alone. TE is a powerful tool for measuring the overall strength of a dependency, but is not suitable for characterizing its dynamics.
In this paper, we detail a new model-free approach for characterizing both the strength and dynamics of a dependency by ‘conditioning out’ the temporal correlations in both signals before assessing the strength of the dependency at different delays. This approach can overcome some of the confounds that are common in studies of neuronal connectivity [10]–[12], as it has the potential to disambiguate statistical dependencies that reflect the connection between neurons from those caused by other sources (e.g. shared inputs or intrinsic cellular or network mechanisms) provided that the dependencies have appropriate timescales. In the following sections, we outline the theory behind our measure, which we call incremental mutual information, illustrate its usage on simulated neuronal activity and experimental recordings from the primate visual system, and consider its relationship to other common measures of dependence.
Matlab code for measuring incremental mutual information is available for download at http://www.ucl.ac.uk/ear/research/lesicalab
In order to characterize the strength and dynamics of the connection between two signals, it is necessary to quantify how much one signal at one point in time influences the other signal at nearby points in time. Most measures of dependence between two signals X and Y seek to quantify the difference between the joint distribution p(X,Y) and the product of the marginal distributions p(X) p(Y). For example, the cross correlation function measures the difference between the mean of the joint distribution and the product of the means of the marginal distributions (the covariance), normalized by the product of the standard deviations for a given delay δ:(1)where CXY[δ] is the correlation coefficient between X[n] and Y[n], which are assumed to be discretized signals, at integer delay δ.
As described in the Introduction, CXY has limitations that are important to consider when studying neuronal connectivity. Most importantly, CXY, as with all dependency measures that operate only on the joint distribution p(X,Y) and the marginal distributions p(X) and p(Y), cannot differentiate between the dynamics of the connection between the neurons and the temporal correlations in their activity that are due to other sources. It is possible to overcome this limitation by conditioning out the temporal correlations in each signal before measuring the dependency between them, i.e. rather than operate on p(X,Y), p(X), and p(Y), operate on p(X,Y|), p(X|), and p(Y|), where is a vector containing the past and future of X[n] and Y[n] relative to the delay of interest(2)as shown in the schematic diagram in figure 1.
The analog of CXY for conditional distributions is the partial cross correlation:(3)While CXY|Z overcomes the major limitation of CXY, it is still a linear measure and may not accurately characterize nonlinear dependencies.
The idea of partial correlation can be generalized for the study of any dependency by formulating the information-theoretic analog of CXY|Z as a partial mutual information [13]: First, the entropy of X is measured after conditioning on its own past and future, as well as the past and future activity of Y relative to the delay of interest. Then, the strength of the influence of Y on X at the delay of interest can be measured as the additional reduction in entropy that occurs after observing Y at that delay:(4)Because this quantity, which we call the incremental mutual information (IMI), reduces the uncertainty of X as much as possible before measuring the influence of Y at each delay, it has the potential to provide an accurate description of both the strength and dynamics of their dependency. In this form, ΔIXY is similar to a partial covariance in that its value is dependent on the properties of the individual signals (e.g. the total entropy of X, the strength of the temporal correlations in X, etc.). In some cases, it may be preferable to use a normalized measure that is more similar to a partial correlation coefficient, i.e. a measure that expresses the incremental mutual information as a fraction of its maximum possible value:(5)To determine whether IMI is appropriate for use in any particular context, it is important to consider the relative timescales of the dependency between the signals and the other dependencies to be conditioned out. At any particular delay, the effects of dependencies with durations that are long relative to the time bins used for discretization will be predictable from the past and future values of the signals, so their contribution to the IMI will be small, i.e. dependencies with a slow timescale will make a relatively large contribution to initial reduction in the entropy of X based on past and future values of X and Y, , but not to the additional reduction in the entropy of X based on the present value of Y, . Conversely, the effects of dependencies that have a duration that is similar to the time bins used for discretization will not be predictable from the past and future values of the signals, so their contribution to the IMI will be large, i.e. dependencies with a fast timescale will make a small contribution to the initial reduction in entropy , but will make a large contribution to the additional reduction in entropy . Thus, IMI will be most useful when the duration of the dependency between the signals is similar to the size of the time bins used for discretization and the durations of the other dependencies to be conditioned out are longer. Fortunately, this is often the case for neurons in sensory systems, as will be illustrated in the examples in the Results.
As with any measure based on entropies, the calculation of IMI requires careful consideration. Because IMI is a model-free approach, the number of samples required to produce a result of a given precision are likely to significantly exceed those of model-based approaches. The bias and variability of the entropy estimates that underlie the computation of IMI can vary substantially depending on the size of the data sample, the number of possible values that a signal can take on, and the signal dimensionality. Fortunately, neuronal activity typically has only a few possible values (e.g. the number of spikes in each time bin). However, the terms , , , and representing the past and future of the signals are vectors. In practice, these vectors must be limited to some finite length, which we term ω, and this length will determine their dimensionality:(6)Thus, the calculation of IMI requires a tradeoff: increasing the value of ω allows the entropy of the first signal to be reduced as much as possible before measuring the influence of the second signal, but also increases the chances that the entropy estimates may be biased or highly variable. There are a number of bias correction techniques available that may be useful in mitigating problems related to sample size [14]. For the examples below, we corrected the entropy estimates using ‘quadratic extrapolation’ bias correction via the information toolbox software available at http://www.ibtb.org [15]. Also, for all of the examples below, time is discretized into sufficiently small bins such that each bin contains no more than one spike, limiting the possible values of X and Y to 0 and 1.
Because the bias and variability of entropy estimates are dependent on sample size, it is critical to establish the validity and precision of any calculation of IMI using statistical methods. In the experimental examples presented below, we use two different bootstrap procedures with random sampling to establish 95% confidence intervals and to determine whether the observed values are significantly different from zero. To establish 95% confidence intervals, we calculated IMI from 100 random samples of the same size drawn with replacement from the original sample. To preserve the temporal dependencies in the data, sampling was performed after the vectors were formed and the three vectors were sampled together. Confidence intervals were defined as the mean ± 2 standard deviations of the values calculated from the random samples. To establish the significance of the observed values, the same procedure was followed, but Y was sampled separately from . This sampling preserved the dependencies between , but removed the dependencies between X and Y (and, thus, in theory, removed any IMI between them). The observed values were considered significantly different from zero if they were greater than 2 standard deviations above the mean of the values calculated from the random samples.
IMI is designed to give accurate measures of the strength and dynamics of the connections between neurons even in cases when the correlation may not, i.e. when the activities of individual neurons contain temporal correlations unrelated to the connection between them. In these cases, the cross correlation function can be ambiguous – its shape can reflect either the true dynamics of the connection, temporal correlations in the activities of the individual neurons, or a combination of both. A simple example of this ambiguity is illustrated in figure 2a.
We first simulated a pair of neurons X and Y with independent, uncorrelated inputs and a dynamic connection, i.e. a spike from neuron Y caused a prolonged increase in the spiking probability of neuron X. We simulated the activity of neuron Y as a dichotomized Gaussian noise and the activity of neuron X as the dichotomized sum of a Gaussian noise and the filtered activity of Y:(7)where and are uncorrelated, ε = 0.5 is a scaling factor determining the overall strength of the connection, θ = 1 is the spiking threshold, and the input from Y to X, , is the convolution of the activity of Y with a Gaussian filter g[n] with a peak delay of 4 samples and a half width of 3 samples (note that , the filtered version of Y, is unobserved). From the simulated activity of this pair of neurons (with a sample size of 220), we estimated the cross correlation function CXY and normalized incremental mutual information (with ω = 2) at delays ranging from δ = −10 to 10 samples. Both CXY and for this pair were broad, reflecting the dynamics of the connection.
We next simulated another pair of neurons that was similar to the first one, except that Υ received input with temporal correlations and the connection between Υ and X was static with a delay of 4 samples:(8)where is the convolution of sy with a Gaussian filter g[n] with a peak at zero delay and a half width of 3 samples. While CXY for this pair was also broad because of the temporal correlations in the activity of Y, was sharp, reflecting the static connection. Thus, while IMI captures the differences in the connections between these two pairs of neurons, correlation conflates connection dynamics with temporal correlations in individual activities and yields ambiguous results.
This example can also be used to illustrate the necessity of conditioning out the both past and future activities of the neurons. A modified version of IMI can be formulated in which only the past activities of the two neurons are conditioned out:(9)In this formulation, the IMI is related to transfer entropy (see Discussion). As shown in figure 2b, correctly conditions out the effects of the temporal correlations in the activity of Y for delays that are smaller than that of the actual connection, but not for delays that are larger than that of the actual connection. This reason for this asymmetry is as follows: Because of the temporal correlations in the activity of Y, its value will be similar for neighboring samples. When the delay of interest δ is smaller than the delay corresponding to the actual connection δ*, Y[n−δ*] is included in the vector of past activity and, since Y[n−δ] carries no information about X beyond that which is carried by Y[n−δ*], Y[n−δ] makes no contribution to the IMI. However, when Y[n−δ*] is not included in the vector of past activities, Y[n−δ], which is similar to Y[n−δ*] because of the temporal correlations in Y, will carry additional information about the activity of X and, thus, will contribute to the IMI.
As a further consequence of the ambiguity in the cross correlation function illustrated in the example above, temporal correlations in individual activities may mask weak connections between neurons entirely. A simple example of this problem is shown in figure 3a. We simulated a pair of neurons that received a shared input with temporal correlations and had a weak static connection between them with a delay of 3 samples:(10)where and are the convolution of Gaussian noise with a Gaussian filter as described above with a correlation coefficient of 0.5 between them, and ε = 0.25 (other parameter values are as described above). CXY for this pair of neurons was broad, with no discernable increase at the delay corresponding to the connection (black arrow), while exhibits a sharp peak at the appropriate delay. Thus, by conditioning out dependencies due to shared input, IMI is able to reveal connections that may not be evident in the cross correlation function.
A slight modification of the previous example can be used to illustrate a situation where shared inputs cannot be conditioned out and contaminate the IMI. As described above, IMI will be most useful when the duration of the dependency between the signals is similar to the size of the time bins used for discretization and the durations of the other dependencies to be conditioned out are longer, as is the case in example 2. If the simulation in example 2 is modified so that the shared input is uncorrelated over time, the dependency resulting from the shared input can no longer be conditioned out, as the past and future activities of the neurons can no longer be used to infer the effects of the input at the delay of interest. As a result, has two peaks, one with no delay reflecting the shared input, and another with a delay reflecting the actual connection, as shown in figure 3b. It should be noted that this type of contamination can potentially arise both from shared external sources such as sensory stimuli as well as from other unobserved neurons.
To test the utility of IMI on experimental data, we analyzed the activity of two pairs of thalamic relay neurons and their retinal ganglion cell (RGC) inputs recorded in the lateral geniculate nucleus (LGN) of an anesthetized monkey as shown in figure 4a. The details of the experimental procedures can be found in Carandini et al. [16]. During the recordings, visual stimulation was presented via an LED that illuminated the receptive field center with an intensity that varied naturally (i.e. with temporal correlations typical of the natural environment). In this example, the stimulus was approximately 12 min in duration and did not repeat.
The histograms in figure 4b show the basic relationship between the activity of the retinal and thalamic neurons in each pair. For the first pair, less than half of the RGC postsynaptic potentials (PSPs) evoked immediate LGN spikes, while the connection between the second pair was stronger, with nearly 75% of PSPs evoking immediate spikes. We calculated the cross correlation function and incremental mutual information for these pairs after binarizing the spike trains in 2 ms time bins. CXY for these pairs has a complex shape with 3 components: a broad positive peak with a half width of approximately 20 ms reflecting the temporal correlations in the visual stimulus, two sharp negative peaks reflecting refractory effects, and a sharp positive peak reflecting the actual connection between the cells. In contrast, for these pairs had one main peak reflecting the connection between the neurons - the effects of statistical dependencies arising from the stimulus correlations have been completely removed and the refractory effects have been largely conditioned out. For the first pair, had a relatively long tail, reflecting temporal summation of RGC PSPs that failed to evoke an immediate LGN spike. For the second pair, was sharper, reflecting the stronger connection between the cells.
In early sensory systems, experiments are often designed such that the activity in response to repeated trials of an identical stimulus are observed so that the correlation between neurons can be separated into two distinct parts known as signal correlation and noise correlation. The signal correlation, which reflects both correlation in the stimulus itself and similarities in neurons' preferred stimulus features, will capture the correlation in the fraction of the response that is repeatable from trial to trial, i.e. the correlation that remains after the trial order has been randomized:(11)where Xi[n] is the response of neuron X on trial i and indicates the expectation over all possible combinations of trials i and j in which their values are not equal. In studies of neuronal connectivity, is often referred to as the ‘shift-predictor’.
The noise correlation, which results from network and intrinsic cellular mechanisms, will capture the remaining correlation in the fraction of the response that is variable from trial to trial(12)and, thus, captures the dependencies between the neurons that are not locked to the external stimulus. However, while may provide a better measure of the strength and dynamics of the connection between two neurons than CXY, it still confounds connection dynamics and temporal correlations that are independent of the stimulus, e.g. refractory effects or coupled oscillations.
For comparison with and , the signal and noise IMI between X and Y can be formulated in an analogous fashion. The signal IMI is the reduction in the entropy of the response of X on trial i that results from observing the response of Y on trial j at the delay of interest, beyond that which results from observing the past and future responses of both neurons on trial i:(13)where . The noise IMI is the difference between the total IMI and the signal IMI, i.e. the reduction in the entropy of the response of X on trial i that results from observing the response of Y on trial i at the delay of interest and the past and future responses of both neurons on trial i, beyond that which results from observing the response of Y at the delay of interest on trial j and the past and future responses of both neurons on trial i:(14)
We estimated the signal and noise correlations and signal and noise IMI for the same two retinogeniculate pairs that were analyzed in experimental example 1 using a different set of responses to 140 repeated trials of identical stimulation in which each trial was 5 seconds in duration. As shown in figure 5, for both pairs was broad, reflecting the temporal correlations in the visual stimulus. In contrast, was nearly zero at all delays – because the temporal correlations in the visual stimulus were slow relative to the bin size used for discretization, there was little information about stimulus-induced dependencies to be gained by observing the RGC activity at any particular delay on a different trial when RGC and LGN activity at surrounding delays on the current trial were already known. While the effects of the stimulus correlations were removed from for both pairs, these functions still had a complex shape, with two negative peaks reflecting refractory effects, and one positive peak reflecting the actual connection between the neurons. Thus, while shuffling removed some of the confounding correlations in CXY, others still remained, while in , which has one main peak reflecting the connection between the neurons, most of the confounding dependencies have been conditioned out.
This example illustrates an important property of IMI. as shown in figure 5 is nearly identical to for the same two pairs shown in figure 4. Thus, unlike the cross correlation function, IMI does not require multiple trials in order to differentiate the temporal correlations in the responses of individual neurons from the dynamics of the connection between them.
We have presented IMI as a new approach to characterizing the strength and dynamics of the connection between neurons. By conditioning out the temporal dependencies in the responses of individual neurons before assessing the connection between them, IMI improves on correlation-based measures in several important ways: 1) IMI has the potential to disambiguate connection dynamics from other temporal dependencies due to shared inputs or intrinsic cellular or network mechanisms provided that the dependencies have appropriate timescales, 2) for the study of sensory systems, IMI does not require responses to repeated trials of identical stimuli, and 3) IMI does not assume that the connection between neurons is linear. Through example applications of IMI to simulated and experimentally recorded neuronal activity, we have demonstrated that IMI has the potential to be both a powerful and practical tool for analyzing the functional connectivity in neuronal circuits.
The major determinant of the ability of IMI to differentiate connection dynamics from other dependencies is the relative timescale of the other dependencies. If the other dependencies have a long duration relative to the time bins used for discretization, then their effects can be conditioned out through observation of past and future neuronal activity, as demonstrated in the experimental examples presented above. If the other dependencies have a duration that is similar to the bin size, then their effects cannot be conditioned out without explicit observation of their source.
As formulated here, IMI is designed to analyze the connection between a pair of neurons. However, in many brain areas, each neuron receives input from a large population, and correlations between these other inputs and the input under study could contaminate the IMI. If the other inputs are unobserved, it will be difficult to account for their effects with a model-free approach, though recent work with model-based approaches has demonstrated some success [17]–[20]. If the other inputs are observed (which is becoming increasingly common with recent advances in recording and imaging technology that allow for simultaneous recording of the activity complete or nearly complete local populations of neurons), there is no reason that, in principle, IMI cannot be extended to condition out dependencies due to the activity of the other neurons. However, adding the activity of additional neurons to the conditioning vector will increase its dimensionality, and, thus, the bias and variability of the entropy estimates that underlie the computation of IMI. While this may not be a problem for a small number of neurons, it is certain to be a problem for large populations. Thus, for large populations, it may be more appropriate to use a model-based approach such as Granger causality within a generalized linear model framework [21].
Of the existing approaches to characterizing dependencies between signals, IMI is most similar to transfer entropy [9]. TE measures the dependency between two signals as the difference in the entropy of one signal after conditioning on its own past and conditioning on its own past and the past of the other signal, or, in the terminology used to define IMI, . From this definition, it is clear that TE and IMI are designed for different purposes: TE measures the overall causal strength of the dependency between two signals by first conditioning out the past of one signal and then measuring how much can be learned about the present value of that signal based on the past of the second signal, while IMI measures the strength and dynamics of the dependency between two signals by first conditioning out past and future of both signals and then measuring how much can be learned about the present value of one signal from the present value of the other relative to some delay. The key difference between TE and IMI, as illustrated in the simulated example presented above, is that, even if computed at a range of delays, TE is not suitable to assess the dynamics of a dependency because it considers only past activity and, as a result, conditions out temporal correlations appropriately for delays that are shorter than that of the actual dependency, but not for delays that are longer than that of the actual dependency.
The most effective model-based approach for studying the functional connectivity in a neuronal circuit is the generalized linear model (GLM) [22]–[24]. The GLM attempts to predict a neuron's activity based not only on its own activity and the activity of other neurons, but also on external inputs. Because all of the filters in the model are fit simultaneously, the influence of the external inputs on the activity of each neuron, as well as those of its own past activity, are separated from the influence of other neurons. The power of the GLM lies in the fact that once the filters have been estimated, the model can be used to predict the activity of the entire group of neurons to any external input, but this power comes at the expense of assuming a particular parametric structure. Relative to IMI, which makes no assumptions about the connections between neurons, the drawback of the GLM is that the interactions between neurons are assumed to be of a particular nature (usually additive). However, this assumption also allows the GLM to be readily applied to large populations.
|
10.1371/journal.pcbi.1006493 | Verbalizing phylogenomic conflict: Representation of node congruence across competing reconstructions of the neoavian explosion | Phylogenomic research is accelerating the publication of landmark studies that aim to resolve deep divergences of major organismal groups. Meanwhile, systems for identifying and integrating the products of phylogenomic inference–such as newly supported clade concepts–have not kept pace. However, the ability to verbalize node concept congruence and conflict across multiple, in effect simultaneously endorsed phylogenomic hypotheses, is a prerequisite for building synthetic data environments for biological systematics and other domains impacted by these conflicting inferences. Here we develop a novel solution to the conflict verbalization challenge, based on a logic representation and reasoning approach that utilizes the language of Region Connection Calculus (RCC–5) to produce consistent alignments of node concepts endorsed by incongruent phylogenomic studies. The approach employs clade concept labels to individuate concepts used by each source, even if these carry identical names. Indirect RCC–5 modeling of intensional (property-based) node concept definitions, facilitated by the local relaxation of coverage constraints, allows parent concepts to attain congruence in spite of their differentially sampled children. To demonstrate the feasibility of this approach, we align two recent phylogenomic reconstructions of higher-level avian groups that entail strong conflict in the "neoavian explosion" region. According to our representations, this conflict is constituted by 26 instances of input "whole concept" overlap. These instances are further resolvable in the output labeling schemes and visualizations as "split concepts", which provide the labels and relations needed to build truly synthetic phylogenomic data environments. Because the RCC–5 alignments fundamentally reflect the trained, logic-enabled judgments of systematic experts, future designs for such environments need to promote a culture where experts routinely assess the intensionalities of node concepts published by our peers–even and especially when we are not in agreement with each other.
| Synthetic platforms for phylogenomic knowledge tend to manage conflict between different evolutionary reconstructions in the following way: "If we do not agree, then it is either our view over yours, or we just collapse all conflicting node concepts into polytomies". We argue that this is not an equitable way to realize synthesis in this domain. For instance, it would not be an adequate solution for building a unified data environment where authors can endorse and yet also reconcile their diverging perspectives, side by side. Hence, we develop a novel system for verbalizing–i.e., consistently identifying and aligning–incongruent node concepts that reflects a more forward-looking attitude: "We may not agree with you, but nevertheless we understand your phylogenomic inference well enough to express our disagreements in a logic-compatible syntax. We can therefore maximize the translatability of data linked to our diverging phylogenomic hypotheses". We show that achieving phylogenomic synthesis fundamentally depends on the application of trained expert judgment to assert parent node congruence in spite of incongruently sampled children.
| Three years ago, Jarvis et al. (2014; henceforth 2014.JEA) [1] published a landmark reconstruction of higher-level bird relationships. Within 12 months, however, another analysis by Prum et al. (2015; henceforth 2015.PEA) [2] failed to support several of the deep divergences recovered in the preceding study, particularly within the Neoaves sec. (secundum = according to) Sibley et al. (1988) [3]. Thomas (2015) [4] used the term "neoavian explosion" to characterize the lack of congruence between inferences of early-diverging lineages (see also [5]). Similarly, after reviewing six phylogenomic studies, Suh [6] concluded that the root region of the Neoaves constitutes a "hard polytomy". Multiple analyses have dissected the impact of differential biases in terminal and genome sampling, as well as evolutionary modeling and analysis constraints, on resolving this complex radiation [7, 8, 9]. Suh [6] argues that a well resolved consensus is not imminent (though see [10]). Brown et al. (2017) [11] analyzed nearly 300 avian phylogenies, finding that the most recent studies "continue to contribute new edges".
These recent advancements provide an opportunity to reflect on how synthesis should be realized in the age of phylogenomics [11, 12, 13]. The neoavian explosion can be considered a use case where multiple studies provide strong signals for conflicting hierarchies. Resolution towards a single, universally adopted tree is unlikely in the short term.
Rather than focusing on the analytical challenges along the path towards unitary resolution [9], we turn to the issue of how the persistence of conflict affects the design of synthetic data infrastructures. In other words, how do we build a data service for phylogenomic knowledge in the face of persistent conflict? This question is of broad relevance to systematists, comparative evolutionary biologists, and designers of biological information services interested in robust, reproducible, and reusable phylogenomic data. And it turns on the issue of improving identifiers and identifier-to-identifier relationships for this domain.
Particularly verbal representations of the neoavian explosion are not well designed for conflict representation and synthesis [14]. To alleviate this, some authors use tree alignment graphs in combination with color and width variations to identify regions (edges) of phylogenomic congruence and conflict [15]. Other authors may show multiple incongruent trees side-by-side, using color schemes for congruent clade sections [9]. Yet others may use tanglegrams that are enhanced to highlight congruence [4], rooted galled networks [16] or neighbor-net visualizations [17] that show split networks for conflicting topology regions, or simply provide a consensus tree in which incongruent bifurcating branch inferences are collapsed into polytomy [6].
Verbalizing phylogenomic congruence and conflict in open, synthetic knowledge environments [13] constitutes a novel challenge for which traditional naming solutions in systematics are inadequate. The aforementioned studies implicitly support this claim. All use overlapping sets of Code-compliant [18] and other higher-level names in the Linnaean tradition, with sources including [19] or [20]. To identify these source-specific name usages, we will utilize the taxonomic concept label convention of [14]. Accordingly, name usages sec. 2014.JEA are prefixed with "2014.", whereas name usages sec. 2015.PEA are prefixed with "2015."
We diagnose the verbalization challenge as follows. (1) In some instances, identical clade names are polysemic–i.e., have multiple meanings–across studies. For instance, 2015.Pelecaniformes excludes 2015.Phalacrocoracidae, yet 2014.Pelecaniformes includes 2014.Phalacrocoracidae; reflecting on two incongruent meanings of "Pelecaniformes". (2) In other cases, two or more non-identical names have congruent meanings, e.g., 2015.Strisores and 2014.Caprimulgimorphae. (3) Names that are unique to just one study–e.g., 2015.Aequorlitornithes or 2014.Cursorimorphae–are not always reconcilable in meaning without additional human effort, thereby adding an element of referential uncertainty to the apparent conflict. (4) Lastly, many of the newly inferred and conflicting edges are not named at all. There is an implicit preference for labeling edges when suitable names are already available. However, unnamed edges can create situations where conflict cannot be verbalized and reconciled in a data environment, due to the lack of syntactic structure ("names").
Jointly, the effects of polysemic names, synonymous names, exclusive yet hard-to-reconcile names, and conflicting unnamed edges are symptomatic of an information culture that is not ready for the identifier and identifier-to-identifier relationship challenges inherent in representing phylogenomic conflict. Suppose we wish to build a collaborative knowledge environment towards inferring "the tree of life" (though see [12]). The design should allow us to individually represent and at the same integrate conflicting hierarchies, from the tips to the root. The system should respond to name-based data queries across these hierarchies, and return whether they are congruent or how they conflict in meaning. Clearly, the name usages of each individual source are not suited for this integration task. Traditional, Linnaean conventions allow for names to have evolving phylogenomic meanings across hierarchies and are therefore too under-powered for our purpose [21].
At root, this is a novel conceptual challenge for systematics and comparative evolutionary biology, made imperative by the accelerated generation and ingestion of phylogenomic trees into open, dynamic knowledge bases for reliable integration and re-use [11, 13, 22, 23, 24]. The services that such environments aspire to provide require an appropriate theory of node identity, and hence a conception of multi-node congruence or incongruence across individual trees and entire synthesis versions.
Here we propose a solution to the phylogenomic conflict representation challenge. This solution requires collaboration between systematic experts, platform designers, and users of phylogenomic information. It is an extension of prior "concept taxonomy" research [14, 25, 26], and deploys logic reasoning to align tree hierarchies based on Region Connection Calculus (RCC–5) assertions of node congruence [27, 28, 29]. We demonstrate the feasibility of this approach by aligning subregions and entire phylogenomic trees inferred by 2015.PEA and 2014.JEA. In doing so, we address key representation challenges; such as the paraphyly of classification schemes used to label tree regions, and the inference of higher-level node congruence in spite of differentially sampled terminals. The alignment products for this use case constitute a novel answer to our central question: "how to build a synthetic knowledge environment in the face of persistent phylogenomic conflict?" The discussion focuses on the feasibility and desirability of creating such an integration service, emphasizing the role of trained expert judgment in providing them [30].
1. Taxa are models, concepts are mimics. We typically refrain from using the terms "taxon", "taxa", or "clade(s)". We take taxa to constitute evolutionary, causally sustained entities whose members are manifested in the natural realm. The task for systematics is to successively approximate the identities and limits of these entities. Thus, we assign the status of 'models' to taxa, which systematists aim to 'mimic' through empirical theory making. This perspective allows for realism about taxa, and also for the possibility to let our representations stand for taxa [31], at any given time and however imperfectly, to support evolutionary inferences.
In reserving a model status for taxa, we can create a separate design space for the human theory- and language-making domain. In the latter, we speak only of taxonomic or phylogenomic concepts–the products of inference making [21].
2. Sameness is limited to the same source. Therefore, for the purpose of aligning the neoavian explosion use case, we need not speak of the "same taxa" or "same clades" at all. Similarly, we need not judge whether one reconstruction or the other more closely aligns with deep-branching avian taxa, i.e., which is (more) 'right'? Instead, our alignment is only concerned with modeling congruence and conflict across two sets of concept hierarchies. The concepts are labeled with the "sec." convention to maintain a one-to-one modeling relationship between concept labels and concepts (clade identity theories). Accordingly, there is also no need to say that, in recognizing each a concept with the taxonomic name Neornithes, the two author teams are authoring "the same concept". Instead, we model the two labels 2015.Neornithes and 2014.Neornithes, each of which symbolizes an individually generated phylogenomic theory region. As an outcome of our alignment, we may say that these two concepts are congruent, or not, reflecting the intensional alignment (to be specified below) of two phylogenomic theories. But, by virtue of their differential sources (authorship provenance), the two concepts 2015.Neornithes and 2014.Neornithes are never "the same". "Sameness" is limited in our approach to concepts whose labels contain an identical taxonomic name and which originate from a single phylogenomic hierarchy and source. That is, 2015.Neornithes and 2015.Neornithes are (labels for) the same concept.
The methods used herein are consistent with [14, 26, 32]. They utilize three core conventions: (1) taxonomic concept labels to identify concepts; (2) is_a relationships to assemble single-source hierarchies via parent/child relationships; and (3) RCC–5 articulations to express the relative congruence of concept regions across multi-sourced hierarchies. The RCC–5 articulation vocabulary entails (with corresponding symbol): congruence (= =), proper inclusion (>), inverse proper inclusion (<), overlap (><), and exclusion (!). Disjunctions of these articulations are a means to express uncertainty; as in: 2015.Neornithes {= = or > or <} 2014.Neornithes. All possible disjunctions generate a lattice of 32 relationships (R32), where the "base five" are the most logically constraining subset [33].
The alignments are generated with the open source Euler/X software toolkit [28]. The toolkit ingests multiple trees (T1, T2, T3, etc.) and articulation sets (A1–2, A2–3, etc.), converting them into a set of logic constraints. Together with other default or facultative constraints (C) needed for modeling tree hierarchies, these input constraints are then submitted to a logic reasoner that provide two main services. First, the reasoner infers whether all input constraints are jointly logically consistent, i.e., whether they permit at least one "possible world". Second, if consistency is attained, the reasoner infers the set of Maximally Informative Relations (MIR). The MIR constitute that unique set of RCC–5 articulations for every possible concept pair across the input sources from which the truth or falseness of any relationship in the R32 lattice can be deduced [14, 26, 33]. Many toolkit options and functions are designed to encode variable alignment input and output conditions, and to interactively obtain adequately constrained alignments. The toolkit also features a stylesheet-driven alignment input/output visualization service that utilizes directed acyclical graphs [28]. A step-wise account of the user/toolkit workflow interaction is provided in [26].
Aligning phylogenomic trees entails several special representation and reasoning challenges. We address three aspects here that have not been dealt with extensively in previous publications.
1. Representing intensional parent concept congruence via locally relaxed coverage. The first challenge relates directly to the issue of parent node identity. Unlike comprehensive classifications or revisions [14, 26, 34], phylogenomic reconstructions typically do not aspire to sample low-level entities exhaustively. Instead, select exemplars are sampled among all possible low-level entities. The aim is to represent lower-lever diversity sufficiently well to infer reliable higher-level relationships. Often, terminal sampling is not only incomplete for any single reconstruction, but purposefully complementary to that of other analyses. Generating informative genome-level data remains resource-intensive [10]. This makes it prudent to coordinate terminal sampling globally, by prioritizing the reduction of gaps over redundant terminal sampling. In the case of 2015.PEA (198 terminals) versus 2014.JEA (48 terminals), only 12 species-level concept pairs have labels with identical taxonomic names.
By default, the logic toolkit applies a coverage constraint to every input concept region. Coverage means that the region of a parent is strictly circumscribed by the union of its children [35]. However, this constraint is relaxable, either globally for all concepts, or locally for select concepts. To relax coverage locally, the prefix "nc_" (no coverage) is used in the input, as in 2014.nc_Psittacidae. This means: either a parent concept's referential extension is circumscribed by the union of its explicitly included children, or there is a possibility of additional children being subsumed under that parent but not mentioned in the source phylogeny. Either scenario can yield consistent alignments. In other words, if a parent concept has relaxed coverage, it can attain congruence with another parent concept in spite of each parent having incongruent sets of child concepts.
Managing coverage in the toolkit input is not trivial. Relaxing coverage globally is akin to saying "anything goes", i.e., any parent could potentially include any child. This would yield innumerable possible worlds, and therefore has no value for our purpose. On the other hand, applying coverage globally means–counter-intuitively in the case of phylogenomic trees–that only parents with completely congruent sets of children can themselves attain congruence. The challenge for experts providing the input is thus to relax coverage locally, and strictly in the service of 'neutralizing' lower-level sampling differences between trees that should not yield conflict at higher levels.
The effect of locally relaxed coverage is illustrated in Figs 1–4, using the example of parrots– 2015./2014.Psittaciformes. At the species level, the author teams sampled wholly exclusive sets of concepts for this alignment region (Figs 1 and 3). Even at the genus level, only 2015./2014.Nestor is redundantly sampled, yet with the articulation: 2015.Nestor_meridionalis ! 2014.Nestor_notabilis at the child level. Therefore, if no species-level concept sec. 2015.PEA has an explicitly sampled and congruent region in 2014.Psittaciformes, and, vice-versa, no species-level concept sec. 2014.JEA has such a region in 2015.Psittaciformes, then under global application of the coverage constraint we obtain the alignment: 2015.Psittaciformes ! 2014.Psittaciformes (Fig 2). The absence of even partial concept region overlap at the terminal level 'propagates up' to the highest-level parent concepts, which are therefore also exclusive of each other.
Asserting higher-level node congruence in light of lower-level node incongruence requires a conception of node identity that affirms counter-factual statements of the following type: if 2014.JEA had sampled 2014.Psittacus_erithacus, then the authors would have included this species-level concept as a child of 2014.Psittacidae. This is to say that 2015./2014.Psittacidae, and hence their respective parents, are intensionally defined [25, 36, 37].
Using a combination of published topological information (and support), more or less direct reiterations of phenotypic traits (cf. discussions and supplementary data of 2015.PEA and 2014.JEA), and trained judgment [30], we align these concept regions as if there are congruent property criteria that each region entails, i.e., something akin to an implicit set of synapomorphies or uniquely diagnostic features. Of course, the phylogenomic data provided by 2015.PEA and 2014.JEA do not signal intensional definitions directly. But neither do their genome-based topologies for parrots provide evidence to challenge the status of such definitions as previously proposed [38]. In addition, particularly 2015.PEA (supplementary information; sections on "detailed justification for fossil calibrations" and "detailed phylogenetic discussion; pp. 3–21) provide a provide an in-depth account of how their preferred topology relates to published, property-centered circumscriptions of dozens of higher-level clade concepts. We have to assume, fallibly and non-trivially, that such topology-to-synapomorphy relations are also implied by JEA.2014, as reflected (inter alia) in their discussion.
Three clarifications are in order. First, Region Connection Calculus is at best a means of translating the signal of an intensional definition. The congruent (= =) symbol means, only: two regions are congruent in their extension. The RCC–5 vocabulary is obviously not appropriate for reasoning directly over genomic or phenomic property statements. The reasoner does not assess whether 2015.Psittacidae, or any included child or aligned concept, has 'the relevant synapomorphies'. Doing so would not be trivial even if property-based definitions were provided for all higher-level node concepts, because we would still have to make theory-laden assumptions about their congruent phylogenomic scopes [26, 39, 40]. Second, we are not providing detailed textual narratives that would justify each assertion of higher-level congruence. Such narratives are possible, and even needed to understand disagreements, because they explain the reasoning process behind an expert-made assertion. However, our main objective here is to focus on the issue of RCC–5 translation of systematic signals; not on a character-by-character dissection of each congruent articulation. Third, a sensible intensional alignment strategy uses a minimal number of instances of locally relaxed coverage in order to compensate for differential child sampling at lower levels, so that parent coverage can remain in place at higher levels to expose incongruent node concepts. The benefits of this strategy will be shown below.
2. Representing clade concept labels. Our modeling approach requires that every region in each source tree receives a taxonomic or clade concept label. However, the source publications only provide such labels for a subset of the inferred nodes. In particular, 2015.PEA (p. 570: Fig 1) obtained 41 nodes above the ordinal level. Of these, 17 nodes (41.5%) were explicitly labeled in either the published figure or supplement (pp. 9–12). The authors also cite [20] as the primary source for valid name usages, yet that list is not concerned with supra-ordinal names. Similarly, 2014.JEA (p. 1322: Fig 1) inferred 37 nodes above the ordinal level, of which 23 nodes (62.2%) were given an explicit label. They provide an account (cf. supplementary materials SM6: 22–24) of their preferred name usages, sourced mainly to [20] and [41].
In assigning clade concept labels at the supra-ordinal level when the authors may have failed to do so (consistently), we nevertheless made a good faith effort–through examination of the supplementary information and additional sources [1, 3, 42, 43, 44, 45, 46, 47]–to represent the authors' preferred name usages. Where usages were not explicit, we selected the only or most commonly applied clade concept name at the time of publication. This effort yielded 13 additional labels for 2015.PEA (Table 1), and 7 such labels for 2014.JEA (Table 2).
If no suitable label was available, we chose a simple naming convention of adding "_Clade1", "_Clade2", etc., to the available and immediately higher-level node label, e.g. 2014.Passerea_Clade1. The numbering of such labels along the tree topology starts with the most immediate child of a properly named parent, and typically follows down one section of the source tree entirely ("depth-first"), before continuing with the higher-level sister section. Using this approach, we added 11 labels for 2015.PEA (Table 1) and 7 labels for JEA.2014 (Table 2). If greater numbers of labels need to be generated, including siblings, then it is sensible to have a rule for ordering sibling nodes, e.g. by assigning the next-lowest number to the sibling whose child's name appears first in the alphabet. Our numbering of the labels 2014.Passerea_Clade2 (child with first-appearing letter: 2014.Ardeae) and 2014.Passerea_Clade3 (child: 2014.Cursorimorphae) adhere to this rule.
The clade concept labeling convention was not applied below the family level, where instead phylogenomic resolution was collapsed into polytomy (exception: Figs 1–4). In the case of 2014.JEA, only four family-level concepts include two children, whereas the remainder have a single child sampled. Resolving the monophyly of subfamilial clade concepts was not the primary aim of 2014.JEA. The same applies to 2015.PEA, who sampled 104/125 family-level concepts with only 1–2 children.
3. Representing phylogeny/classification paraphyly. A third, relatively minor challenge is the occurrence of clade concepts in 2015.PEA's phylogenomic tree that are not congruently aligned with higher-level concepts of [20]. We highlight these instances here because they represent a widespread phenomenon in phylogenomics. It is useful to understand how such discrepancies can be modeled with RCC–5 alignments (Figs 5 and 6).
Fig 5 exemplifies the phylogenetic tree/classification incongruence observed in 2015.PEA. The authors state (supplementary Table 1, p. 1): "Taxonomy follows Gill and Donsker (2015; fifth ed)". As shown in Fig 5, their phylogeny accommodates four sampled genus-level concepts that would correspond to children of the family-level concept Eurylaimidae sec. Gill & Donsker (2015) [20]. However, these concepts are arranged paraphyletically in relation to the reference classification. There is no parent concept that can be labeled 2015.Eurylaimidae and would not also (1) include 2015.Pittidae, i.e., 2015.Passeriformes_Clade1 in Fig 6, or (2) just represent aligned subset of the Eurylaimidae sec. Gill and Donsker (2015) [20], i.e., 2015.Passeriformes_Clade2 or 2015.Passeriformes_Clade3 in Fig 6. The concept Eurylaimidae sec. Gill and Donsker (2015) [20] has an overlapping (><) articulation with 2015. Passeriformes_Clade1.
In summary, our approach represents non-monophyly as an incongruent alignment of the phylogenomic tree and the source classification used to provide labels for that tree's monophyletic clade concepts. There are four distinct regions in the phylogeny of 2015.PEA where such alignments are needed: {Caprimulgiformes, Eurylaimidae, Hydrobatidae, Procellariidae, Tityridae} sec. Gill & Donsker (2015) [20]. Each of these is provided in the S7–S9 Files.
The source phylogenies specify 703 and 216 clade or taxonomic concepts, respectively. The frequent instances of locally relaxed coverage increase the reasoning complexity in relation to multi-classification alignments [14], making specialized RCC–5 reasoning useful [48]. The reasoning and visualization challenges commend a partitioned alignment approach. To keep the Results concise, we show visualizations of the larger input and alignment partitions only in the Supporting Information. A detailed account of the input configuration and partitioning workflow is given below.
Underlying all alignments is the presumption that at the terminal (species) level, the taxonomic concept labels of 2015.PEA and 2014.JEA are reliable indicators of either pairwise congruence or exclusion [14, 26, 32]. That is, e.g., 2015.Cariama_cristata = = 2014.Cariama_cristata, or 2015.Charadrius_hiaticula ! 2014.Charadrius_vociferus. Because the time interval separating the two publications is short in comparison to the time needed for taxonomic revisions to effect changes in classificatory practice, the genus- or species-level taxonomic concepts are unlikely to show much incongruence; though see [49] or [50]. We note that 2015.PEA (p. 571) use the label 2015.Urocolius(_indicus) in their phylogenomic tree, which also corresponds to the genus-level name endorsed in [20] Gill & Donsker (2015). However, in their Supplementary Table 1 the authors use 2015.Colius_indicus. We chose 2015.Urocolius and 2015.Urocolius_indicus as the labels to apply in the alignments.
The toolkit workflow favors a partitioned, bottom-up approach [29]. The process of generating, checking, and regenerating input files must be handled 'manually' on the desktop (note: improved workflow documentation and semi-automation of input-output-input changes are highly desirable). The performance of different toolkit reasoners was benchmarked in [28].
To work efficiently, the large problem of aligning all concepts at once is broken down into multiple smaller alignment problems, e.g. 2015./2014.Psittaciformes (Figs 3 and 4). To manage one particular order-level alignment, we start with assembling each input phylogeny separately, with relaxed coverage applied as needed (Fig 3). The RCC–5 articulations for low-level concept pairs are provided incrementally, e.g., in sets of 1–5 articulations at a time. Following such an increment, the toolkit reasoning process is re-/deployed to validate input consistency and infer the number of possible worlds. There is an option to specify that only one possible world is sought as output, which is equivalent to just checking for input consistency, as opposed to inferring all possible worlds. Doing so saves time as long as the input remains (vastly) under-specified. The stepwise approach of adding a small number of articulations at a time leads to increasingly constrained alignments, while minimizing the risk of introducing many new. difficult-to-diagnose inconsistencies.
Once a set of small, topographically adjacent alignment partitions is well specified, these can serve as building blocks for the next, larger partition. Hence, the basic sequence of building up larger alignments is: (1) obtain a well-specified low- (order- or family-) level alignment; (2) record the inferred parent-level articulations from this alignment; (3) propagate the latter–now as low-level input articulations–for the next, more inclusive alignment; (4) as needed, prune the lowest-level (sub-ordinal) input concepts and articulations of (1) from this alignment; (5) repeat (1) to (4) for another paired region; (6) assemble the more inclusive alignment by (manually) connecting the pruned, propagated concepts and articulations from two or more lower-level alignments, by adding to them the higher-level concepts from each input phylogeny. Depending on the interplay between (ranked) higher-level names recognized in each phylogeny and the number of terminal concepts sampled, steps (1) to (6) may be iterated once (e.g., 2015./2014.{Falconiformes, Psittaciformes}) or multiple times (e.g., 2015./2014.Passeriformes) to cover a supra-/ordinal alignment. An example of the latter is the 2015./2014.Passerimorphae alignment, which includes two order-level concepts and their children in each source phylogeny. Such mid-level partitions eventually form the basis for the largest alignment partitions, e.g. 2015./2014.Telluraves.
Sometimes, coverage will have to be relaxed even at higher levels. In all, 2014.JEA sample children of 34 order-level concepts in their phylogeny, whereas 2015.PEA recognize 40 order-level concepts. The latter authors represent four order-level concepts for which no analogous children are included in 2014.JEA, i.e.: 2015.{Apterygiformes, Casuariiformes, Ciconiiformes, Rheiformes}. Three of these are assigned to 2015.Palaeognathae, whereas 2015.Ciconiiformes are subsumed under 2015.Pelecanimorphae–in each case under relaxed parent coverage. The remaining 36 order-level concepts sec. 2015.PEA show some child-level overlap with those of 2014.JEA.
Our partitioning approach for this use case started with specifying the input constraints for nearly 35 paired order-level concepts and their respective children, as demonstrated in Figs 3 and 4. The largest order-level partition is 2015./2014.Passeriformes, with 148 x 22 input concepts, seven instances of relaxed parent coverage, and 101 input articulations. This alignment completes in less than 15 seconds on an individual 2.0 GHz processor, yielding 3,256 MIR.
As the partitions grew, we configured the following six, non-overlapping alignments as building blocks for the global alignment: 2015./2014.Palaeognathae (34 x 12 input concepts, four instances of relaxed coverage, and 25 articulations; same data sequence used for following alignments), 2015./2014.Galloanserae (49, 16, 7, 46), 2015.Columbaves/2014.Columbimorphae + 2014.Otidimorphae (53, 37, 13, 37), 2015.Strisores/2014.Caprimulgimorphae (44, 17, 8, 32), 2015./2014.Ardeae (100, 55, 19, 75), and the largest partition of 2015./2014.Telluraves (316, 104, 37, 241).
At the next more inclusive level, the inferred congruence of 2015.Telluraves = = 2014.Telluraves presented an opportunity to partition the entire alignment into two similarly sized regions, where the complementary region includes all 2015./2014.Neornithes concepts (392, 174, 58, 259), except those subsumed under 2015./2014.Telluraves, which are therein only represented with two concepts labels and one congruent articulation. These two complements are the core partitions that inform our use case alignment, globally. The corresponding S10 and S11 Files include the input constraint (.txt) and visualization (.pdf) files, along with the alignment visualization (.pdf) and MIR (.csv).
The two large partitions yield unambiguous RCC–5 articulations from the species concept level to that of 2015./2014.Neornithes. They can be aggregated into a synthetic, root-to-order level alignment, where all subordinal concepts and articulations are secondarily pruned away (see above). Such an alignment retains the logic signal derived from the bottom-up approach, but represents only congruent order-level concept labels as terminal regions, except in cases where there is incongruence. We present this alignment as an analogue to Fig 1 in [4] (p. 515), and compare how each conveys information about congruent and conflicting higher-level clade concepts.
Lastly, we further reduce the root-to-order alignment to display only 5–6 clade concept levels below the congruent 2015./2014.Neoaves. This region of the alignment is the most conflicting, and therefore forms the basis for our Discussion.
Our alignments show widespread higher-level congruence across the neoavian explosion use case; along with several minor regions of conflict and one strongly conflicting region between concepts placed immediately below the 2015./2014.Neoaves.
We focus first on the large complementary partitions, i.e. 2015./2014.Neornithes (without) / 2015./2014.Telluraves (see S10 and S11 Files). Jointly, they entail 707 concepts sec. 2015.PEA and 283 concepts sec. 2014.JEA. Among these, 34 "no coverage" regions were added to 2015.PEA's phylogeny, whereas 61 instances of relaxing parent coverage were assigned to 2014.JEA's phylogeny, for a total of 95 instances of relaxing this constraint. The 2015./2014.Neornithes partition shows 305 aligned regions– 247 without the "no coverage" regions–of which 60 congruently carry at least one concept label from each source phylogeny. This alignment also shows eight congruent species-level concept regions. These would be the only instances of congruence if coverage were globally applied (Figs 1 and 2). Therefore, relaxing the coverage constraint yields 52 additional instances of higher-level node congruence. Similarly, the 2015./2014.Telluraves partition has 231 aligned regions– 194 without the "no coverage" regions–of which 38 are congruent. This corresponds to an increase of 34 regions, compared to four congruent species-level concept regions present under strict coverage. Correcting for the redundant 2015./2014.Telluraves region, we 'gain' 85 congruent parent node regions across the two phylogenies if node identity is encoded intensionally (Figs 3 and 4). Indeed, this approach yields the intuitive articulation 2015.Neornithes = = 2014.Neornithes at the highest level.
We now focus on characterizing the conflict between 2015.PEA and 2014.JEA. Phylogenomic incongruence can be divided into two general categories: (1) differential granularity or resolution of clade concepts (RCC–5 translation: < or >), and (2) overlapping clade concepts (RCC–5 translation: ><).
The first of these is less problematic from a standpoint of achieving integration: for a given alignment subregion, the more densely sampled phylogeny will entail additional, more finely resolved clade concepts in comparison to its counterpart. Typically, this distinction belongs to the phylogeny of 2015.PEA, due to the 4:1 ratio of terminals sampled. There are 83 above species-level clade concepts sec. 2015.PEA that can be interpreted as congruent refinements of the 2014.JEA topology (see S10 and S11 Files). Conversely, only two such instances of added resolution are contributed by 2014.JEA: (1) 2014.Passeriformes_Clade3 which entails 2014.Passeridae and 2014.Thraupidae; and (2) 2014.Haliaeetus with two subsumed species-level concepts. Nevertheless, the joint 97 congruent node regions and 85 refining node regions cover a large section of the alignment where integration is either reciprocally (= =) or unilaterally (< or >) feasible.
The remaining 38 instances of overlapping articulations between constitute the most profound conflict. These instances are clustered in four distinct regions, i.e.: 2015./2014.Pelecanimorphae (8 overlaps; Fig 7 and S12 File); 2015.Passeri/2014.Passeriformes_Clade2 (3 overlaps; Fig 8 and S13 File); 2015.Eutelluraves/2014.Afroaves (1 overlap; Figs 9 and 10, and S14 and S15 Files); and finally, 2015./2014.Neoaves (26 overlaps; Figs 11–13, and S16–S18 Files). We will examine each of these in sequence.
1. 2015./2014.Pelecanimorphae. The two author teams sampled four family-level concepts congruently for this alignment region (Fig 7). However, 2015.PEA's phylogeny entails six additional family-level concepts that have no apparent match in 2014.JEA. Moreover, the latter authors recognize only one order-level concept, 2014.Pelecaniformes, under which all four family-level concepts are subsumed, including 2014.Phalacrocoracidae. In contrast, 2015.PEA infer an intensionally less inclusive concept of 2015.Pelecaniformes, and place their congruent 2015.Phalacrocoracidae in the order-level concept 2015.Suliformes. This is the first instance where we may plausibly reject the proposition: "Had 2014.JEA sampled 2014.Phalacrocoracidae, they would have assigned this concept to 2014.Suliformes". The assertion is no longer counter-factual: 2014.JEA did sample the corresponding child concept (2014.Phalacrocoracidae), but did not assign it to a parent concept separate from 2014.Pelecaniformes. Accordingly, we obtain three overlapping, 'cascading' articulations between concepts that form the 2015.Suliformes higher-level topology and 2014.Pelecaniformes. Meanwhile, the uniquely sampled 2015.Ciconiiformes are subsumed under 2014.Pelecanimorphae with relaxed parent coverage.
Within 2015.Pelecaniformes, we obtain five additional overlapping articulations between five concepts that make up the 2015/2014 supra-familial topologies in this alignment (Fig 7). This conflict is due to the differential assignment of 2015./2014.Pelecanidae. Specifically, 2015.PEA inferred a sister relationship of 2015.Pelecanidae with 2015.Balaenicipitidae, for which 2014.JEA have no sampled match. Meanwhile, the latter authors inferred a sister relationship of 2014.Pelecanidae with 2014.Ardeidae. The latter concept is matched in 2015.PEA with 2015.Ardeidae, though not as the most immediate sister concept of 2015.Pelecanidae. Of course, we may posit that a 2015.Ardeidae/2015.Pelecanidae sister relationship is what 2015.PEA would have obtained, had these authors not also sampled 2015.Balaenicipitidae and 2015.Scopidae. But they did, and hence obtained two clade concepts that include 2015.Pelecanidae yet exclude 2015.Ardeidae; i.e., 2015.Pelecanoidea_Clade1 and 2015.Pelecanoidea_Clade2. While relaxing parent coverage for 2014.Pelecaniformes_Clade2 could serve to mitigate this conflict, we deem the overlapping relationship to better represent 2015.PEA's phylogenomic signal, which happens to 'break up' the lowest supra-familiar clade concept supported by 2014.PEA.
2. 2015.Passeri/2014.Passeriformes_Clade2. This alignment region is another instance where relaxing parent coverage can only partially mitigate conflict (Fig 8). In this case, 2015.PEA and 2014.JEA sampled two sets of family-level concepts that are wholly exclusive of each other, except for 2015./2014.Corvidae. Regarding the only two additional family-level concepts recognized in 2014.JEA–i.e., 2014.Passeridae and 2014.Thraupidae–we may posit counter-factually that these would be subsumed under 2015.Passeroidea with relaxed coverage [47]. However, further assertions of congruence are difficult to justify, given the limited sampling of 2014.JEA. Thus, in our current representation, 2014.Passeriformes_Clade2 shows an overlapping relationship with 2015.Passeroidea, its immediate parent 2015.Passerida, and also with 2015.Corvoidea.
3. 2015.Eutelluraves/2014.Afroaves. A single overlap occurs just within the congruent parent concepts 2015./2014.Telluraves (Fig 9). Two levels below this paired parent region, both author teams recognize three congruent children; viz. 2015./2014.{Australaves, Coracornithia, Accipitrimorphae/Accipitriformes}. However, 2015.Prum group the former two concepts under 2015.Eutelluraves, with 2015.Accipitriformes as sister; whereas 2014.JEA cluster the latter two concepts under 2014.Afroaves, with 2014.Australaves as sister. This is the first occurrence of conflict that cannot justifiably be resolved by relaxing parent coverage, but instead reflects divergent phylogenomic signals.
How to speak of such overlap? In Fig 9, we utilize clade concept labels that pertain to each input phylogeny. In the resulting alignment, the articulation 2015.Eutelluraves >< 2014.Afroaves is visualized as a dashed blue line between these regions. Yet Fig 9 also specifies the extent of regional overlap at the next lower level. Accordingly, only the region 2015./2014.Coracornithia is subsumed under each of the overlapping parents. This is indicated by the two inclusion arrows that extend 'upward' from this region. The other two paired child regions are respectively members of one parent region.
If we call the input regions 2015.Eutelluraves "A" and 2014.Afroaves "B", we can use the following syntax to identify output regions that result from overlapping input concepts [26]: A*B (read: "A and B") constitutes the output region shared by two parents, whereas A\b ("A, not b") and B\a ("B, not a") are output regions with only one parent. We call this more granular syntax split-concept resolution ("merge concepts" in [26]), as opposed to whole-concept resolution which preserves the syntax and granularity provided by the input concept labels.
In Fig 10, the 2015./2014.Telluraves overlap is represented with split-concept resolution. This eliminates the need to visualize a dashed blue line between 2015.Eutelluraves and 2014.Afroaves (Fig 9). Moreover, in this case the split-concept resolution syntax is redundant or unnecessary, because each of the three resolved regions under "A" (2015.Eutelluraves) and "B" (2014.Afroaves) is congruent with two regions already labeled in the corresponding input phylogenies. We will see, however, that this granular syntax is essential for verbalizing the outcomes of more complex alignments that contain many overlapping regions.
4. 2015./2014.Neoaves. The remaining 26 instances of overlap are shown under different alignment visualizations in Figs 11–13. They occur 1–5 levels below the congruent concept pair 2015./2014.Neoaves, and jointly make up the primary region of conflict between these reconstructions. Because parent coverage was already and selectively applied at lower levels, none of the 26 overlaps in the alignment are caused by differential child sampling. Therefore parent coverage must hold here, resulting in genuine conflict in the higher-level arrangement of congruent sets of children.
Our Fig 11 is intended to be an RCC–5 alignment analogue to Fig 1 in [3]. The alignment reaches from the root to the ordinal level, and to the family level in the two subregions where order-level concepts are conflicting (see Fig 4 and S12 File). The visualization provides an intuitive signal of the distribution of in-/congruence throughout the alignment. In all, 66/111 regions (59.5%) are congruent, of which 22 are located in the 2015./2014.Telluraves; 15 are contained in the 2015./2014.Ardeae; and 5 are part of the 2015./2014.Columbimorphae. Outside of the 2015./2014.Neoaves, 8 such regions are present. In other words, the two phylogenies are congruent at the highest level and also in several intermediate regions above the ordinal level.
Fig 12 shows just the neoavian explosion region under whole-concept resolution. Each phylogeny contributes 21 input concepts to this 'zoomed-in' alignment, which yields 13 congruent regions. Of these, only 2015./2014.Neoaves and 2015./2014.Otidimorphae represent non-terminal concepts.
Unpacking the complexity of this conflict region requires a stepwise analysis. From the perspective of 2015.PEA, the 2015.Neoaves are split into a sequence of three unnamed, higher-level clade concepts, i.e. 2015.{Neoaves_Clade1, Neoaves_Clade2, Neoaves_Clade3}, with 2015.{Strisores, Columbaves, Gruiformes} as corresponding sister concepts. The two children of 2015.Neoaves_Clade3 are 2015.{Aequorlitornithes, Inopinaves}. The authors accept the nomenclature of [44] for 2015.Strisores, with is congruent with 2014.Caprimulgimorphae; and the region 2015./2014.Gruiformes is congruent as well. However, the remaining six high-level concepts of 2015.PEA are in conflict with the two highest-level neoavian concepts of 2014.JEA, i.e. 2014.{Columbea, Passerea}, and also with any of the four unnamed clade concepts below 2014.Passerea. In particular, the node sequence 2015.{Neoaves_Clade3, Aequorlithornites, Aequorlithornites_Clade1} participates in 16/26 overlaps, as summarized in Table 3. Loosely corresponding to this sequence are the concepts 2014.{Passerea_Clade1, Passerea_Clade2, Cursorimorphae}, jointly with 10 overlaps. These overlaps are grounded in the incongruent assignment of five paired, lower-level concept regions; viz. 2015./2014.{Ardeae, Charadriiformes, Opisthocomiformes, Phoenicopterimorphae, Telluraves}. Two conflicting placements contribute most to the number of overlaps: (1) 2015./2014.Charadriiformes in 2015.Aequorlithornites_Clade1 (sister to 2015.Phoenicopterimorphae) versus 2014.Cursorimorphae (sister to 2014.Gruiformes); and (2) 2015./2014.Phoenicopterimorphae in 2015.Aequorlithornites_Clade1 versus 2014.Columbea (sister to 2014.Columbimorphae). The newly proposed yet unnamed 2015. Aequorlithornites_Clade1, consisting of certain "waterbirds", in effect causes the most topological incongruence with 2014.JEA. This concept, together with its four superseding parents, 'triggers' 20/26 overlaps with the phylogenomic tree of 2014.JEA.
Two additional clusters of conflict are identifiable in Fig 12. The first concerns the alignment of the two concepts 2015.Inopinaves and 2014.Passerea_Clade2, which share the child regions 2015./2014.Telluraves, yet which differentially accommodate the congruent regions 2015./2014.Ardeae and 2015./2014.Opisthocomiformes. This further contributes to the abundance of overlaps along the respective 2015.Neoaves_Clade{1–3}/Aequorlithornites/_Clade1 and 2014.Passerea/_Clade{1–3}/Cursorimorphae chains. Second, the two paired regions 2015./2014.Columbimorphae and 2015./2014.Otidimorphae are incongruently assigned to three overlapping parents, i.e. 2015.Columbaves and 2014.{Columbea, Passerea_Clade4}. From the perspective of 2015.PEA, 2014.JEA's bifurcation of 2014.Columbea and 2014.Passerea is the most conflicting, as these two concepts participate in 11 overlaps. A third and more minor incongruence concerns the placement of three concept regions within the 2015./2014.Oditimorphae.
In Fig 13, the same 'zoomed-in' alignment is shown under split-concept resolution. This permits identifying all output regions created by the 26 overlaps of the neoavian explosion (see Table 4). The entire set consists of 78 labels; i.e., 26 labels for each split-resolution product {A*B, A\b, B\a} for one instance of input region overlap. Not all of these split-concept resolution labels are semantically redundant with those provided in the input. Specifically, 51 labels are generated 'in addition' for the 12 terminal congruent regions (compare with Fig 12). These are indeed unnecessary synonyms for regions already identified in the input. However, the relative number of additional labels generated per input region is telling. This number will be highest for regions whose differential placements are the primary drivers of incongruence. As explained above, these are: 2015./2014.{Phoenicopterimorphae, Charadriiformes, Columbimorphae}, respectively with 14, 8, and 7 additional labels. Six redundant split-concept resolution labels are further produced for input regions that are unique to one phylogeny; e.g., 2014.Columbea is also labeled 2015.Neoaves_Clade1 \ 2014.Passerea (where the "\" means: not).
The remaining 21 split-concept resolution labels identify 15 salmon-colored alignment regions– 11 uniquely and 4 redundantly with 2–3 labels each–for which there are no suitable labels in either of the phylogenomic input trees (Table 4). Forty-six additional articulations are inferred to align these regions to those displayed in Fig 12. Although these novel regions are not congruent with any clade concepts recognized by the source phylogenies, they are needed to express how exactly the authors' respective clade concepts overlap.
Three distinct reference services are gained by generating the split-concept resolution labels. First, in cases where no whole-concept resolution labels are available, we obtain appropriately short and consistent labels to identify the split regions caused by overlapping clade concepts. Second, the {A*B, A\b, B\a} triplets have an explanatory function, by using the same syntactic set of input labels (A, B) to divide complementary alignment subregions of an overlap. If we focus on one label of a triplet, we can find the two complements, and thereby systematically explore the 'reach' of each split in the alignment. Third, the clade concept labels (A, B) used in the split-concept resolution labels will be exactly those that identify overlapping regions across the source phylogenies.
We can now also ask to what extent the clade names (syntax) used by the two author teams succeed or fail to identify congruent and incongruent concept regions (semantics). Such name:meaning (read: "name-to-meaning") analyses were carried out in three previous alignment use cases, with rather unfavorable outcomes for the respective names in use [14, 32, 51]. Here, based on the alignment of Fig 11, the 97 x 83 input concepts yield a set of 8,051 MIR (S16 File). Of these, 384 MIR involve one of four "no coverage" regions added to 2014.JEA concepts. We therefore restrict the name:meaning analysis to the remaining 7,667 MIR (Table 5).
Interestingly, the clades names used by the respective author teams fare rather well. Only nine of 7,667 pairings in the MIR (0.12%) are unreliable as identifiers of in-/congruence of the respective RCC–5 articulation. In seven instances, two congruent concepts have different names. Four of these merely involve changes in name endings, viz.: 2015.Accipitriformes = = 2014.Accipitrimorphae, 2015.Galloanserae = = 2014.Galloanseres, 2015.Gaviiformes = = 2014.Gaviimorphae, and 2015.Pteroclidiformes = = 2014.Pterocliformes. The other three instances involve the respectively preferred roots 2015.Strisores and 2014.Caprimulgi-{formes, morphae}. The articulation 2015.Pelecaniformes < 2014.Pelecaniformes is the single instance in which the meaning of the same name is less inclusive in one source (Fig 7). Lastly, the overlapping relationship 2015.Otidimorphae_Clade1 >< 2014.Otidimorphae_Clade1 involves the same name (Figs 12 and 13), though it is not actually used by the author teams (see Methods).
In summary, the clade concept names used by 2015.PEA and 2014.JEA rarely provide an incorrect signal regarding in-/congruence. This desirable outcome seems to reflect their recognition that newly inferred clade concepts merit the use of unique names.
We now compare these results with conflict analysis and visualization tools created for the Open Tree of Life project (OToL)–a community-curated tree synthesis platform [13, 22, 23, 24]. The OToL approach is explained in [11, 15, 23, 52, 53]. The method starts off with 'normalizing' all terminal names in the source trees to a common taxonomy [24]. Having the same terminal name means taxonomic concept congruence (= =). To assess conflict from the perspective of one rooted input tree (A), a source edge j of that tree is taken to define a rooted bipartition S(j) = Sin | Sout, where Sin and Sout are the tip sets of the ingroup and outgroup, respectively. The algorithm progresses sectionally from the leaves to the root. Concordance or conflict for a given edge j in tree A with that of tree B is a function of the relative overlap of the corresponding tip sets, as follows [23]. Concordance between two edges in the input trees A and B is obtained when Bin is a proper subset (⊂) of Ain and Bout ⊂ Aout. On the other hand, two edges in trees A and B are conflicting if none of these sets are empty: Ain intersects (⋂) with Bin, Ain ⋂ Bout, or Bin ⋂ Aout. In other words, conflict means that there is reciprocal overlap in the ingroup and outgroup bipartitions across the two trees.
We applied this approach in both directions, i.e. starting with 2014.JEA as primary source and identifying edges therein that conflict with those of 2015.PEA, and vice-versa. The visualizations are shown in Figs 14 and 15, respectively.
Most of the red edges in Fig 15, which is based on the more densely sampled tree sec. 2015.PEA, are consistent with the overlapping RCC–5 relationships shown in Figs 7 to 13. However, within the 2015.Pelicanimorphae, certain RCC–5 overlaps (Fig 7) are not recovered ("false positives"). In addition, numerous edges within the 2015.Passeriformes are shown as conflicting ("false negatives") but are congruent refinements based on the RCC–5 alignment (Fig 8).
Using the less densely sampled tree sec. 2014.JEA as the base topology creates is instructive (Fig 14). Here, a much larger subset of the topology 'backbone' is inferred by the OToL algorithm as conflicting–an outcome that would appear inconsistent. For instance, 2014.{Neoaves, Ardeae, Coracornithia} are shown as conflicting edges in Fig 14, when 2015.{Neoaves, Ardeae, Coracornithia} are concordant edges in Fig 15. The inconsistencies are caused by the addition of terminals sec. 2015.PEA that have no matches in 2014.JEA's sampled tips and tree, and will therefore attach as children to a higher-level parent in the OToL taxonomy. The latter is used to place terminals that are differentially sampled between sources. For instance, 2015.Ciconiiformes–which has no close match in 2014.JEA–may end up attaching as a child of 2014.Neognathae instead of 2014.Pelecanimorphae (Fig 7). Hence the OToL taxonomy is used to represent concept intensionality, but it cannot do so reliably if it lacks relevant input concepts. At the time of analysis, the OToL taxonomy lacked a name/concept for "Neoaves". This means that the 2015./2014.Neoaves ingroup/outgroup bipartitions will be inconsistent in evaluating the placement of 2015.Ciconiiformes, showing conflict in Fig 14 but not in Fig 15.
We review the key conventions of our approach before discussing services that can be derived from our alignments.
What can we gain from this approach, both narrowly for this use case and for future data integration in systematics?
Data representation designs have inherent trade-offs. Unlike other semi-/automated phylogenomic conflict visualization methods [13, 23, 24], the above approach requires extensive upfront application of human expertise to obtain the intended outcomes. In return, the RCC–5 alignments deliver a level of explicitness and verbal precision exceeding that of published alternatives [4, 5, 6, 9, 16, 17]. We can not just verbalize all instances of congruence and conflict, but transparently document and therefore understand their provenance in a global alignment (Figs 11 and 13). In other words, the RCC–5 alignments provide a logically tractable means to identify and also explain the extent of conflict.
We can derive novel data services from the alignment products. (Note that these services are envisioned but not yet implemented in a web-based platform.) Example queries include the following. (1) Show all congruent regions of the alignment and their clade concept labels. (2) Modify this query to only apply to alignment regions subsumed under one particular concept and source, such as 2014.Columbea. (3) For any subset region of the global alignment (e.g., 2015./2014.Australaves), show the lowest-level pairs of children that are sampled congruently, versus those that are sampled incongruently. (4) Highlight within such an alignment region all clade concepts for which parent coverage is relaxed, and which show congruence as a result of this action. (5) Highlight sets of concepts where incongruence is due to differential granularity (sampling), versus actual overlap. (6) Identify and rank concepts that participate in the greatest number of overlapping relationships (Table 3). (7) Identify and rank the longest chains of nested, overlapping concept sets (Fig 12). (8) Highlight the congruent, lowest-level concept pairs whose incongruent placement into higher-level regions causes the chains of overlap. (9) List all split-concept resolution labels in complementary triplets {A*B, A\b, B\a}, and provide for each the two immediate children and (again) the set of lower-level, whole-concept resolution regions that are differentially distributed by the split (Fig 13 and Table 4). (10) Identify clade names that are unreliable across the source phylogenies; including identical clade name pairs that participate in concept labels with an incongruent relationship, or different clade names whose concept labels have a congruent relationship (Table 5).
All of the above queries, and many others we could propose, are enabled by our RCC–5 representation and reasoning conventions, which therefore present a new foundation for building logic-based, machine-scalable data integration services for the age of phylogenomics. Conceptualizing node identity and congruence this way addresses a gap in current systematic theory that is not adequately filled by other syntactic solutions.
Linnaean naming. We have shown elsewhere that homonymy and synonymy relationships are unreliable indicators of congruence [14, 26, 32]. Code-enforced Linnaean naming is designed to fixate the meaning of names by ostension, while allowing the intensional components to remain ambiguous [21, 54, 55, 56, 57]. This trade-off effectively shifts the burden of disambiguating varying intensionalities associated with Linnaean names onto an additional, interpreting agent–typically human experts. Our RCC–5 alignment approach can be viewed as a way to formalize the disambiguation effort, so that it can attain machine-interpretability.
Phyloreferencing. Similarly, node-based phyloreferences [58, 59, 60] are not well suited to reconstruct an alignment such as that of 2015./2014.Pelecanimorphae (Fig 7). This would require: (1) an elaborate notion of phyloreference homonymy and synonymy (e.g., 2015.Pelecanifores versus 2014.Pelecaniformes, or 2015.Strisores versus 2014.Caprimulgimorphae); (2) node-based definitions with inclusion/exclusion constraints that cover all terminals in the phylogeny; and (3) synapomorphy-based definitions at higher levels to model the local relaxation of coverage constraints. All of these functions may be feasible in principle with phyloreferences, provided that human experts are permitted to enact them. However, it may be fair to say that phyloreferences were not mainly designed to bring out fine differences between node concepts across multiple phylogenies. They are best utilized when concept evolution and conflict are not the main drivers of an information system design.
The two largest alignments of 2015./2014.Neornithes (without) / 2015./2014.Telluraves jointly entail 895 concepts and 95 instances of relaxed parent coverage. They provide us with 97 congruent regions in the global alignment, of which 85 regions are obtained only because of the indirect modeling of intensional node definitions. The contingency of the alignment outcome on expert intentions is neither surprising nor trivial. We should therefore explore this dependency more deeply.
Redelings and Holder [23: pp. 5–6] comment on the OToL synthesis method: "Any approach to supertree construction must deal with the need to adjudicate between conflicting input trees. We choose to deal with conflict by ranking the input trees, and preferring to include edges from higher-ranked trees. The merits of using tree ranking are questionable because the system does not mediate conflicts based on the relative amount of evidence for each alternative. […] In order to produce a comprehensive supertree, we also require a rooted taxonomy tree in addition to the ranked list of rooted input trees. Unlike other input trees, the taxonomy tree is required to contain all taxa, and thus has the maximal leaf set. We make the taxonomy tree the lowest ranked tree. […] Our method must resolve conflicts in order to construct a single supertree. However, the rank information used to resolve conflicts is an input to the method, not an output from the method. We thus perform curation-based conflict resolution, not inference-based conflict resolution."
Clearly, the outcomes of the OToL synthesis method are also deeply dependent on expert input regarding the relative ranking of input phylogenies and of the OToL taxonomy [24]. We have shown (Figs 14 and 15) that these choices can lead to inconsistent outcomes whenever the sequence of input trees determines how concordance and conflict are negotiated by the algorithms. If the less densely sampled tree is prioritized, and the taxonomy cannot accommodate all components of a lower-ranked tree, then the method will show more conflict in comparison to an inverse input sequence. Any global rule of priority among trees is a poor proxy for modeling individual node concept intensionality, which requires making reliable, local decisions between (1) conflict due to differential granularity versus (2) conflict due to overlap.
We can now return to the challenge posed in the Introduction. How do we build a data service for phylogenomic knowledge in the face of persistent conflict? Our answer is novel in the following sense. Assuming that such a service is desirable, we show that achieving it fundamentally depends on making and expressing upfront empirical commitments about the intensionalities of clade concepts whose children are incongruently sampled. Without embedding these judgments into the alignment input, we lose the 85 congruent parent regions recovered under relaxed parent coverage. We furthermore lose the ability to distinguish the former from more than 340 alignment regions that are not congruent. And we lose the power to express the nature of this residual conflict–granularity versus overlaps–and how to resolve it.
In other words, the first step for building the phylogenomic data knowledge service will be to recognize that conceptualizations of node identity within such a system just cannot be provided through some mechanical, 'objective' criterion. Instead, we need an inclusive standard of objectivity that embraces trained judgment as an integral part of identifying and linking node concepts [30]. In that sense, phylogenomic syntheses are inference-based (contra [23]) and also driven by a specific purpose. As integrative biologists, our goal in providing RCC–5 alignments is to maximize intensional node congruence. There may not be a more reliable criterion for achieving this than expert judgment, which draws on complex and context-specific theoretical knowledge [40, 43, 61]. Logic representation and reasoning can help render these constraints explicit and consistent, and expose implicit articulations through the MIR which encompass all node concepts in an alignment. But logic cannot substitute the expert aligners' intensional aims and definitions.
Building a phylogenomic data knowledge service forces us to become experts about externally generated results that conflict with those which we may (currently) publish or endorse. We need to become experts of another author team's node concepts, to the point where we are comfortable with expressing counter-factual statements regarding their intensionalities, in spite of incongruent child sampling. This will require a profound but necessary adjustment in achieving a culture of synthesis in systematics that no longer manages conflict this way: "If we do not agree, then it is either our view over yours, or we just collapse all conflicting node concepts into polytomies". In contrast, we need to develop the following culture of synthesis: "We may not agree with you, but we understand your phylogenomic inference well enough to express our dis-/agreements in a logic-compatible syntax. Therefore, we are prepared to assert and refine articulations from our concepts to yours for the purpose of maximizing intensional node congruence". Only then can we expect to also maximize the empirical translatability of biological data linked to diverging phylogenomic hypotheses.
Shifting towards the latter attitude will be more challenging than providing the operational logic to enable scalable alignments. Automation of certain workflow components is certainly possible. Ultimately, the logic or technical issues are not the hardest bottlenecks to overcome. Designers of future data environments capable of verbalizing phylogenomic conflict and synthesis need to reflect on how to promote a culture where experts routinely re-/assess the intensionalities of node concepts published by peers. If we wish to track progress and conflict across phylogenomic inferences, we first need to design a value system that better enables and motivates experts to do so.
He we discuss various reviewer comments that merit a response but would break up the main flow of the narrative if inserted earlier. We take liberty to assign a header to each comment.
Phylogenetic clade definitions and taxonomic concepts are fundamentally mismatched. One reviewer pointed out that clade hypotheses are about branching patterns and relationships of descent, and therefore are mismatched with our notion of node intensionality. We disagree in the following sense. We believe that we are not conflating two fundamentally different kinds of clade conceptualizations, as much as bringing out with the RCC–5 alignments one aspect in the dual, or hybrid nature of clade concepts. The latter are not either this or that–with parallels to the taxa as classes-versus-individuals literature–both can be both, depending on the pragmatic interest [36, 37, 62]. For the purpose of synthesis and integration, modeling the intensional aspect of clade concepts is critical. We see this purpose reflected (e.g.) in the matching of high-level terminals in [3].
No mechanism for quantitatively expressing uncertainty about tree topology. The same reviewer pointed out that we select single point estimate topologies for each author team, thereby not accounting for the complex likelihood surfaces of the reconstructions and the relative uncertainty of each topology. Applied to what we show here, this criticism is valid. However, it would be feasible perform RCC–5 alignments on a cluster of paired topology alternatives with similar likelihood values. The products can be compared in order to manage uncertainty, through identification of stable versus variable regions across multiple alignments. If most of the variation occurs at higher levels, this would mean that the vast majority of our low-level RCC–5 input articulations could be reused.
Phylogenetic conflict is not limited to two trees. Another reviewed pointed out the need to align more than two phylogenies in situations where many recent reconstructions are available to inform a synthesis [5, 6, 11]. While the current logic toolkit handles three or more input trees in principle, there certainly are unrealized opportunities to model transitive relationships (example: for concepts A, B, C in the input trees T1, T2, T3: if AT1 = = BT2 and BT2 = = CT3 then AT1 = = CT3). 'Smartly' breaking down alignments of three or more trees while exploiting transitive relationships, as well as visualizing the outcomes accessible ways, are important future improvements for this approach.
"Not every clade [concept] is worth labeling and discussing". We can agree with that assessment. But, having a framework to do so is critical to evaluating the feasibility of a phylogenomic data knowledge service, and should not trail behind discussions regarding its desirability. If we have no formalized means of translating Fig 1 of [3] into a machine-accessible language (Fig 11), then we cannot fully understand the costs and benefits of building the service.
Incentivizing alignment production. One reviewer pointed out that efforts to align multiple trees are costly, and inquired about our suggestions for incentivizing such expert contributions. An initial answer would point to the creation of an e-journal, where multi-phylogeny and -taxonomy alignments can be published either as stand-alone articles or in association with separate publications of new tree reconstructions. The platform of a formal journal best responds to expert needs to receive academic credit [63]. Knowledge systems such as [64] could represent the information input and output. The most valuable product of such an e-journal are the expert-vetted sets of RCC–5 articulations, which represent a new kind of "systematic intelligence". Scientists and commercial publishers may utilize this intelligence to improve the precision and recall of systematically structured data [54], where business models would focus on the latter clients for revenue. Needless to say, these are ideas that will take time to concretize and test.
|
10.1371/journal.ppat.1000627 | Probing the HIV-1 Genomic RNA Trafficking Pathway and Dimerization by Genetic Recombination and Single Virion Analyses | Once transcribed, the nascent full-length RNA of HIV-1 must travel to the appropriate host cell sites to be translated or to find a partner RNA for copackaging to form newly generated viruses. In this report, we sought to delineate the location where HIV-1 RNA initiates dimerization and the influence of the RNA transport pathway used by the virus on downstream events essential to viral replication. Using a cell-fusion-dependent recombination assay, we demonstrate that the two RNAs destined for copackaging into the same virion select each other mostly within the cytoplasm. Moreover, by manipulating the RNA export element in the viral genome, we show that the export pathway taken is important for the ability of RNA molecules derived from two viruses to interact and be copackaged. These results further illustrate that at the point of dimerization the two main cellular export pathways are partially distinct. Lastly, by providing Gag in trans, we have demonstrated that Gag is able to package RNA from either export pathway, irrespective of the transport pathway used by the gag mRNA. These findings provide unique insights into the process of RNA export in general, and more specifically, of HIV-1 genomic RNA trafficking.
| High genetic diversity of HIV-1 presents a difficult barrier for drug treatment and vaccine development. HIV-1 efficiently reassorts mutations in its genome by packaging two copies of RNA into one virion and using portions of each RNA as the template for DNA synthesis. Thus, a key to HIV-1's genetic diversity is how it packages two RNAs into one virion. We have previously shown that HIV-1 packages RNA as a dimer, rather than two monomers. In this report, we determined that HIV-1 RNA finds its copackaged dimer partner mostly in the cytoplasm, not the nucleus, of the infected cells. We also found that the host cell machinery used to transport the HIV-1 RNA out of the nucleus affects the ability of two RNAs to copackage into one virion: HIV-1 RNAs transported using the CRM-1 pathway does not copackage efficiently with RNA transported by the NXF1 pathway. This is also the first demonstration that RNAs transported by the two major cellular pathways, CRM-1 and NXF1, are partially segregated at some point in the cytoplasm. Therefore, this work provides novel insights to both HIV replication mechanisms and cellular RNA transport pathways.
| HIV-1 full-length RNAs serve at least two functions: as a template for Gag/Gag-Pol translation, and as genetic material packaged in the virion. Many cellular factors ensure the correct macromolecular trafficking between nucleus and cytoplasm; specifically, mechanisms exist to prevent the export of intron-containing transcripts, such as the full-length HIV-1 RNA [1],[2]. Most cellular mRNAs are fully spliced before export and many are believed to exit the nucleus via the NXF1-dependent pathway [3]. However, many proteins and some RNAs use an alternative, CRM-1-dependent pathway to migrate out of the nucleus [3]. The extent to which these two pathways are linked or overlap is currently unknown, and the reason for their differential use is subject to speculation. It is interesting to note that members of the Retroviridae family can use different transport pathways for the export of their intron-containing RNA. Some viruses, such as Mason-Pfizer monkey virus (MPMV), use the NXF1 pathway by binding the NXF1 protein directly to its RNA via a structured motif (constitutive transport element or CTE) [4]. Other viruses, such as HIV-1, use the CRM-1 pathway by indirectly linking the CRM-1 protein to the viral RNA via a virally encoded protein. In HIV, the viral protein is Rev and it acts as a bridge between a structured RNA motif (the Rev-response element (RRE)) and the CRM-1 protein [5],[6]. Additionally, some retroviruses use pathways that have yet to be identified [7].
Many studies have shown that it is possible to alter the retroviral RNA transport pathway by manipulating the cis- and trans-acting viral elements. Interestingly, a recent study indicates that the transport pathway the RNA takes can influence the function of the protein it encodes. By altering the transport pathway of the HIV-1 Gag-encoding RNA from CRM-1 to NXF1, one can relieve the HIV-1 assembly block in murine cells [8]. Intriguingly, Gag was expressed efficiently in murine cells by RNA using either pathway. However, efficient assembly only occurred when Gag was translated from RNA using the NXF1 pathway; Gag translated from the CRM-1 pathway was mistargeted and could not assemble efficiently [9]. These studies revealed that the transport pathway may play a far more intricate role during viral replication than we previously realized.
Retroviruses package two copies of full-length RNA in one virion. One of the consequences of packaging two RNAs is frequent recombination during DNA synthesis when reverse transcriptase uses parts of both RNAs as templates [10],[11]. Although frequent recombination can occur during DNA synthesis of all virions, a genotypically different recombinant can only be generated from virions that package two different RNAs (heterozygous virions) [10]. Using a recombination assay, we have shown that RNA molecules derived from two similar HIV-1 proviruses can randomly assort and be efficiently copackaged into virions [12],[13]. However, heterozygous virions are formed less efficiently when the two proviruses contain variations in their dimerization initiation signal (DIS). Located at the loop of stem-loop 1 of the 5′ untranslated region, the DIS is a 6-nt palindromic sequence that forms the initial interaction between the two HIV-1 RNAs [14]. The Gag polyproteins of HIV-1 interact with, and specifically package, the viral RNA to generate infectious viruses. We have previously examined whether RNA dimerization occurs prior to virus assembly using HIV-1 variants with DIS mutations that abolish their palindromic nature (for example, from GCGCGC to GGGGGG) but can form perfect base pairs with the DIS of a partner virus (such as a virus with CCCCCC at the DIS). We reasoned that in the coinfected cells if dimeric RNAs are packaged, then the GGGGGG viral RNA would preferentially pair with CCCCCC viral RNA, and we would therefore observe an increase in the formation of heterozygous virions. In contrast, if two monomeric RNAs are packaged, then we would not observe an increase in heterozygous viruses. Our results revealed that most of the virions from coinfected cells were heterozygous, indicating that copackaged RNA partner selection, i.e. dimerization, occurred prior to the packaging of virion RNA [15].
Many questions remain concerning how HIV-1 full-length RNA traffics through the cell to fulfill its various roles. For example, it has been unclear at which point along the RNA trafficking pathway, from the site of transcription to the site of viral budding, HIV-1 RNA selects its copackaged RNA to initiate the dimerization process. It remained unclear whether HIV-1 RNAs are dimerized in the nucleus and traffic out of the nucleus as a dimer, or exit the nucleus as a monomer and dimerize in the cytoplasm. Furthermore, it is not known whether the RNA transport pathway used by the viral RNA affects its interactions with other viral RNA or Gag proteins, and thereby influences the abilities of the RNA to dimerize and be packaged into a nascent virion. In this current work, we addressed these questions by using three systems: a cell-fusion-dependent recombination assay to establish the cellular location of the dimerization event, a Rev-independent HIV-1 to investigate the influence of the NXF1 and CRM-1 export pathways on these processes, and a single virion visualization assay to directly analyze the RNA content of each viral particle. These studies resulted in three novel findings. We demonstrate that the majority of HIV-1 RNA dimerization occurs within the cytoplasm, not within the nucleus. Moreover, the RNA transport pathway taken by the viral RNA has a major effect on the ability of the RNA to be copackaged; these results revealed that RNAs trafficking through the CRM-1 and NXF1 transport pathways are partially separated at least at the stage of RNA commitment to dimerization. Finally, we demonstrate that although the RNA pathways are partially distinct at the point when initial dimerization occurs, selection of packaged RNA by Gag does not discriminate between RNA in these two pathways. These findings provide insights into the mechanisms of HIV replication including RNA-RNA and RNA-Gag interactions, and demonstrate a new approach to the unraveling of the detailed mechanisms of the cell's nuclear-cytoplasmic trafficking pathways.
Our previous experiment revealed that the selection of the copackaged RNA occurs prior to encapsidation into the virion [15]. In the current study, we sought to determine the subcellular location where RNA-RNA interactions take place to initiate the dimerization process. To do this, we developed a cell fusion-based recombination assay to examine whether RNA partner selection is initiated in the cytoplasm. Our previously established recombination assay uses dual-infected cell lines to provide accurate and reproducible measurements of recombination between two viruses [13]. In this system, the near-full-length HIV-1 genome encodes two markers in nef, an inactivated green fluorescent protein gene (gfp) and a second gene that is either a mouse heat-stable antigen gene (hsa) or a mouse thy1.2 gene (thy) (Fig. 1). The two inactivating mutations of gfp in these two viruses are separated by a distance of 600 bp. A proportion of the viral particles released from the dual-infected cell lines contain one RNA from each provirus (heterozygous viruses); recombination between which can occur during reverse transcription to generate a recombinant progeny expressing a functional gfp. The multiplicity of infection (MOI) of the GFP+ events, as a percentage of the total infection MOI (determined using the expression of HSA and Thy), provides a measure of the recombination efficiency between the two viruses. As retroviral recombination occurs between two copackaged RNAs, the rate of recombination is dependent on the formation of heterozygous virions; therefore, recombination can be used as a measurement of the dimerization efficiency between two viral RNAs, as we have previously shown [15].
We modified the recombination system to examine whether we could detect cytoplasmic RNA dimerization events (Fig. 2). In this system, two cell lines are generated, each infected with a gag mutant virus that has a severe replication defect. When the two cell lines are fused using conditions that facilitate cytoplasmic but not nuclear fusion, the Gag proteins from these two viruses can coassemble and complement each other's function to rescue virus replication. Our previous study demonstrated that HIV-1 RNA dimerization occurs prior to Gag packaging [15], and we envision two differential outcomes that would be reflective of the locations of the dimerization event (Fig. 2A). If the selection of the copackaged RNA mostly occurs in the nucleus, the majority of the viruses generated from this system will contain two RNAs derived from the same provirus (homozygous virions) that have the same inactivating gfp mutation. Therefore, very few GFP+ phenotypes will be formed by recombination. In contrast, if the selection of the copackaged RNA mostly occurs in the cytoplasm, fusing the cytoplasm of the two cell lines will allow the RNAs derived from the two nuclei to interact and dimerize prior to packaging by Gag. In this scenario, heterozygous virions can be formed, which can generate numerous GFP+ recombinants. Therefore, in the fusion experiment, the location of the RNA dimerization event, in the nucleus or in the cytoplasm, predicts different levels of the recombinant GFP+ phenotype.
We generated cell lines containing either a nucleocapsid (NC) mutant virus (T6-NC*) or a late domain mutant virus (H0-PTAP–). The NC* mutation reduces RNA packaging to 5% of wild-type and abolishes infectivity [16] (Fig. 2B panels II and VI), whereas the PTAP– mutation reduces viral budding and viral titer drops to ∼2% of wild type [17],[18] (Fig. 2B panels III and VII). To facilitate fusion, one cell line was based on modified 293T cells that express CD4 and CCR5 (293T.CC), whereas the other cell line was 293T based and was transiently transfected, prior to fusion, with a plasmid expressing CCR5-tropic HIV-1 Env. Viruses were harvested 16 h after the coculture of the two cell lines; at this time point, cell fusion but not nuclear fusion was visible by microscopy (Fig. S1 and data not shown). Target cells were infected with these viruses, and flow cytometry was performed to measure virus infection and recombination. Our results (Fig. 2C) demonstrate that the recombination rates of infectious viruses produced by cell fusion (5–6%) are similar to that of our standard recombination assay (7%), which uses viruses produced from cells containing two proviruses [13]. To confirm that nuclear fusion did not affect the results of this assay, the cyclin-dependent kinase inhibitor roscovitine was added after mixing the cell lines. Roscovitine was previously shown, in a detailed study of HIV-1-directed cell fusion, to prevent nuclear envelope breakdown and nuclear fusion or karyogamy [19]. We observed similar recombination rates in samples with or without roscovitine treatment (Fig. 2C), indicating that nuclear fusion was not a factor in the high levels of recombination detected. Taken together, these results indicate that most of the RNA dimerization and partner selection process occurs in the cytoplasm of the producer cells.
Having demonstrated that the majority of HIV-1 RNA dimerization occurs within the cytoplasm, we further investigated whether the pathway of entry into the cytoplasm is important for RNA dimerization and packaging. To this end, we generated two HIV-1 variants, H0-CTE and B6-CTE, that use the NXF1, and not the CRM-1, transport pathway by removing the RRE motif and replacing it with four copies of the MPMV CTE [20]. H0-CTE encodes the same HSA and mutated gfp marker as H0-RRE (Fig. 1); B6-CTE encodes a mouse B7 gene and a gfp that has the inactivating mutation located at the 3′ end of the gene (Fig. 1). Using positive control viruses with two functional genes, we have previously showed that the HSA and GFP or Thy and GFP markers are well-expressed and can be simultaneously detected efficiently by flow cytometry [13]; we have also examined the B7 and GFP expression and showed these two markers were detected simultaneously and efficiently by flow cytometry (Fig. S2).
To assess the effect of the nuclear export pathway on HIV-1 RNA dimerization and copackaging, six types of dual-infected cell lines were generated containing either two RRE-dependent proviruses (H0-RRE.T6-RRE; H0-RRE.B6-RRE), two CTE-dependent proviruses (H0-CTE.B6-CTE), or one RRE- and one CTE-dependent provirus (H0-CTE.T6-RRE, H0-CTE.B6-RRE, and H0-RRE.B6-CTE). Flow cytometry analyses of the markers encoded by the two viruses indicated that most of the cells (>90%) in each cell line express both viruses (data not shown). These cell lines were transiently transfected with a plasmid expressing HIV-1 Env and the released viruses were used to infect target cells; the titers generated from these infections are shown in Fig. 3A. In general, the two proviruses in each cell line generated titers within two fold of each other; the only notable exception was H0-RRE.B6-CTE, from which the B6-CTE titer was much lower than that of H0-RRE. We also measured the recombination rates generated from these cell lines; results from three experiments are summarized in Fig. 3B. When two viruses both used the CRM-1 pathway (H0-RRE.T6-RRE and H0-RRE.B6-RRE), recombination rates were 6.2–6.6%, similar to the previously measured 7% rate using the H0-RRE.T6-RRE cell line. When both viruses used the NXF1 pathway (H0-CTE.B6-CTE), the recombination rate was similar to that of the two RRE viruses; these results indicated that HIV-1 RNAs from two proviruses both using the NXF1 export pathway can copackage together efficiently. In contrast, recombination between a CTE-using and an RRE-using HIV-1 is lower than expected from random mixing of the viral genomes. The two viruses from H0-CTE.T6-RRE cell lines have similar titers, and if their RNAs were copackaged together randomly, we would have expected a 6–7% recombination rate; however only 2.5% of the viruses generated GFP+ recombinants. Similar low rates of recombination were also observed in viruses harvested from H0-CTE.B6-RRE cell line and H0-RRE.B6-CTE cell line. Therefore, HIV-1 variants separately transported by CRM-1 and NXF1 pathways recombined less frequently than two variants using the same pathway, either by CRM-1 or by NXF1. These results suggest that HIV-1 RNAs exported by the CRM-1 and NXF1 pathways are not copackaged together efficiently.
To directly examine whether RNA exiting the nucleus by different RNA transport pathways can be copackaged together efficiently, we analyzed the RNA content of the viral particles. Utilizing the specific interactions between RNA binding proteins and RNA secondary structures, we recently developed a method to simultaneously detect two RNA species in virion at single RNA molecule sensitivity [21]. This system uses modified HIV-1 genomes that contain the stem-loop sequences, referred to as MS2SL and BglSL, which are recognized by bacteriophage MS2 coat protein and E. coli antitermination protein BglG, respectively (Fig. 4A). These HIV-1 genomes also express Gag or a Gag-cerulean fluorescent protein (CeFP) fusion protein. When these HIV-1 genomes are coexpressed with the MS2 coat protein fused to a yellow fluorescent protein (MS2-YFP), or the N-terminal fragment of BglG fused to a red fluorescent protein (Bgl-mCherry) (Fig. 4B), the presence of viral RNA can be visualized by the fluorescent signals. To analyze the virion RNA, plasmids expressing the MS2SL- or BglSL-containing HIV-1 genomes were cotransfected with plasmids expressing MS2-YFP or Bgl-mCherry. Viral particles were harvested and visualized; representative images are shown in Fig. S3 and summarized in Table 1. In these experiments, HIV-1 particles are detected by CeFP signals, whereas RNA molecules are detected by YFP or mCherry signals. It has been shown that when coexpressed with wild-type Gag, Gag-GFP fusion protein can coassemble into virus-like particles indistinguishable from wild-type immature particles [22],[23]. Thus, in all RNA visualization experiments, a pair of nearly-identical HIV-1 constructs were used, one expressing Gag-CeFP and the other expressing wild-type Gag; for concise description of these experiments, only the name of the CeFP-tagged construct is used in the text.
In agreement with our previous results [21], when Bgl-mCherry was coexpressed with GagCeFP-BglSL-RRE, most of the CeFP+ particles also have the mCherry signal (>96%; Table 1, Group 1). The mCherry signal is specific to the BglSL in the viral RNA genome; particles derived from HIV-1 with similar structures but containing MS2SL did not have mCherry signals when coexpressed with Bgl-mCherry (Table 1, Group 5). Similarly, MS2-YFP also specifically labeled the HIV-1 genome with MS2SL at high efficiency (∼95%) but not HIV-1 RNA with Bgl SL (Table 1, Group 2 and 6, respectively). When we examined particles derived from HIV-1 genomes using the CTE for export, we observed similar RNA labeling efficiency by the Bgl-mCherry and MS2-YFP proteins (both ∼92%; Table 1, Group 3 and 4).
Examination of viral particles generated from coexpression of the Bgl and MS2 systems can determine the distribution of the viral genomes at a single RNA level [21]. When both BglSL and MS2SL containing HIV-1 genomes used the RRE element for RNA transport, these two RNA species were copackaged together efficiently (Table 1, Group 9); ∼49% of the CeFP+ particles have both the mCherry and YFP signals (expected to be 50% by Hardy-Weinberg equilibrium). Similarly, when both HIV-1 genomes used the CTE element for RNA export, these RNA species were also copackaged together efficiently; ∼42% of the CeFP+ particles have both RNA signals (Table 1, Group 10). This finding agrees with our recombination rate studies that efficient copackaging of HIV-1 RNA is not dependent on the CRM-1 transport pathway; two RNA species transported by the NXF1 pathway can also copackage together efficiently.
We then examined particles generated from coexpression of HIV-1 genomes containing RRE or CTE. As shown in Table 1 (Group 11 and 12), RRE-containing HIV-1 RNA did not copackage efficiently with CTE-containing HIV-1 RNA; most of the CeFP+ particles were labeled with one RNA signal, either mCherry or YFP, and only ∼22% of the CeFP+ particles were labeled with both YFP and mCherry signals regardless of whether the RRE-containing RNA was detected by MS2-YFP or by Bgl-mCherry. These results indicate that the RRE- and CTE-containing RNAs were not copackaged together as frequently as two RRE-containing RNAs or two CTE-containing RNAs. These findings are in agreement with our data demonstrating a decreased recombination rate between RRE- and CTE-containing HIV-1 variants (Fig. 3B).
Taken together, our results show that HIV-1 RNAs exported by the CRM-1 and NXF1 pathways are not copackaged together efficiently; therefore, RNA dimerization occurs at a point where RNAs from the two pathways are partially segregated and not completely mixed. However, both the recombination assay and single virion analyses revealed that the RNAs transported by the different pathways can still be copackaged, albeit less efficiently, indicating a partial overlap of the RNAs from the two pathways at the time of RNA dimerization and packaging.
The transport pathway used by the RNA encoding HIV-1 Gag has been shown to affect the downstream assembly events in rodent cells, indicating a link between RNA trafficking and protein function [8],[9]. Our results showed that RNAs using different export pathways are not copackaged randomly; however, it was unclear whether the use of different RNA export pathways also affects the interaction between the RNA and viral Gag proteins, thereby affecting the selection of packaged RNAs. If the RNA export pathway dictates the location of Gag translation and nascent Gag molecules favor the packaging of RNA within the same locality, then Gag would be expected to demonstrate a preference for packaging viral RNA exported via the same pathway as that of the gag mRNA (Fig. 5A). Conversely, Gag could demonstrate no preference for selecting and packaging viral RNA exported by different pathways, irrespective of the transport pathway used by its own mRNA.
To explore the impact of the RNA export pathway used by gag mRNA, we generated a cell line infected with two viruses, H0-CTE-Spe and T6-RRE-Spe (Fig. 1). Each of these viruses used a different RNA transport pathway and contained an inactivating frameshift mutation in gag. We then provided Gag in trans by transfection of the cell line with various Gag-Pol expression plasmids. The Gag-encoding mRNA is exported from the nucleus by the CRM-1 pathway (RRE-Gag), the NXF1 pathway (CTE-Gag), or both pathways (RRE-Gag and CTE-Gag). We then compared the relative viral titers generated from these experiments with those from the H0-CTE.T6-RRE cell line, in which both viruses expressed functional Gag and the relative viral titers were 44%:56% (HSA: Thy; 1st lane Fig. 5B). If preferential packaging occurred, we would expect to see an alteration in the relative viral titers of the HSA and Thy viruses, with the direction of the shift dependent on the export element used to provide Gag. Alternatively, efficient trans-packaging across RNA transport pathways would result in no alteration in the ratios of HSA:Thy released from the H0-CTE-Spe.T6-RRE-Spe cell line.
Our results, summarized from three independent experiments, demonstrate that the same relative levels of the two viruses (H0-CTE-Spe and T6-RRE-Spe) were detected, irrespective of the RNA export pathway used by the gag mRNA (Fig. 5B). Similar results were obtained by infecting this cell line with a virus expressing functional Gag (data not shown). Therefore, Gag packages viral genomes with similar efficiency regardless of the RNA transport pathways used by the full-length viral RNA or the Gag mRNA. This observation further extends our previous findings, demonstrating the ability of Gag to efficiently package RNA in trans [24]. Moreover, we also observed that the resultant recombination between the two Gag mutant viruses was unaffected by the RNA transport pathway used by the supplied Gag (Fig. 5C), indicating that the proportion of heterozygous virions in the viral population was not altered. These results are consistent with our previous observation that copackaged RNA selection and dimerization occur prior to encapsidation into the virion.
The diverse roles played by the full-length HIV-1 RNA necessitate strict control over each aspect of its biogenesis and fate. Initially, newly synthesized RNA must be spliced for expression of early gene products such as Rev and Tat. Once accumulated to sufficient levels, Rev directs the export of the RRE-containing transcripts, including the unspliced RNA. Once in the cytoplasm, full-length viral RNA can be used for Gag/Gag-Pol translation, or can be packaged into nascent particles as the template for DNA synthesis in the next replication cycle. At least two viral proteins, Rev and Gag, are involved in the trafficking of the full-length RNA from the nucleus to the viral particle. Additionally, interactions between the viral components and the cellular pathways play a crucial role in the process, the extent and specifics of which are not yet completely understood. Our findings in this work provide an outline of events that lead to packaging of the HIV-1 full-length RNAs. Following export from the nucleus, HIV-1 full-length RNA selects its copackaged RNA partner by forming base-pairing at the DIS. Our data demonstrated that the initial dimerization event occurs in the cytoplasm, at a point where RNA molecules exported from NXF1 and CRM-1 pathways are not fully mixed. Although HIV-1 normally uses the CRM-1 transport pathway, this is not a prerequisite for efficient copackaging of RNAs from two viruses; HIV-1 RNAs derived from two viruses both using the NXF1 pathway for RNA transport can also be efficiently copackaged. The dimeric RNAs then interact with and are packaged by the Gag polyproteins; furthermore, Gag expressed in trans can efficiently package HIV-1RNAs exported using either the CRM-1 or the NXF1 pathway. In addition to gaining an increased understanding of the HIV-1 replication cycle, our demonstration that RNAs from the NXF1 and CRM-1 pathways are partially segregated in specific cellular locations also provides insights into the nuclear trafficking pathways themselves.
Our finding that selection of HIV-1 copackaged RNAs and initial dimerization events mainly occur in the cytoplasm provide a contrast to the suggested nuclear MLV dimerization events [25]–[27]. The difference between these two viruses in the location of their RNA dimerization event provides a possible explanation for the stark difference in their respective recombination potentials. We have directly compared the recombination rates of HIV-1 and MLV using the same target sequences separated by 1 kb [13],[28]; the HIV-1 rate under these conditions is 42.4%, whereas the MLV rate is 10-fold lower at 4.7%, even though the reverse transcriptase complex of each virus has a similar capacity to switch templates during DNA synthesis [29]–[31]. As the initial dimerization event of HIV-1 occurs in the cytoplasm, the RNAs from two different HIV-1 proviruses have a great opportunity to mix fully prior to copackaging, thereby allowing efficient formation of heterozygous virions and a high recombination potential. In contrast, MLV appears to have a limited ability to copackage RNAs generated from two separate proviruses. Although frequent crossovers can be observed in MLV recombinants, only a limited number of viruses are capable of generating phenotypically different recombinants. If, as suggested, MLV RNA dimerizes within the nucleus, particularly if it occurs at the site of transcription, then the RNA from two proviruses would be unable to fully mix and heterozygous virions would not be formed efficiently, reducing the recombination potential of the viral population. Hence, the location of the initial dimerization event may directly influence the recombination potential and the ability of the virus to generate variation and evolve. It is curious that two viruses from the same virus family appear to initiate the dimerization events at rather different locations. Interestingly, HIV-1 and MLV may also differ in whether separate pools of RNAs are used for translation and RNA packaging. Elegant early studies have suggested that MLV uses two different pools of full-length RNAs, one for translation of Gag/Gag-Pol and the other for encapsidation into newly synthesized particles [32]. Such separation of RNA pools is not apparent for HIV-1 [33],[34]. We speculate that these differences may reflect distinct strategies used by MLV and HIV-1 to regulate the function of the full-length RNA. Experimental evidence supports that both HIV-1 and MLV Gag package RNA dimers. Therefore, both viruses have to face the dilemma of incompatibility between dimer formation and ribosomal scanning through the dimeric region during translation. We speculate that MLV solves this issue by using early nuclear dimerization events to separate the two RNA pools such that the packaged RNA is dimeric and the translated RNA is monomeric. In contrast, HIV-1 adopts a different strategy, having one pool of full-length RNA capable of translation, dimerization and packaging [33],[34]. Although the exact determinants of the fate of each RNA are unclear, it is probable that dimerization within the cytoplasm plays a key role in distinguishing RNA used for translation from RNA packaged into virions. Both of these proposed strategies bypass the conundrum of translating through dimerized regions and maintain the dual role of the full-length genomic RNA.
In order to transport the full-length or singly spliced RNAs from the nucleus to the cytoplasm, HIV-1 uses the cellular protein-shuttling CRM-1 pathway instead of the NXF1-dependent pathway. To determine the amount of overlap between these two pathways and investigate the reliance of HIV-1 on its particular pathway to assort RNAs transcribed from different proviruses, we generated an RRE-independent HIV-1 that uses the MPMV CTE and the NXF1 pathway for RNA export from the nucleus. Using these viruses, we showed that the ability of RNA from different HIV-1 proviruses to fully assort does not depend on using the CRM-1 pathway. Furthermore, we showed that the two pathways (CRM-1 and NXF1) are distinct at the point of HIV-1 dimerization, although some recombination was detected, indicating a partial overlap. To our knowledge, this is the first time that an interaction between RNAs in these two pathways has been addressed. The partial overlap between the CRM-1 and NXF1 pathways also raises the interesting possibility of RNA-RNA interactions between two differentially regulated viruses, such as simian immunodeficiency virus and MPMV, during the nuclear export of their full-length RNAs within dual-infected cells.
Although the results presented here, which were generated from three different cell lines, clearly and consistently indicate that the recombination rates between an RRE- and a CTE-dependent virus are lower than those from two viruses using the same RNA transport pathway, it is worth noting that this lower recombination rate was not consistently observed when viruses were produced by cotransfection of an RRE- and a CTE-dependent viral DNA. Viral titers generated by our transfection experiments are routinely much higher than those from cell lines containing integrated proviruses, most likely because high expression from multiple copies of transfected DNA. We hypothesize that overexpression in the transiently transfected cells can overload the system and blur the boundaries between the two transport pathways. We believe the results from the cell lines containing integrated proviruses most closely mimic the events during HIV-1 replication in the human population and are more biologically relevant than any transient transfection data.
Using an HIV-1-based minimal vector and a helper construct, it was recently shown that in the absence of the Rev-RRE interaction, the level of vector RNA in the cytoplasm was slightly reduced, but the virion RNA detected in the supernatant was severely affected [35]. Therefore, the RRE-Rev interaction is important for efficient RNA packaging. One possible explanation is that cytoplasmic viral RNA was mistargeted in the absence of functional RRE-Rev interaction and was unable to be efficiently packaged by Gag. In our single virion analyses, the viral particles derived from CTE-using HIV-1 contained viral RNA at similar levels to those derived from RRE-using viruses (Table 1). Additionally, we show that RRE- or CTE-containing RNA can be packaged regardless of the RNA export pathway used by the Gag-expressing RNA (Fig. 5). Together, these findings further support the view that the RRE- or CTE-mediated RNA transport pathway not only exports the RNA from the nucleus but also targets these RNAs to proper compartments where downstream events necessary for viral replication can occur.
Why some retroviruses, such as HIV-1, have evolved a dependence on the CRM-1 pathway whereas other retroviruses, such as MPMV, use the NXF1 pathway remains a mystery. However, the use of two differentially exported HIV-1 viruses within the same coinfected cell line provides a unique tool to analyze this and many other questions. Moreover, the two pathways could be further dissected to define novel cellular factors involved in RNA trafficking, which may lead to the identification of novel targets for the development of antiviral compounds specific for HIV-1 RNA export. This cell line might also provide a basis for a high-throughput screen to identify antiviral compounds that specifically target the interplay of HIV-1 RNA with the CRM-1 pathway.
For clarity, previously described pON-H0 and pON-T6 [13] are referred to as H0-RRE and T6-RRE, respectively. T6-NC* and H0-PTAP– contain a CCHC/CCHC-to-CCHH/CCCC mutation in NC and a PTAP-to-LIRL mutation in p6, respectively. H0-ΔRRE was derived from H0-RRE by deleting the entire RRE (nt 7644–8330 in NL4-3). H0-CTE was generated by inserting a DNA fragment containing four copies of the MPMV CTE (4XCTE) obtained from pGPV-CTEx4 [8] into H0-ΔRRE at the location of the previously removed RRE. B6-RRE was derived from T6-RRE by replacing thy with mouse B7 gene; B6-CTE was generated by replacing the NotI-to-XhoI fragment of H0-CTE with that of B6-RRE. T6-RRE-Spe and H0-CTE-Spe each contains a premature stop in gag located in the middle of the CA domain. RNA derived from H0-Spe, which has identical structure as H0-RRE except containing the aforementioned premature gag stop codon mutation were packaged at a similar efficiency to RNA derived from H0-RRE [24].
Plasmid MS2-YFP was a gift from Robert Singer [36]; Bgl-mCherry has been recently described [21]. For clarity, the previously described GagCeFP-MS2SL and GagCeFP-BglSL [21] are referred to as GagCeFP-MS2SL-RRE and GagCeFP-BglSL-RRE, respectively. These modified NL43-based HIV-1 genomes encode functional gag, tat and rev, and cis-acting elements such as LTRs, 5′ untranslated regions, and RRE; however, they have inactivating deletions in pol, vif, vpr, vpu, and env. To generate GagCeFP-MS2SL-CTE and GagCeFP-BglSL-CTE, a CTE-containing DNA fragment from B6-CTE was used to replace the corresponding RRE-containing DNA fragment in GagCeFP-MS2SL-RRE and GagCeFP-BglSL-RRE, respectively, using a two-step cloning process. The general structures of newly generated vectors were characterized by restriction enzyme mapping; inserted fragments that were amplified by PCR were verified by DNA sequencing.
The cell line 293T.CC, a kind gift from Dr. Robert Doms, is a derivative of the 293T cell line that expresses CD4 and CCR5. Hut/CCR5 and 293T cells were maintained as described previously [15]. Transfection of 293T cells was performed with the TransIT-LT1 reagent (Mirrus).
Cell lines containing two proviruses were produced as previously described [13] by sequential infection of viruses at low MOIs, followed by multiple rounds of cell sorting. To measure virus titers and recombination rates, viruses were harvested from producer cell lines transfected with the HIV-1 Env-expressing plasmid pIIINL(AD8) [37] and used to infect Hut/CCR5 target cells. Marker gene expression of target cells was detected by flow cytometry using cell staining with antibodies [13]. Infected cells were identified by expression of encoded markers (HSA, Thy, or B7), whereas cells infected with recombinant viruses were identified by the GFP+ phenotype using flow cytometry. The numbers of infected cells or cells infected with recombinants in the population were then converted to MOI using the Poison distribution. Recombination rates were calculated as the MOI of GFP+ cells divided by the MOI of infected cells. Additionally, infections were carried out at relatively low MOI when each parental viral titer was less than 0.5 and a minimum of 750 GFP+ cells were sampled in each measured rate.
To generate the cell lines used for the fusion assay, pseuodotyped viruses containing T6-NC* or H0-PTAP– genomes were produced by cotransfection with helper plasmids expressing functional Gag. These viruses were used to infect 293T or 293T.CC cells separately; provirus-containing cells, identified by their expression of HSA or Thy marker, were enriched by two to three rounds of cell sorting. To perform the fusion assay, the provirus-containing 293T-based cell line was transfected with pIIINL(AD8). The provirus-containing 293T.CC-based cell line were cocultured with equal numbers of successfully transfected 293T cells at 24 h posttransfection. Viruses were harvested from the coculture 16 h after mixing the two cell lines, clarified through a 0.45-µm-pore-size filter, and used to infect Hut/CCR5 cells by spinnoculation (1200×g for 1 h at 15°C) in the presence of 10 µg/ml polybrene. Where indicated, the mixed cells were cultured in the presence of roscovitine, added at the time of mixing, at a final concentration of 10 µM.
Human 293T cells were maintained as described above; transient transfection was performed with FuGENE HD (Roche) according to the manufacturer's recommendation. For all visualization experiments, 293T cells were cotransfected with a pair of modified HIV-1 plasmid DNA, one expressing GagCeFP and the other wild-type Gag at equal weight ratios; although only the plasmid with GagCeFP was mentioned for concise description. Virus particles were harvested 17 h posttransfection, and clarified by filtration as described above. To visualize particles, polybrene was added to the culture supernatant (25 µg/ml final concentration); this mixture was placed in a glass-bottom dish and incubated for 2 h at 37°C with 5% CO2 before image acquisition.
Epifluorescence microscopy was performed with an inverted Nikon TE2000E2 microscope and a 100×1.40 numerical aperture oil objective, using an X-Cite 120 system (EXFO Photonic Solution Inc.) for illumination. Digital images were acquired using a Hammamatsu ORCA-ER camera and Openlab software (Improvision) with the excitation and emission filter sets 427/10 nm and 472/30 nm for CeFP, 504/12 nm and 542/27 nm for YFP, and 577/25 nm and 632/60 nm for mCherry, respectively. Virus particle and RNA signal analysis and colocalization were performed by using custom software developed with Matlab™ and dipimage as previously described [21]. Merged images and pseudo-colored images were generated using ImageJ software.
|
10.1371/journal.pgen.0030144 | Linkage Disequilibrium in Wild Mice | Crosses between laboratory strains of mice provide a powerful way of detecting quantitative trait loci for complex traits related to human disease. Hundreds of these loci have been detected, but only a small number of the underlying causative genes have been identified. The main difficulty is the extensive linkage disequilibrium (LD) in intercross progeny and the slow process of fine-scale mapping by traditional methods. Recently, new approaches have been introduced, such as association studies with inbred lines and multigenerational crosses. These approaches are very useful for interval reduction, but generally do not provide single-gene resolution because of strong LD extending over one to several megabases. Here, we investigate the genetic structure of a natural population of mice in Arizona to determine its suitability for fine-scale LD mapping and association studies. There are three main findings: (1) Arizona mice have a high level of genetic variation, which includes a large fraction of the sequence variation present in classical strains of laboratory mice; (2) they show clear evidence of local inbreeding but appear to lack stable population structure across the study area; and (3) LD decays with distance at a rate similar to human populations, which is considerably more rapid than in laboratory populations of mice. Strong associations in Arizona mice are limited primarily to markers less than 100 kb apart, which provides the possibility of fine-scale association mapping at the level of one or a few genes. Although other considerations, such as sample size requirements and marker discovery, are serious issues in the implementation of association studies, the genetic variation and LD results indicate that wild mice could provide a useful tool for identifying genes that cause variation in complex traits.
| Linkage disequilibrium (LD) refers to the nonrandom association of variants at different sites in the genome. In recent years, LD has been of great interest in biomedical research because of its utility in “association studies,” where DNA sequence variants associated with disease traits are used to identify susceptibility genes. The resolution of this gene-finding tool depends on how the LD decays with distance between the associated sites. The pattern of LD decay is well known in human populations, where it provides high resolution on the order of one or a few genes. This paper shows that the pattern of LD in wild house mice (in contrast to laboratory mice) is very similar to that in human populations. This result means that wild mice (reared in the laboratory) could be used in association studies to identify genes that cause trait variation. Wild mouse association studies might complement those in humans by dealing with traits that are difficult to measure in humans (such as response to carcinogen exposure) and by filtering human associations for subsequent validation with genetically engineered mouse models.
| The house mouse, Mus musculus, consists of three principal subspecies, with native populations of M. m. musculus in Eastern Europe and Asia, M. m. castaneus in Southeast Asia and India, and M. m. domesticus in Western Europe and the Middle East [1]. Populations in the Americas, Africa, and Australia are mainly of domesticus origin, due to recent transportation by Western European seafarers [1], although an admixture of domesticus and castaneus has been found in one California population [2], and other such cases may occur elsewhere. M. musculus developed a commensal relationship with humans at the dawn of agriculture, and natural populations (“wild” mice) of all three subspecies now live primarily in close association with human dwellings [3]. Laboratory strains were developed from domesticated pets and appear to be an admixture of all three subspecies [4,5].
Crosses between inbred strains of laboratory mice have been used very successfully to identify quantitative trait loci (QTL) that affect a variety of complex traits, including many related to human diseases such as atherosclerosis, diabetes, and obesity. Flint et al. [6] note that more than 2,000 mouse QTL have been detected, whereas only about 20 of the causal genes have been identified. The basic problem in gene identification is that most QTL have been discovered in F2 mapping populations, in which the confidence region for a QTL generally spans 20–40 cM (containing hundreds of genes), and efficient methods for fine-mapping have not been available. The classical approach to fine-mapping is construction of congenic lines, which can take several years to achieve a resolution of 1 cM (containing about 15 genes) [6].
In recent years, new genetic and genomic approaches have been used to reduce the time and improve the resolution of QTL mapping in mice (Table 1). These approaches include haplotype analysis of parental lines to exclude regions of identity-by-descent [7,8], association studies with a set of inbred lines [9–11], and admixture mapping in laboratory stocks with multiple inbred parents and multiple generations of recombination [12]. The expected resolution of several methods is an interval containing roughly ten to 20 genes, although the actual resolution varies with other factors such as gene density. The use of transcriptional profiling and other functional annotation sometimes can reduce the number of likely candidates to just one or a few genes [12–15]. However, useful annotations are not always available, and could be misleading. Hence, there is a need for an efficient genetic method that achieves even finer mapping resolution than existing methods.
In humans, association studies are being used extensively to identify genes that cause variation in complex traits [16–18]. This method relies on either genotyping the causative polymorphism directly, or on genotyping “tag” markers that are in strong linkage disequilibrium (LD) with the causative site. In human populations, strong LD occurs over a distance that varies depending on the genomic region, but is usually on the order of tens of kilobases [19–21]. Therefore, a significant marker–trait association in humans usually indicates that a causative polymorphism for the trait is located within about 100 kb of the marker, providing a resolution on the order of one or a few genes for LD mapping. This pattern of LD has been shaped by recombination, population structure, and demographic history. Although the house mouse has less recombination than humans [22], the commensal relationship between the two species has resulted in parallels in demographic history that might have led to similarities in the LD pattern.
Here, we investigate the genetic structure of a natural population of mice in Arizona to determine its suitability for fine-scale LD mapping and association studies. There are three main findings: (1) Arizona mice have a high level of genetic variation, which includes a large fraction of the sequence variation present in classical strains of laboratory mice; (2) they show clear evidence of local inbreeding but appear to lack stable population structure across the study area; and (3) LD decays with distance at a rate similar to human populations, which is considerably more rapid than in laboratory populations of mice. These results indicate that wild mice could provide a useful tool for identifying genes that cause variation in complex traits.
In this study, wild mice (M. musculus domesticus) were sampled from a natural population in the vicinity of Tucson, Arizona. A total of 94 mice were collected, each from a different site to avoid sampling close relatives, and we also collected small samples of the three M. musculus subspecies from their native ranges. To assess population structure and long-range LD, these mice were genotyped for a genome-wide set of 4,581 single nucleotide polymorphisms (SNPs) with a median distance between adjacent markers of ∼500 kb. These SNPs were ascertained in laboratory strains. In the Arizona mice, 89% of the SNPs are polymorphic, and 67% are “common” (i.e., minor allele frequency [MAF] > 0.05).
To assess short-range LD, we resequenced segments in four genomic regions, each on a different autosome, and each focused on a candidate gene from previous QTL studies of metabolic traits in laboratory mice (Alox15 [23], Apoa2 [13], C3ar1 [24], and Nr1h3 [25]). A total of 25.7 kb of sequence per mouse was obtained in segments of ∼1–2 kb, with one segment in each gene of interest and the others at distances of 25, 100, and 200 kb away from the gene in a single direction. The observed nucleotide diversity (π, the average proportion of nucleotide differences between pairs of sequences) is 0.10% in Alox15, 0.37% in Apoa2, 0.45% in C3ar1, and 0.09% in Nr1h3 (excluding coding sequence). Although the sample of loci sequenced in the Arizona mice is small, the mean π (0.25%) is very similar to the mean of 0.21% (SD = 0.20) for 21 autosomal loci in Western European populations of domesticus (the source of North American populations) [26,27]. This result indicates a high level of genetic variation in Arizona mice. For reference, human populations of Asian and European descent have an estimated π of 0.08% to 0.10% for autosomal loci [28,29].
The wild Arizona mice are polymorphic for a large fraction of SNPs found in classical inbred lines of laboratory mice. A total of 285 SNPs were discovered by resequencing the four regions in Arizona mice, and 163 of these are common (MAF > 0.05). In the same segments, public databases contain 63 SNPs that are polymorphic across classical inbred lines. The public databases do not contain a complete inventory of all SNPs that occur in laboratory strains, so these SNPs should be regarded as a sample. Within the sample of 63 classical line SNPs, 51% (32) occur with MAF > 0.05 in Arizona mice. Similarly, 67% of the 4,581 SNPs in the genome-wide panel (ascertained in laboratory strains) are common in the Arizona population. These results suggest that roughly 50% of the QTL found in classical line crosses will be segregating in the Arizona population, assuming that most QTL are due to SNPs (rather than structural variation [30]) and that causal SNPs are distributed in the same way as other SNPs.
In a related study, T. Salcedo and M. Nachman (unpublished data) examined haplotypes of five X-linked loci in European domesticus and musculus subspecies, as well as eight classical inbred strains. They found that laboratory strain haplotypes often are common in the wild mouse populations, consistent with earlier studies comparing wild-derived and classical inbred lines [4]. These haplotype studies support the notion that many of the lab mouse QTL are common in wild mouse populations. In addition, because laboratory mice contain a limited sample of the genetic diversity that occurs in nature [5], wild mice also provide an opportunity for new QTL discovery.
The Arizona mice show significant deviations from the Hardy-Weinberg (HW) equilibrium, with an excess of homozygotes at 45% of autosomal loci (α = 0.05) in the genome-wide panel. The inbreeding coefficient estimated from HW deviations is 0.20 averaged across loci. A similar study by Ihle et al. [31] of 204 microsatellite loci in domesticus populations in France, Germany, and Africa also revealed excess homozygosity. These results suggest the presence of population structure and/or inbreeding in wild mouse populations, as previously indicated by allozyme and other field studies of domesticus populations [32,33].
The relationships among wild mice were investigated by clustering genetic distances (one minus the mean fraction of alleles shared per locus). Figure 1 shows a neighbor-joining tree for samples from Arizona and for the three subspecies from their native ranges. The star-shaped pattern for Arizona mice indicates a lack of geographic differentiation within the study area and suggests that most pairs are unrelated. Very similar trees were obtained by Ihle et al. [31] for samples from French, German, and African populations of domesticus. There is no overall correlation between genetic distance and collection site distance in Arizona (Figure S1), although several pairs of individuals with low genetic distances also have low collection site distances. This result indicates some local inbreeding, but argues against a gradient of differentiation across the study area.
We also looked for cryptic structure in the Arizona population using the model-based clustering method of Pritchard et al. as implemented in the Structure 2.2 program [34]. When all the individuals in Figure 1 are used in the analysis, there is a large increase in the likelihood going from a model of one to a model of two subpopulations, but very small changes going from two to three or from three to four subpopulations (Figure S2A). The two-subpopulation model separates the musculus, castaneus pair from the Arizona, western European domesticus pair, such that each individual is clearly assigned to one or the other group. The three- and four-subpopulation models also maintain this division. Analysis of the Arizona population alone gives only small changes in the likelihood with each increase in the number of subpopulations modeled (Figure S2B). Therefore, the model-based structure analysis fails to support differentiation of the Arizona population into a small number of subpopulations.
We also estimated the inbreeding coefficient of each individual and the kinship coefficient of each pair of individuals using a model that allows for inbreeding, but no population structure. The mean of the inbreeding coefficients is 0.21 (consistent with the observed deviations from the HW equilibrium) and 90 of the 94 estimates are less than 0.5 (Figure S3). Four mice have inbreeding coefficients greater than 0.50, which is equivalent to three generations of full-sibling mating. The mean kinship coefficient is 0.0055, and 91% of the 4,371 pairs have an estimate less than 0.0156, which is the expected value for second cousins in a population with no inbreeding (Figure S4). Therefore, most pairs of mice are essentially unrelated, as indicated by the neighbor-joining tree in Figure 1, but a small number are rather highly inbred, and several pairs are closely related (see Figures S3 and S4).
Early studies of allozyme variation in North American domesticus showed genetic differentiation on a fine spatial scale, such as different buildings on the same farm, suggesting highly structured populations and low dispersal rates [32,35]. However, recent ecological studies have shown that wild mouse demes are very transient in nature, and there is considerable long-distance migration [36]. The results presented here are consistent with the ecological studies in suggesting some local inbreeding (which may result in transient differentiation on a fine scale), but also considerable gene flow across distances on the order of tens of kilometers so that stable population structure within our study area appears to be absent.
The decay of LD for autosomal SNPs from the genome-wide set is summarized in Figure 2 (and Figure S5), with comparisons to previous studies of human and laboratory mouse populations. The figures show that LD decays with physical distance in Arizona mice on a scale similar to samples from human populations of Asian and European descent [19], although somewhat less rapidly. In the human samples, there is essentially no significant LD between markers >1 Mb apart, but in Arizona mice this distance is >2 Mb. However, in Arizona mice, when markers are more than 200 kb apart, r2 is nearly always less than 0.3, which provides very low power for detecting associations unless the effect is very large [37]. The LD for X-linked SNPs may be slightly higher than for autosomal SNPs. In the range of 0–2 Mb, the mean of r2 for X-linked pairs is 0.072, and that for autosomal pairs is 0.021, but the sample size of X-linked pairs is small (82 X-linked versus 7,669 autosomal). Permutation testing shows essentially no significant LD between pairs of markers on different chromosomes in Arizona mice (1.3% are significant at the nominal 1% level). This result is consistent with the apparent lack of population structure (despite the excess homozygosity).
Figure 2 also shows LD in two types of laboratory mice that are being used to refine QTL location. One type is a set of 54 inbred lines used for association studies (also known as “in silico” mapping [9]), and the other is an outbred “heterogeneous stock” (HS) founded from an eight-way cross of inbred lines and used for admixture mapping [12]. In contrast to natural populations of mice and humans, these laboratory mice have strong LD for distances up to several megabases, and r2 values of 1 are sometimes observed for very distant (even unlinked) markers, as noted previously [12,14,38]. The smaller sample size of inbred lines does not account for the much higher levels of LD, since comparisons with wild mice and humans of equal sample size show similar differences (Figure S5). These results indicate that wild mice have the potential to deliver much finer mapping resolution than laboratory populations.
The genome-wide set of SNPs assayed in Arizona mice has relatively few pairs at short distances (96 pairs less than 100 kb apart). To assess short-range LD in more detail, we used resequencing data from four genomic regions. Figure 3 shows the pattern of LD decline over 200 kb in 77 Arizona mice for 163 common SNPs. The decrease in r2 is similar in all four regions, although somewhat less sharp for C3ar1. The 95th percentile of r2 falls to less than 0.4 at 100 kb. Since the mouse genome has, on average, about one gene per 100 kb, these results indicate that the pattern of LD in the Arizona population can provide a mapping resolution on the order of one or a few genes.
Figure 3 also compares LD in 60 Arizona mice with 60 unrelated European humans [19]. The human data include 3,891 SNPs selected from ten resequenced regions in order to match the allelic frequency distribution in the Arizona mice. These regions are from the Encyclopedia of DNA Elements (ENCODE) project. The mean r2 for SNPs within a 0–2 kb distance is somewhat lower in the Arizona mice (0.38) than in European humans (0.47), but the ranges overlap (0.29 to 0.43 for the four mouse regions and 0.30 to 0.61 for the ten human regions). The pattern of LD decline in these samples is very similar for the two species. This result appears to be somewhat different than the slower decline of LD in mice over longer distances (Figure 2), which may be due to the small sample of genomic regions resequenced in the Arizona mice, since LD is known to vary considerably among regions in humans [19]. In any case, the differences are fairly small at both long- and short-range distances, suggesting that the resolution of association mapping in Arizona mice would be similar to that in human populations of European and Asian descent.
The expected level of LD in a sample at equilibrium under the neutral model depends on the rate of recombination (c), the effective population size (Ne), and the diploid sample size (n): E (r2) = (1 / (1 + 4Nec)) + (1 / n) [39]. The human genome has an average of about 1.20 cM/Mb (based on a genetic map from families of European descent [40], whereas the mouse has about 0.62 cM/Mb (based on a genetic map from outbred laboratory mouse families [22]). This difference in recombination rate may account, at least in part, for the apparently slower rate of decline in LD over long physical distances in the Arizona mice compared with Asian humans (Figure 2). Figure 4 shows the relationship between LD and genetic distance, in which the latter was calculated from the genome-wide average estimates of cM/Mb. The decline of LD with genetic distance is very similar, except that when the genetic distance is less than 0.05 cM, the mean r2 of Arizona mice is less than in Asian humans. Although the sample size for mice in the genome-wide SNP set is small (69 pairs between 0 to 0.05 cM, compared with 230 pairs in Asian humans), the resequencing data in Figure 3 also suggest less LD at this distance. The pattern of LD decline over a short physical distance is very similar for the two species, but the local recombination rates are, on the average, less in the mouse than in humans (0.77 cM/Mb and 0.99 cM/Mb, respectively, in a ∼10 cM window around each region). Nevertheless, because LD varies considerably among regions, firm conclusions about a possible difference in short-range versus long-range LD will require data from additional genomic regions in wild mice.
Although differences in recombination rate appear to account for much of the difference in LD decline in Arizona mice and non-African humans, expected LD is also dependent on population size. Figure 4 shows the expected decline in LD with genetic distance for Ne values of 1,000, 3,000, and 10,000. The expectation for Ne = 3,000 fits the human data very well, but is not consistent with Ne estimates based on nucleotide diversity, which are 8,000–10,000 [29,41]. This observation is well known—i.e., that LD levels in humans of European and Asian descent suggest a smaller effective population size than that indicated by polymorphism levels [41]. The discrepancy appears to be due to departures from demographic equilibrium, and both types of data are compatible with a range of simple bottleneck models for non-African humans [28]. The discrepancy for Arizona mice is even larger, since polymorphism levels are higher (∼0.2% versus ∼0.1%), implying a higher Ne. Wild populations of domesticus, like humans, have experienced large range expansion in the past few thousand generations (humans out of Africa and domesticus out of the Middle East with agricultural humans [3,42]) and possibly also bottlenecks associated with colonization. Therefore, further work on the comparative population genetics of wild mice and humans may contribute significantly to our understanding of sequence variation and evolution in both species.
The results presented here show that the Arizona population of wild mice has a genetic structure that can complement and extend existing methods for identifying quantitative trait genes. It has a high level of genetic variation that captures a large fraction of the polymorphisms (and presumably also the QTL) in laboratory mice. It also has a favorable pattern of LD in that strong associations are limited primarily to markers less than 100 kb apart, which provides the possibility of fine-scale association mapping at the level of one or a few genes.
Favorable patterns of genetic variation and LD are two basic requirements for useful association studies, but there are also important practical considerations such as obtaining an adequate sample size of interesting phenotypes and a sufficiently high density of SNPs to take advantage of the fine-scale LD. These are serious issues, but potentially tractable. Wild mice can be bred easily in the laboratory and phenotyped under controlled conditions, which should reduce the sample size requirements relative to human association studies. Family-based designs could be used to reduce the need to collect large numbers of mice in the wild. Furthermore, transgenic and knockout mouse technologies provide an immediate functional test within the same species for genes with putative associations, thus reducing the need for large replicate cohorts. In the short term, SNPs for association studies could be discovered in candidate genes by standard methods of resequencing. In the long term, new sequencing technologies may make this process much more efficient and less costly.
In recent years, there has been a great deal of investment in whole-genome association studies in humans, and these efforts are coming to fruition [17,18,43,44]. There are two areas in which mouse association studies could complement those in humans. (1) Recent reports of whole-genome association in humans show a small number of hits that are highly reproducible and a much larger number that are promising, but require validation [18,43]. One possibility for validation is transgenic testing in mice, but this process is time consuming and may not be appropriate when homologous genes play different physiological roles in the two species. However, a SNP–trait association for a given gene in both humans and mice would suggest a true positive and also indicate that the trait is sensitive to variation of that gene in both species. Thus, candidate association studies in mice using human whole-genome association hits could provide an indication of whether transgenic testing in mice is likely to provide useful results. (2) Association studies in wild mice could be used for traits that cannot be measured easily in humans, such as response to carcinogen exposure, adverse effects of drugs at high dosage, susceptibility to disease agents, or gene expression in multiple tissues. Furthermore, the ability to control diet and other environmental exposures allows unbiased detection of genotype–environment interactions. Therefore, despite the growing success of human association studies, wild mouse studies have potential for contributing to our understanding of the genetic basis of complex traits of medical importance.
All aspects of the study were approved by the Institutional Animal Care and Use Committee of the University of Arizona and were performed in accordance with institutional policy and National Institutes of Health guidelines governing the humane treatment of vertebrate animals. Arizona mice were collected with Sherman traps at 94 different sites (mostly barns and houses) covering an area of 93 × 60 km in and around Tucson. These sites are spaced at least 100 m apart (except for one pair at 60 m), with a median intersite distance of 13.1 km, and only one mouse per site was genotyped. A small sample (seven to ten animals each) of each of the three M. musculus subspecies were trapped from more widely dispersed sites in their native ranges: ten M. m. domesticus in Western Europe (Italy, Greece, and Spain), seven M. m. musculus in Eastern Europe (Slovakia, Poland, and Hungary), and nine M. m. castaneus in India (Katrain, Dehardun, Mandi, and Siliguri). Animals were killed and DNA was extracted from liver using PureGene (Gentra Systems, http://www.qiagen.com). Dataset S1 provides sample annotation.
A genome-wide set of SNPs was genotyped for the 94 Arizona mice and seven to ten each of the subspecies from their native ranges. The genotyping was performed by Affymetrix using their GeneChip Mouse Mapping 5K SNP Kit (http://www.affymetrix.com/support/technical/datasheets/mouse_5k_datasheet.pdf). This chip includes 5,071 assays for SNPs ascertained in laboratory strains of mice, of which 4,581 were successful with the wild mouse samples. The dataset for successful assays is 98% complete over all sites and individuals, and is provided in Datasets S2 and S3. For LD estimates in Figure 2 (94 Arizona mice), we selected 2,974 autosomal SNPs with MAF > 0.05, <10% missing values, and unique genomic location on mouse genome build 36. These markers are well distributed across the genome, with median, mean, and standard deviation of distance between adjacent markers (or marker and chromosome terminus) of 536, 857, and 1,158 kb, respectively. For LD estimates in Figure S5, we selected a subset of 54 Arizona mice at random and 2,859 SNPs with MAF > 0.05 and six or fewer missing values in those individuals.
Sequence was obtained from 77 Arizona mice in each of four genomic regions (a subset of animals genotyped using the Affymetrix chip). Each region consists of four segments of 1–2 kb in length, with one segment in a gene of interest and the others at distances of 25, 100, and 200 kb away from the gene in one direction. Partially overlapping polymerase chain reaction products (two to four per segment) were sequenced using standard Big Dye terminator chemistry and capillary electrophoresis on ABI3730 (Applied Biosystems, http://www.appliedbiosystems.com). Sequence traces were base-called using Phred, assembled into contigs onto the mouse reference sequence, and then scanned for SNPs with Polyphred, version 5.01 [45]. Assembled traces and variant sites were visually inspected using Consed [46] to ensure the accuracy of the alignments and variant calls. This dataset is 90% complete over all sites and individuals and is provided in Datasets S4 and S5. For LD comparison with ENCODE data from 60 humans of European descent, we selected a subset of 60 mice and 141 mouse SNPs with MAF > 0.05 and six or fewer missing genotypes for each SNP in those individuals.
Two public databases, dbSNP mouse build 126 (http://www.ncbi.nlm.nih.gov/projects/SNP) and Perlegen mouse release 3 (http://mouse.perlegen.com/mouse), were searched for mouse SNPs in the genomic regions that were resequenced in the Arizona mice. Overlap between the public and Arizona mouse SNPs was determined by flanking sequence alignment using Cross_Match (http://www.phrap.org). Genotypes were downloaded, and SNPs that occur in classical inbred lines were identified as sites with at least two different genotypes among lines that are not wild derived (i.e., not in the list of wild derived lines provided on the Jackson Laboratory Web site; http://jaxmice.jax.org/list/cat481389.html).
Genotypic data were analyzed with R statistical software [47]. Exact HW tests were performed with the function “HWE.exact” in the “genetics” package [48]. The within-population inbreeding coefficient, which is the correlation between alleles at one locus within an individual [49], was calculated using the “diseq” function. LD was estimated as the composite (genotypic) r2, which is the squared correlation of genotypic indicators at two loci in a diploid individual, whereas the usual gametic r2 is the squared correlation of allelic indicators at two loci in a haploid gamete [49]. Since a large fraction of loci show significant deviation from the HW expectation in the Arizona mice, we prefer the genotypic rather than the usual gametic correlation, because its calculation does not require an assumption of random mating. However, to evaluate the potential difference, we used a maximum likelihood method (implemented in the “LD” function of the R genetics package) to estimate the gametic correlation. Although random mating is assumed for this estimation, deviations in the direction of excess homozygosity (as observed in the mice) are not likely to bias the estimate [50]. Figure S6 shows that there is very little difference between the squared genotypic and gametic correlation estimates in either Arizona mouse or human populations. Nevertheless, we present the assumption-free genotypic measure, which may be more relevant in the context of association studies [49]. Permutations (n = 1,000) were used to obtain genome-wide significance thresholds for composite r2.
Genotypes from the genome-wide set of SNPs were used to construct a genetic distance matrix as the mean fraction over loci of the number of shared alleles between each pair of individuals. The loci consist of all SNPs (4,158) that have at least some nonmissing values within each of the four groups of wild-caught mice. This matrix was used to construct a neighbor-joining tree [51] with the “nj” function of the R statistics package “ape” [52].
Structure 2.2 (http://pritch.bsd.uchicago.edu/structure.html) was used to detect cryptic population structure using the model-based approach of Pritchard et al. [34]. This method assumes LD within subpopulations, so the full set of autosomal markers were thinned to a set of 752 in which no two markers were closer than 2 Mb apart (the distance at which there is very little LD). Two sets of individuals were analyzed: (1) the full set of 120 mice shown in Figure 1 and (2) just the 94 Arizona mice. For each set of mice, the admixture model having one, two, three, or four subpopulations was analyzed. For each model and set of mice, three to seven independent runs of the program were made, with burn-in and subsequent steps each numbering 25,000 (or, in some cases, 50,000).
Degrees of inbreeding and relatedness in the mice in our study were estimated by maximum likelihood using a model that allows for inbreeding but no population structure, as described in Milligan [53], Hepler [54], and Weir et al. [55]. This model is based on Jacquard's nine identity coefficients, Δ = Δ1,…, Δ9. Maximum likelihood estimates of the nine coefficients were obtained for each pair of individuals in our sample using an expectation–maximization algorithm. The estimated kinship coefficient, θXY for individuals X and Y, was obtained from the estimates of Δ and the relationship
. The inbreeding coefficient for individual X (FX) was estimated as FX = 2θXX − 1, where θXX is the kinship coefficient of an individual with itself. The data used for relatedness estimation consist of 3,928 autosomal SNPs from the genome-wide set.
We performed a simulation study in which the performance of the estimators was evaluated with a set of markers similar to those used in this study. Data were simulated via gene-dropping, with 3,928 autosomal markers having allelic frequencies that matched those observed in the Arizona mice. We obtained genetic map positions for 2,087 of the 3,928 markers directly from a map of 8,513 markers [22], and the remaining positions were estimated by interpolation. Recombination frequencies for pairs of adjacent loci were obtained using Haldane's map function. In the simulations of meioses from parent to offspring, crossovers occurred along each chromosome according to a no-interference model. We simulated individuals and pairs of individuals with a variety of different levels of relatedness and inbreeding. In general, estimated kinship coefficients were within 0.063 units of the true values in 95% of all simulated pairs of relatives. Estimated inbreeding coefficients were within 0.120 units of the true values in 95% of all simulated individuals. Estimation was more accurate for certain situations: the estimated inbreeding coefficient was below 0.031 for 95% of all simulated non-inbred individuals, and the estimated kinship coefficient was below 0.015 for 95% of all simulated pairs of non-inbred unrelated individuals.
We started with the 60 classical and wild-derived inbred lines of Petkov et al. [38], which were selected for their genetic diversity and to remove closely related “sibling” strains. Six of these lines were eliminated for lack of sufficient genotypic data (LT/SvEiJ, CAST/EiJ, MOLD/RkJ, CZECH/EiJ, SKIVE/EiJ, and PWK/PhJ). Genotypes for the remaining 54 lines were obtained from the Wellcome-CTC Mouse Strain SNP Genotype Set Web site (http://www.well.ox.ac.uk/mouse/INBREDS). From a total of 12,614 SNPs that are polymorphic across the 54 lines, we selected 2,855 autosomal SNPs with six or fewer missing values to match the allelic frequency distribution (within 5% bins) in a random subset of 54 Arizona mice, while maximizing overlap with the Arizona SNPs.
These data are from the HS QTL project [12], and genotypes were downloaded from their Web site (http://gscan.well.ox.ac.uk). For the LD estimation, we selected either 94 (Figure 2) or 54 (Figure S5) animals at random from a pool of 1,940 animals with identifiers beginning with “A0” (i.e., not ancestors of other animals in the set). These 1,940 animals belong to 85 different families. The samples of 94 or 54 animals belong to 33 or 22 families, respectively. We also analyzed a sample of 85 animals, consisting of one randomly chosen individual per family, and the LD results are very similar. For example, at an intermarker distance of 2 Mb, the mean r2 = 0.31 and 0.26 and the 95th percentile of r2 is 1.00 and 0.96 for the samples of 85 and 94, respectively. From a total of 12,112 SNPs genotyped in the HS animals, we selected a subset to match the corresponding Arizona set in terms of number and allelic frequency distribution (within 5% bins) while maximizing overlap with the Arizona SNPs. All selected SNPs had <10% missing values.
To estimate long-distance LD for a set of SNPs comparable to the Arizona mouse data, we analyzed a selected set of genotypic data from the International HapMap Project [19] (release 21; http://www.hapmap.org). Data for three population samples were used: 60 unrelated individuals with European ancestry (Centre d'Etude du Polymorphisme Humain samples of Utah residents with ancestry from northern and western Europe [CEU]), 45 Japanese from Tokyo, Japan, and 45 Han Chinese from Beijing, China. The two Asian samples were combined for allelic frequency and LD estimation. A SNP set was selected separately for each human group (Asian and European) to match the Arizona mouse set in terms of number, allelic frequency distribution (within 5% bins), and chromosomal distribution (20 linkage groups, excluding human Chromosomes 20–22), but otherwise at random. The two Asian samples were combined for LD estimation by first calculating the genotypic correlations for the Japanese and Chinese samples separately, testing for homogeneity, and then averaging the two correlations using Fisher's z-transformation. The two sets of correlations are very homogenous, since 1.1% of the homogeneity tests are significant at the nominal level of 1%. For comparison to the sets of 54 mouse samples, 54 of the 60 CEU samples were chosen at random.
For comparison to LD in SNPs discovered by resequencing Arizona mice, we selected SNPs discovered by resequencing a panel of 16 CEU individuals in 10 ENCODE regions and subsequently genotyped on 90 CEU individuals [19]. For the 60 unrelated CEU individuals, we selected a set of 3,891 SNPs with MAF > 0.05, six or fewer missing values, and an allelic frequency distribution matching the 141 SNPs selected for a set of 60 Arizona mice. Genotypes were obtained from the HapMap Web site.
The GenBank (http://www.ncbi.nlm.nih.gov/Genbank) accession numbers for the regions resequenced in this study are EU007907 for Alox15, EU007908 for Apoa2, EU007909 for C3ar1, and EU007910 for Nr1h3. |
10.1371/journal.pgen.1007283 | Controlled branched-chain amino acids auxotrophy in Listeria monocytogenes allows isoleucine to serve as a host signal and virulence effector | Listeria monocytogenes (Lm) is a saprophyte and intracellular pathogen. Transition to the pathogenic state relies on sensing of host-derived metabolites, yet it remains unclear how these are recognized and how they mediate virulence gene regulation. We previously found that low availability of isoleucine signals Lm to activate the virulent state. This response is dependent on CodY, a global regulator and isoleucine sensor. Isoleucine-bound CodY represses metabolic pathways including branched-chain amino acids (BCAA) biosynthesis, however under BCAA depletion, as occurs during infection, BCAA biosynthesis is upregulated and isoleucine-unbound CodY activates virulence genes. While isoleucine was revealed as an important input signal, it was not identified how internal levels are controlled during infection. Here we show that Lm regulates BCAA biosynthesis via CodY and via a riboregulator located upstream to the BCAA biosynthesis genes, named Rli60. rli60 is transcribed when BCAA levels drop, forming a ribosome-mediated attenuator that cis-regulates the downstream genes according to BCAA supply. Notably, we found that Rli60 restricts BCAA production, essentially starving Lm, a mechanism that is directly linked to virulence, as it controls the internal isoleucine pool and thereby CodY activity. This controlled BCAA auxotrophy likely evolved to enable isoleucine to serve as a host signal and virulence effector.
| Bacterial pathogens must adapt to their host environment to carry out a successful infection. Sensing host-derived signals precedes adaptation, and triggers switching to the virulent state. Within mammalian cells L. monocytogenes responds to branched-chain amino acids (BCAA) deficiency by inducing virulence gene expression. In this study, we provide compelling evidence that fine tuning BCAA biosynthesis in L. monocytogenes allows the bacteria to sense isoleucine as a host-specific signal. Tightly controlled BCAA production depends on Rli60, a riboregulator, which is transcribed upstream to the BCAA biosynthesis genes. Rli60 functions as a ribosome mediated attenuator that cis-regulates BCAA production under limiting conditions. This study highlights the remarkable cross-regulation of metabolism and virulence in bacterial pathogens.
| Listeria monocytogenes (Lm) is a facultative intracellular bacterial pathogen and the causative agent of listeriosis disease [1]. It invades host cells via phagocytosis, or by induction of endocytosis [2]. Upon invasion, it is initially found in a membrane-bound vacuole, from which it escapes into the host cell cytosol using the pore-forming toxin listeriolysin O (encoded by the hly gene) and two phospholipases [3–5]. Once in the host cell cytosol, Lm replicates and spreads into neighboring cells using actin-based motility that is mediated by the virulence factor ActA [6,7]. The transcription of the aforementioned virulence factors (and other factors) is regulated by the master virulence activator, PrfA [8].
Lm is also a saprophyte, highly abundant in the soil and vegetation. The transition to the pathogenic state relies on sensing of host-specific signals, that together inform the bacterium of its intracellular location. To date, all signals were shown to affect PrfA, directly or indirectly [9]. The signals include temperature [10], availability of carbon sources [11–13], iron [14,15], glutathione [16,17], L-glutamine [18] and BCAA (isoleucine, leucine and valine) [19,20]. We previously found that BCAA, particularly isoleucine, are important metabolic signals for Lm in the mammalian niche. Lm senses the low availability of BCAA within the host cell cytosol and responds by triggering virulence gene expression [19]. This response is dependent on the global transcription regulator and metabolic sensor, CodY, which directly binds isoleucine and activates or represses genes [21,22]. While classically CodY was shown to gain function upon binding of isoleucine, acting as repressor of metabolic genes, we found it retains a regulatory activity also when unbound to isoleucine [23]. Under this condition, CodY repression of the metabolic genes is alleviated and the unbound CodY becomes an activator of PrfA and thereby the downstream virulence genes [19,20,23]. Notably, while these findings placed CodY at the crossroad of metabolism and virulence, they revealed isoleucine to be a key signaling molecule within the host that influences gene expression. This discovery prompted us to hypothesize that BCAA biosynthesis in Lm must be tightly regulated. BCAA biosynthesis in Lm is strongly repressed by CodY under high BCAA conditions, and is transcriptionally up-regulated when BCAA levels drop [19,23]. Notwithstanding, despite encoding all the BCAA biosynthesis genes, Lm still requires BCAA supplement to support optimal growth under nutrient limiting conditions [24,25]. Considering this observation, we speculated that Lm may have evolved additional mechanisms that finely tunes BCAA biosynthesis, enabling isoleucine to serve as a host signal and effector of virulence.
Several transcriptome studies identified a putative small RNA, named Rli60, upstream to the BCAA biosynthesis genes (Fig 1A) [23,26–28]. Rli60 was predicted to function as a riboswitch [26] or as a sRNA [27], though these predictions were not validated. Other studies suggested a role for Rli60 in biofilm formation and virulence, by a mechanism that is not known [29,30]. Here we found that Rli60 functions as a ribosome-mediated attenuator that cis- regulates BCAA biosynthesis genes. Importantly, we found this riboregulator to restrict BCAA production even under BCAA depletion. This property is important for Lm virulence, as it limits the internal pools of BCAA, thus maintaining isoleucine signaling function via CodY. This controlled BCAA-auxotrophy in Lm may thus represent an adaptive mechanism to the life within the host.
In Lm the BCAA biosynthesis genes are encoded in one operon consisting of nine genes (ilvDBHC-leuABCD-ilvA), named the ilv-leu operon. rli60 transcript was previously identified upstream to ilvD, the first gene of this operon, and was suggested to consist of 184 to 339 nt, raising the question whether it functions as a sRNA or cis-regulatory element [26–28]. Examining the rli60-ilvD genomic region, we identified a single promoter upstream to rli60 with no additional promoter upstream to ilvD (Fig 1A and S1 Fig). Rapid amplification of cDNA 5'-Ends (5'-RACE) analysis confirmed that rli60 and ilvD are co-transcribed and share a single transcription start site (TSS) (Fig 1A and S1 and S2 Figs). Importantly, the co-transcript was detected under BCAA limiting conditions, whereas a shorter transcript (~200 nt) representing only Rli60 RNA was detected under rich BCAA conditions, suggesting a BCAA-dependent transcription regulation (S2 Fig).
We employed quantitative reverse-transcription PCR (qRT-PCR) to analyze the transcription pattern of rli60 and ilvD in Lm bacteria grown under varying BCAA concentrations. Three types of media were used: brain heart infusion (BHI), a rich medium containing excess amounts of BCAA; a minimal defined medium (MM) containing 800 μM of each BCAA; and a low BCAA minimal defined medium (LBMM) containing 80 μM of each BCAA. As shown in Fig 1B, under rich BCAA conditions (i.e., in BHI) both rli60 and ilvD were repressed, whereas under low BCAA conditions (i.e., in LBMM) their transcription was up-regulated (~140-fold). Notably, in the MM medium a differential transcription pattern was observed, where rli60 exhibited a higher transcription level in comparison to ilvD, suggesting a premature termination of transcription may occurs upstream to ilvD. Further analysis of rli60-ilvD transcription in bacteria grown intracellularly in bone marrow-derived macrophage cells (BMDM) revealed a similar transcription pattern to that seen upon Lm growth in LBMM (Fig 1C), insinuating low availability of BCAA within the macrophage cytosol. Of note, ilvD up-regulation was specific to conditions were BCAA were limited, and was not observed upon limitation of other amino acids e.g., arginine, or both tryptophan and phenylalanine [19] (S3 Fig).
To further corroborate the premise that rli60 and ilvD are regulated in a BCAA-dependent manner, Northern blot analyses of Rli60 and ilvD were performed on RNA extracted from WT bacteria grown in BHI, MM and LBMM. This analysis confirmed the existence of Rli60 RNA at the size of ~200 nt, and its transcription under BCAA limiting conditions (Fig 1D). In accordance with the 5’-RACE and the qRT-PCR analyses, a longer transcript of a ~1000 nt was detected in LBMM (Fig 1D). Northern blot analysis using an ilvD specific probe suggested that this ~1000 nt transcript may represent the rli60-ilvD co-transcript. Additional longer transcripts of the ilv-leu operon were also detected, not including rli60, suggesting it may be processed (cleaved). (Fig 1D). Altogether, these findings establish that rli60 is co-transcribed with the ilv-leu genes in a BCAA-dependent manner, suggesting it may function as a cis-regulatory element. To address the question whether rli60 is also repressed by CodY under high BCAA conditions, as known for the ilv-leu genes [19], a Northern blot analysis of Rli60 was performed on RNA extracts from ΔcodY bacteria. Higher levels of Rli60 were observed in ΔcodY bacteria under high BCAA conditions compared to WT bacteria, demonstrating that CodY represses rli60 when BCAA are plentiful (Fig 1D). Accordingly, two putative CodY binding-sites were identified upstream to rli60 and ilvD genes (Fig 1A and S1 Fig).
To further characterize the role of Rli60 as a regulator of the ilv-leu operon and its relationship with CodY, we examined the transcription of ilvD in WT, ΔcodY, Δrli60 and in a ΔcodY/Δrli60 double mutant strain under the different BCAA growth conditions. Under high BCAA conditions (i.e., in BHI), ΔcodY and Δrli60 mutants transcribed ilvD to a similar level (~40-fold more than WT bacteria), whereas the double mutant (ΔcodY/Δrli60) up-regulated ilvD transcription by ~600-fold, as compared to WT bacteria (Fig 2A). In MM medium, where the BCAA levels are lower, ilvD was only slightly upregulated in ΔcodY in comparison to WT bacteria, since under this condition CodY repression is lessened (Fig 2A). Remarkably, Rli60 was found to be the main repressor of the ilv-leu genes under this condition, as evidenced by the enhanced ilvD transcription in Δrli60 and ΔcodY/Δrli60 bacteria (~10-fold in comparison to WT bacteria) (Fig 2A). Upon low BCAA conditions (i.e., in LBMM), ilvD transcription was upregulated in WT bacteria (~220-fold as compared to WT bacteria grown in BHI) (Fig 2A), since under this condition the transcription continues through rli60, transcribing the ilv-leu genes (Fig 1B and 1D). That said, Rli60 still repressed the ilv-leu genes under this condition, as ilvD transcription was even higher in Δrli60 and ΔcodY/Δrli60 bacteria (by ~3-fold), overall indicating that Rli60 prevents the full activation of this operon (Fig 2A). Taken together, these findings suggest that two BCAA-dependent mechanisms regulate the ilv-leu operon; the first is CodY, which represses both rli60 and the ilv-leu genes under high BCAA conditions, and the second is Rli60, which kicks in when BCAA levels drop, further repressing the transcription of the ilv-leu genes. Notably, an overall similar expression pattern was observed with IlvD protein, under the same growth conditions and strains, using Western blot analysis (Fig 2B), corroborating the premise that CodY and Rli60 are the primary regulators of BCAA biosynthesis.
We reasoned that Rli60 may have the ability to directly sense BCAA and to act as a cis-regulatory RNA. To investigate this hypothesis, rli60 in the context of its native regulatory region (consisting 675 bp upstream to IlvD start codon) was cloned upstream to luciferase reporter genes on a pPL2 plasmid (pPL2-rli60-luxABCDE), which was further transformed into E. coli bacteria auxotrophic for BCAA (E. coli K-12 ilvC::Km strain) (Fig 3A). Using this heterologous system, Rli60 regulation of the downstream lux genes could be examined under varying BCAA concentrations (supplemented in the media) in the absence of de novo BCAA synthesis and CodY (E. coli bacteria are devoid of CodY, as it is a Gram-positive specific regulator). We observed that luminescence increased as BCAA concentrations were lowered (Fig 3A), demonstrating that Rli60 directly senses BCAA availability and regulates its downstream genes in a concentration-dependent manner. As a control, a similar plasmid was used, this time deleted of rli60 sequence (pPL2-Δrli60-luxABCDE), which demonstrated high luminescence levels independent of BCAA concentrations (Fig 3A). These findings indicated that Rli60 directly senses BCAA levels and accordingly negatively regulates its downstream genes.
Taken together, the data suggested that Rli60 cis-regulates BCAA biosynthesis in response to BCAA availability, but the mode of regulation was not clear. To search for clues for the regulation type, we examined rli60 sequence and found a short coding sequence of 13 amino acids that is followed by putative stem-loop structures. Notably, the identified peptide was enriched in BCAA codons (Fig 3B), implying that a reduced rate of translation caused by BCAA limitation may lead to a transcription attenuation. In such a mechanism of ribosome-mediated attenuation, the translation rate of the leader peptide dictates the secondary structure of the leader transcript. When the regulatory amino acids are in short supply, translation is slow, allowing the RNA to form an anti-terminator structure that permits transcription to continue into downstream genes; however, when amino acids supply is in excess, translation is rapid, preventing the formation of the anti-termination loop and causing the RNA to assume a terminator structure [31].
To examine the existence of the leader peptide in rli60, it was fused to EGFP (a translational fusion) with its native promoter and cloned into pPL2 plasmid (pPL2-rli60-peptide-EGFP) (Fig 3B). Translation of the fused protein was analyzed in WT and ΔcodY bacteria grown in BHI, MM, and LBMM using Western blot analysis. We observed that the fused protein was indeed translated and that its expression is CodY-dependent under BHI (high BCAA conditions), similarly to Rli60 (Fig 3B). The fused protein was also detected under MM and LBMM conditions, but as expected, to a lower extent. Next, we analyzed Rli60 sequence using PASIFIC server [32] and identified two alternative RNA structures downstream to the leader peptide, consisting of overlapping hairpins that may serve as a terminator and an anti-terminator (Fig 3C). We then performed a mutational analysis of the hairpins in the Lm genome. A series of nucleotide substitution mutations were made in the putative terminator hairpin to impair its stability, that do not interfere with the anti-terminator structure (rli60-ter mutant) (Fig 3C). Additional mutations were made in the peptide’s ribosome binding site (RBS) and start codon (ATG) to hinder its translation (rli60-rbs and rli60-atg mutants, respectively) (Fig 3C). The latter mutations are known as “super-attenuators”, as in the absence of engaging ribosomes the terminator hairpin is hyper-stabilized, leading to a premature termination [33]. We next used these mutants to analyze ilvD transcription during growth in MM and LBMM, conditions in which rli60 is transcribed (Fig 1B and 1D). As predicted, under LBMM conditions (where the anti-terminator should be formed) the "super-attenuator" mutants (rli60-atg and rli60-rbs) demonstrated a significant reduction in ilvD transcription (Fig 3D), whereas under MM conditions (where the terminator should be stabilized) the rli60-ter mutant demonstrated an enhanced ilvD transcription (Fig 3E). As expected, no significant effect was observed for each mutant in the other medium (Fig 3D and 3E), suggesting Rli60 regulates the ilv-leu operon via a ribosome-mediated attenuation mechanism.
In the literature, Lm is described as a BCAA auxotroph, or a partial auxotroph, since it requires BCAA supplement for optimal growth [25]. To examine whether Rli60 is the cause for BCAA requirement in Lm, we compared the growth of WT, Δrli60 and ΔilvC bacteria in minimal medium supplemented with increasing concentrations of BCAA (0, 20, 80 and 800 μM of each) (Fig 4A and S4 Fig). Of note, ilvC encodes a central enzyme in the BCAA biosynthesis pathway [19]. We found that ΔilvC behaves like a true auxotroph, failing to grow when BCAA levels drop, whereas WT bacteria exhibit a moderate phenotype, behaving like semi-auxotrophs, and Δrli60 bacteria grow like prototrophs, less affected by external BCAA levels (Fig 4A and S4 Fig). The different phenotypes were most evident under conditions where no BCAA were added, as Δrli60 exhibited a significant growth advantage over WT bacteria, while ΔilvC did not grow (Fig 4B). Further support for the premise that indeed Rli60 restricts BCAA biosynthesis was provided by the finding that the rli60-ter mutant grew better under low BCAA conditions, like Δrli60 (i.e., better than WT bacteria), whereas the “super-attenuator” mutants grew similarly to ΔilvC, (i.e., worse than WT bacteria), in accordance with their predicted ilv-leu gene regulation (Fig 4 and S4 Fig). Taken together, these results demonstrated that Lm is capable of relying completely on de novo BCAA synthesis and grow independently of external BCAA, though this capability is restricted by Rli60.
To investigate whether BCAA semi-auxotrophy supports Lm virulence, we next analyzed the transcription of three major virulence genes, prfA, hly and actA, in WT and Δrli60 bacteria grown in LBMM (that mimicks intracellular conditions [19]). Notably, the transcription level of the virulence genes was reduced in Δrli60 in comparison to WT bacteria (Fig 5A), suggesting that over-production of BCAA hinders virulence gene expression. Of note, in a previous study that examined the impact of Rli60 on virulence gene expression, an enhanced prfA transcription was detected in a Δrli60 mutant [29]. A close examination of this rli60 deletion mutant indicated that the ilvD TSS was also deleted together with the rli60 sequence (ilvD TSS was identified here by 5'-RACE analysis, S2 Fig), therefore it is most likely that BCAA biosynthesis was impaired in this resulting mutant, which can indeed further lead to enhanced prfA transcription by CodY.
To examine whether the reduction in transcription of virulence genes in response to Rli60 deletion is mediated by CodY, we combined Δrli60 deletion with R61A mutation in CodY, which considerably reduces CodY’s affinity to isoleucine (codY-R61A/Δrli60 mutant) [20,34], and tested this double mutant for virulence gene expression. We reasoned that the mutated CodY, will be “blind” to the increase in isoleucine and therefore, virulence genes will be activated. In line with our prediction, we found the double mutant to induce virulence gene transcription similarly to WT bacteria (Fig 5A), supporting the premise that uncontrolled production of BCAA directly affects CodY regulation, hindering its ability to activate virulence gene expression under low BCAA conditions. Unlike the experiments in defined medium, examination of Δrli60 and codY-R61A/Δrli60 mutants in vivo in mice infections implied a more complex picture. Both mutants were slightly attenuated for virulence in comparison to WT bacteria, demonstrating ~40% reduction in competitive fitness, as evaluated using a competitive index assay (Fig 5B). While we previously found that ΔcodY is 10-fold less virulent in mice (whereas prfA mutant is >100-fold less virulent) [20,35], it is likely that the codY-R61A and rli60 mutations only partially alter CodY activity and thus lead to a slight effect in vivo. We previously demonstrated that CodY functions both in its isoleucine-bound and -unbound form, simultaneously activating and repressing different genes, some of which are important for virulence independently of PrfA, therefore affecting Lm gene expression in a highly complex manner [23]. Moreover, within the intracellular niche BCAA are not the sole signal for prfA activation, and multiple signals were shown to play a role, which together orchestrate virulence gene expression. This study focuses on one such signal, overall demonstrating that BCAA biosynthesis fine tunes CodY activity and thereby virulence gene transcription.
Bacteria rarely encounter rich nutrient conditions in natural environments. Bacterial pathogens that traverse freely between extracellular and intracellular environments are frequently subjected to massive changes in nutrient availability. The ability to sense nutrients, remodel metabolic pathways and change behavior via gene regulation is therefore fundamental to bacterial adaptation and growth. Furthermore, nutrient sensing provides essential information regarding the physiological condition of the environment, signaling a niche specific “signature” that informs the bacteria of their exact location (e.g., extracellular vs. intracellular). This added information is particularly critical during host invasion, as pathogens need to quickly express virulence factors to counteract host defense mechanisms in order to survive. In line with this premise, this study demonstrates that sensing of BCAA is an important feature of Lm not only to support growth but also to promote virulence, and that the ability to control BCAA production is fundamental to successful invasion. It is generally accepted that controlled metabolite production is crucial for cell functioning and growth by providing competitive advantage in natural environments. Yet, here we show that Lm has evolved a regulatory mechanism for BCAA biosynthesis that hampers growth in extracellular environments but gives an advantage within the host. Limiting de novo BCAA biosynthesis enables CodY to accurately sense the external level of isoleucine and to regulate genes in a BCAA-concentration dependent manner. In a sense, this tightly regulated BCAA auxotrophy of Lm has become a control point that shapes not only metabolic networks but also virulence gene expression and thus the ability of Lm to infect its host. We propose that this adaptive mechanism may be the result of co-evolution of Lm with its host, allowing isoleucine to be used as a host specific signal.
The finding that isoleucine deficiency is the signal for virulence gene activation, prompted us to look for mechanisms that control isoleucine biosynthesis during infection. We knew that BCAA biosynthesis in Lm is intact and functional and that the ilv-leu genes are up-regulated when BCAA levels drop [19,25]. However, while CodY was shown to regulate the ilv-leu genes under rich nutrient conditions, it was not clear if and how BCAA biosynthesis is controlled under poor nutrient conditions, e.g. within the host. In this study, we characterized Rli60 as a ribosome-mediated attenuator that controls the ilv-leu gene transcription in a BCAA-dependent manner. While many bacteria use attenuation mechanisms as ON/OFF switches to regulate amino acid biosynthesis [31,33,36], we found Rli60 to limit BCAA production such that internal levels are insufficient to support optimal growth. This BCAA auxotrophy of Lm is partial, fully dependent on Rli60, thus falling into the category of 'phenotypic auxotrophy', whereby auxotrophy is a result of gene dysregulation rather than loss of function [37]. Overall, our findings indicate that BCAA biosynthesis in Lm is regulated by two mechanisms, the first involving classical CodY repression under nutrient rich conditions and the second using Rli60-ribosome-mediated attenuation under poor BCAA conditions (Fig 6). This model relies on two types of regulations; a global (via CodY) and a specific (via Rli60), which is typical for metabolic pathways. However, since isoleucine (the end product of this pathway) is also the input signal of CodY, Rli60 has the capacity to impact CodY activity, and thereby global gene expression, strengthening the premise that BCAA production must be tightly regulated. In support of this idea, a previous study in B. subtillis has demonstrated that changes in endogenous BCAA biosynthesis indeed affect CodY global regulation [38].
Regulation of bacterial and host behaviors via amino acid auxotrophy is an emerging concept. For example, Group A Streptococcus bacteria (GAS) requires supplementation of asparagine to support growth and depends on the host supply [39]. Notably, it was shown that GAS stimulates host asparagine synthesis via secretion of hemolysin toxins that trigger endoplasmic reticulum stress. In parallel, GAS senses host derived asparagine, using a two-component system, and regulates metabolic and virulence genes, including the hemolysin toxin genes [39]. Francisella tularensis and Legionella pneumophila are additional example, as these bacteria have lost their ability to synthesize certain amino acids, but developed unique mechanisms to obtain them from the host [40]. F. tularensis, auxotrophic for BCAA, triggers the host macroautophagy degradation machinery to increase the intracellular pool of these amino acids [41]. Similarly, L. pneumophila, auxotrophic for seven amino acids (Arg, Cys, Ile, Leu, Met, Thr and Val) [42,43], triggers proteasomal degradation of polyubiquitinated proteins and activate mammalian transporters to import the required amino acids into the Legionella containing vacuole [44–46]. Although it is still not clear how Legionella and Francisella sense the availability of host nutrients and whether they use this information to regulate virulence, it is likely that such a mechanism exists. Of note, it was previously suggested that threonine signals Legionella to differentiate and replicate within macrophage cells [47]. Together these examples demonstrate how amino acid auxotrophy in bacterial pathogens can be a driving force of pathogenic evolution or an adaptive mechanism to life within the host, supporting the premise that metabolism and virulence are tightly interlinked.
Interestingly, the idea of amino acid auxotrophy as a system that regulates cellular responses exists also in mammalian cells. Humans and most mammals are auxotrophic for certain amino acids and acquire them from the microbiota and diet. It is clear now that this amino acid auxotrophy, particularly of immune cells, is involved in regulation of immune responses, production of antimicrobial effectors, T cell responses and additional mechanisms [48]. Several amino acids were shown to function as immuno-modulators such as arginine, tryptophan and glutamine. Arginine plays a role in macrophage activation and blocks tumor growth mainly via its conversion to nitric oxide (NO), which by itself is toxic to microbes and targets intracellular pathogens in addition to its signaling roles [49–51]. Tryptophan is degraded to kynurenines that were suggested to regulate T cells and glutamine was shown to be important for T cells proliferation [52–54]. In light of these findings, it could be interesting to examine how bacterial pathogens compete for these amino acids within the host taking into account their regulatory roles, potentially manipulating them to subvert host responses.
In conclusion, controlled BCAA auxotrophy in Lm likely represents an adaptive mechanism to the life within the host. This study places Rli60 at the cross-road of metabolism and virulence and validates the role of BCAA in Lm regulation of virulence. A better understanding of bacterial pathogens metabolism during infection and its links to virulence and host cell modulation is critical for our understanding of bacterial pathogenesis and for the identification of new metabolic targets that can be the basis for the development of novel drugs and therapeutic approaches.
Experimental protocols were approved by the Tel Aviv University Animal Care and Use Committee (01-15-052, 04-13-039) according to the Israel Welfare Law (1994) and the National Research Council guide (Guide for the Care and Use of Laboratory Animals 2010).
Listeria monocytogenes 10403S was used as the WT strain and as the parental strain to generate allelic exchange mutant strains (S1 Table). E. coli XL-1 Blue strain (Stratagene) was used for generation of vectors, and E. coli SM-10 strain [55] was used for plasmid conjugation to Lm. Plasmids and primers used in this study are listed in S1 and S2 Tables, respectively.
Lm bacteria were grown at 37°C with agitation in brain heart infusion (BHI) as a rich medium or in minimal defined medium (MM), and harvested at mid logarithmic growth (OD600 of ~0.3). MM was prepared as described previously [56]: phosphate buffer (48.2 mM KH2PO4 and 153 mM Na2HPO4, pH 7), 0.41 mg/ml magnesium sulfate, 10 mg/ml glucose, 100 μg/ml of each amino acid (methionine, arginine, histidine, tryptophan, phenylalanine, cysteine, isoleucine, leucine and valine), 600 μg/ml glutamine, 0.5 mg/ml biotin, 0.5 mg/ml riboflavin, 20 mg/ml ferric citrate, 1 mg/ml para-aminobenzoic acid, 5 ng/ml lipoic acid, 2.5 mg/ml adenine, 1 mg/ml thiamine, 1 mg/ml pyridoxal, 1 mg/ml calcium pantothenate and 1 mg/ml nicotinamine. For growth under limiting concentrations of branched-chain amino acids (BCAA), MM was freshly made with 10-fold less of isoleucine, leucine and valine (resulting in a final concentration of 10 μg/ml or 80 μM of each amino acid) and named low-BCAA minimal defined medium (LBMM). For growth under limiting concentrations of either arginine or both phenylalanine and tryptophan, MM was freshly made with 10-fold less of these amino acids (resulting in a final concentration of 10 μg/ml). For growth curves, bacteria from overnight MM cultures were washed 3 times with PBS to remove excess BCAA and adjusted to OD600 of 0.03 in fresh MM without BCAA or supplemented with 2.5, 10, or 100 μg/ml of BCAA (20, 80 and 800 μM, respectively). Bacterial growth was measured by Synergy HT BioTek plate reader at 37°C for 55 h. OD600 measurements were taken every 15 min after shaking for 2 min.
Total RNA was extracted from bacteria using the RNAsnap method [57]. Briefly, bacterial pellets were washed with AE Buffer (50 mM NaOAc pH 5.2, 10 mM EDTA) and then resuspended in 95% formamide, 18 mM EDTA, 1% 2-mercaptoethanol and 0.025% SDS. Bacterial lysis was performed by vortexing with 100 μm of zirconia beads (OPS Diagnostics) followed by incubation at 95°C. Nucleic acids were precipitated with ethanol and treated with Turbo-DNase (Ambion), followed by standard phenol extraction.
PCR products of 152 and 970 bp for rli60 and ilvD, respectively, were amplified from Lm genomic DNA using gene-specific primers for rli60 and ilvD (S2 Table). Thirty nano-grams of each PCR product was used as a template for synthesis of 32P-labeled probes using NEblot kit (New England Biolabs) and ɑ-32P dCTP (PerkinElmer), according to manufacturer’s instructions. Equal amounts of total RNA (5–10 μg) were separated on 1% agarose gel containing 7.4% formaldehyde and stained with ethidium bromide for visualization of rRNA. RNA was transferred to Biodyne B 0.45 μM nylon membrane (Pall Life Sciences) and cross-linked by UV (0.12 Joules). Pre-hybridization was performed at 65°C for 2 h in Church buffer (sodium phosphate buffer 0.25 M pH = 7.2, 1% BSA, 1 mM EDTA, and 7% SDS). Probes were added to Church buffer and hybridization was performed overnight at 65°C. Membranes were washed with 2XSSC 0.1% SDS, 1XSSC 0.1% SDS and 1XSSC. Light sensitive films (Fuji) were exposed to radioactive membranes for visualization of RNA-probe hybridizations. Sizes of RNA bands were evaluated using Transcript RNA Markers 0.2–10 kb (Sigma-Aldrich).
One μg of total RNA was reverse transcribed to cDNA using qScript (Quanta). qRT-PCR was performed on 10 ng of cDNA using FastStart Universal SYBR Green Master (Roche) in a StepOne Plus real time PCR system (Applied Biosystems). The transcription level of each gene was normalized to that of the reference gene rpoD. For the comparative analysis of rli60 and ilvD transcripts, a standard curve was prepared using Lm genomic DNA.
5'-RACE analysis was performed on total RNA extracts from Lm bacteria as described previously [58]. Briefly, 6 μg of total RNA were treated with Tobacco acid pyrophosphatase (TAP, Epicentre) and then ligated to a RNA linker using T4 RNA ligase 1 (Epicentre). TAP-untreated samples were analyzed in parallel in order to identify processed transcripts. Two μg of linker-ligated RNA were used for first-strand cDNA synthesis with random hexamers (Invitrogen) and Superscript III reverse transcriptase (Invitrogen). PCR amplification of the first-strand cDNA products was performed using a gene-specific primer (either rli60 or ilvD) and a linker specific primer. PCR products were then separated on 3% agarose gels, and TAP-specific bands were purified and cloned into pUC-18 for sequence analysis.
RNA was purified from intracellularly grown bacteria in bone marrow-derived macrophage cells (BMDM) as described previously [59]. BMDM cells used for infection experiments were isolated from 6–8 week-old female C57BL/6 mice (Envigo, Israel) as described previously [60] and cultured in Dulbecco’s Modified Eagle Medium (DMEM)-based media supplemented with 20% fetal bovine serum, sodium pyruvate (1 mM), L-glutamine (2 mM), β- Mercaptoethanol (0.05 mM) and monocyte-colony stimulating factor (M-CSF, L929-conditioned medium. Briefly, WT Lm bacteria were used to infect BMDM seeded in a 145 mm dish, resulting in a MOI of ~100. Thirty minutes after infection, BMDM monolayers were washed twice with PBS to remove unattached bacteria and fresh medium was added. At 1 h post-infection, gentamicin (50 μg/ml) was added to limit bacterial extracellular growth. 2 hours post infection, intracellular bacteria were collected using 0.45 μM filter membranes and flash-freezed in liquid nitrogen. Bacteria were recovered from filters by vortexing into AE buffer (50 mM NaOAc pH 5.2, 10 mM EDTA), and bacterial nucleic acids were extracted using hot (65°C) phenol with 1% SDS followed by ethanol precipitation. Rneasy Mini Kit Dnase on column (Qiagen) was used for Dnase treatment. Transcription levels of rli60 and ilvD in total RNA samples were measured with specific probes using the NanoString nCounter system, according to manufacturer standard procedures [61]. Total RNA extracted from bacteria grown in BHI was analyzed in parallel as a control.
WT Lm or indicated mutants (ΔcodY, Δrli60 or ΔcodY/Δrli60) harboring 6his-tagged ilvD at its native locus (ilvD-6his) or the translational fusion of the leader peptide to enhanced green florescent protein (EGFP) on the integrative pPL2 plasmid (pPL2 rli60-peptide-EGFP) were grown as indicated. Cultures were washed with Buffer-A (20mM Tris-HCl pH = 8, 0.5M NaCl, and 1 mM EDTA), resuspended in 20 ml of Buffer-A supplemented with 1 mM PMSF and lysed by an ultra-high-pressure homogenizer (Stansted Fluid Power) at 12000 psi. Lysates were centrifuged at 3,000 rpm for 10 min at 4°C. Proteins from the supernatants were precipitated on ice for 1 hour using 10% TCA and centrifuged at 3,800 rpm for 30 min at 4°C. Supernatants were discarded and the pellets were washed in Buffer-A with 5% TCA, then washed with ice-cold acetone twice. Dried pellets were resuspended in water with 2% SDS and analyzed for total protein content by modified Lowry assay. Samples with equal amounts of total proteins were separated on 15% SDS-polyacrylamide gels and transferred to nitrocellulose membranes. Proteins were probed either with mouse anti-6His tag (Abcam ab18184) or anti-GFP (Covance, a kind gift from E. Bacharach lab, Tel Aviv University) antibody used at a 1:1000 dilution, followed by HRP-conjugated goat anti-mouse IgG (Jackson ImmunoResearch, USA) at a 1:20,000 dilution. Homemade anti-GroEL antibody (a kind gift from A. Azem lab, Tel Aviv University) was used as an internal control at a dilution of 1:20,000, followed by HRP-conjugated goat anti-rabbit IgG at a dilution of 1:20,000. Western blots were developed by enhanced chemiluminescence reaction (ECL). ImageJ software (https://imagej.nih.gov/ij/) was used for densitometry of obtained bands.
Overnight E. coli K-12 ilvC::Km bacteria (Keio collection, a kind gift from U. Qimron lab, Tel Aviv University) harboring the rli60-luciferase reporter system (pPL2-rli60-luxABCDE or pPL2-Δrli60-luxABCDE) grown in MM cultures were adjusted to OD600 of 0.03 in fresh MM medium supplemented with 1, 10, or 100 μg/ml of BCAA (8, 80 and 800 μM, respectively), and grown in a Synergy HT BioTek plate reader at 37°C for 12 h. Luminescence measurements at 12 h time point at the different BCAA concentrations were normalized to the corresponding OD600.
Lm ilvD promoter was predicted using BPROM [62]. The leader peptide was predicted using ApE (http://biologylabs.utah.edu/jorgensen/wayned/ape). Terminator and anti-terminator structures were predicted using PASIFIC [32], with the kind help of Adi Millman from the Rotem Sorek lab, Weizmann institute. A scheme of the structures was prepared using Mfold [63].
Competitive index assay was performed as described previously [64]. Briefly, WT Lm, Δrli60 and codY-R61A/Δrli60 bacteria harboring the integrative pPL2 plasmid containing a kanamycin or spectinomycin resistance genes were grown in BHI medium at 30°C overnight. Bacterial cultures were washed in PBS, measured for OD600 and mutant culture (either Δrli60 or codY-R61A/Δrli60) was mixed with WT culture at a 1:1 ratio. Eight weeks old C57BL/6 female mice (Envigo) were infected via tail vein injections with 4 × 104 total bacteria in 200 μl of PBS. Animals were observed daily for any signs of illnesses and were euthanized 2 days post-infection. Spleens and livers were harvested and homogenized in 0.2% Triton X-100 in PBS, and the numbers of viable bacteria in each organ were determined by plating serial dilutions of homogenates onto BHI agar plates containing kanamycin or spectinomycin. The experiment was performed twice using five mice in each group per experiment.
|
10.1371/journal.pgen.1002468 | Nucleolar Association and Transcriptional Inhibition through 5S rDNA in Mammals | Changes in the spatial positioning of genes within the mammalian nucleus have been associated with transcriptional differences and thus have been hypothesized as a mode of regulation. In particular, the localization of genes to the nuclear and nucleolar peripheries is associated with transcriptional repression. However, the mechanistic basis, including the pertinent cis- elements, for such associations remains largely unknown. Here, we provide evidence that demonstrates a 119 bp 5S rDNA can influence nucleolar association in mammals. We found that integration of transgenes with 5S rDNA significantly increases the association of the host region with the nucleolus, and their degree of association correlates strongly with repression of a linked reporter gene. We further show that this mechanism may be functional in endogenous contexts: pseudogenes derived from 5S rDNA show biased conservation of their internal transcription factor binding sites and, in some cases, are frequently associated with the nucleolus. These results demonstrate that 5S rDNA sequence can significantly contribute to the positioning of a locus and suggest a novel, endogenous mechanism for nuclear organization in mammals.
| Eukaryotic genomes are compartmentalized within nuclei such that physiological events, including transcription and DNA replication, can efficiently occur. The mechanisms that regulate this organization represent an exciting, and equally enigmatic, subject of research. In mammals, the identification of elements that influence these associations has been impeded by the complex nature of the genomes. Here, we report the identification and characterization of such an element. We demonstrate that the integration of a 5S rDNA gene, a 119 base pair noncoding RNA transcribed by RNA polymerase III, into a new genomic location can significantly influence the association of the host region with the nucleolus. This positioning has drastic, inhibitory effects on the transcription of a neighboring protein coding gene transcribed by RNA polymerase II, demonstrating a functional relationship between localization and gene expression. We also provide data that suggest this may be an endogenous phenomenon, through a class of repetitive sequences derived from 5S rDNA. Together, our data not only demonstrate a structural role for 5S rDNA but also suggest that nuclear organization of mammalian genomes may be strongly influenced by repetitive sequences.
| The organization of DNA within mammalian nuclei is considered nonrandom [1]. A number of characteristics have been proposed to influence the position of a gene or chromosomal region within the nucleus, including gene density and transcriptional activity [2]. However, the parameters that drive nuclear organization are likely complex and remain largely enigmatic. Significant proportions of mammalian genomes are comprised of noncoding, repetitive elements, many of which are derived from RNA polymerase III (pol III) transcripts. An increasing number of examples have suggested diverse roles for repetitive elements in modulating transcription of neighboring protein-coding genes transcribed by RNA polymerase II (pol II) [3], [4], [5], [6]. In yeast, binding sites for the pol III transcription factor complex, TFIIIC, play a significant role in chromatin structure and nuclear organization: tRNA genes and tRNA-like sequences function as chromatin barriers to prevent the spread of heterochromatin, while in other contexts these elements cluster together often at the nuclear and nucleolar peripheries [7], [8]. This latter phenomenon typically results in silencing of nearby pol II-transcribed genes [9].
Moreover, just as pol II genes are thought to cluster in transcription ‘factories’ [10], active pol III also forms distinct foci in mammalian nuclei that contain a number of active pol III genes [11]. Since most pol III transcribed genes, including those of repetitive elements, carry internal promoters, they could confer intrinsic structural and regulatory properties to the surrounding genomic sequence upon insertion. Given their widespread and nonuniform distribution in mammalian genomes through repetitive elements, pol III promoters may have significant influence on chromatin structure. Furthermore, binding sites for pol III transcription factors within these elements may be under positive selection if beneficial for host genome fitness. To test these hypotheses, we focused on 5S rRNA genes (Figure 1A), which have long been known to possess unique qualities with regard to chromatin structure. We use a number of complimentary approaches to demonstrate that ectopic 5S rDNA sequence can mediate nucleolar association of a genomic region, with significant effects on local transcription. We also provide evidence that this mechanism may be active in endogenous contexts in the mouse genome: psuedogenes that are derived from 5S rDNA show preferential conservation of internal transcription factor binding sites can be bound by TFIIIC and localize to the nucleolar periphery.
A well-known nucleosome positioning sequence, 5S rDNA genes (endogenously present as multi-copy arrays in most eukaryotic genomes) have been observed to form large chromatin loops in Xenopus and mammalian systems [12], [13]. In agreement with observations in other eukaryotes, and recently published descriptions of chromatin associated with nucleoli in human cells [14], [15], [16], we found the mouse 5S rDNA gene array (located on the distal end of chromosome 8) associated with the nucleolar periphery in ∼40% of mouse embryonic stem (ES) cells (Figure S1A). If localization to the nucleolar periphery is an intrinsic quality of the 5S rRNA genes, then de novo insertion of these sequences into new genomic contexts should recapitulate this phenomenon. To study the effect of 5S rDNA sequence on sub-nuclear localization, we generated ES cell lines with stable, multicopy insertions of a reporter construct containing a single 5S rRNA gene (Tg5S) (Figure 1B). To determine whether transgenes with 5S sequence would be found at the nucleolar periphery, we then assessed localization of the stable transgenes by DNA FISH with a probe for the vector backbone relative to immunofluoresence against Nucleolin, a marker for the nucleolus [17]. In support of our hypothesis, we observed significantly more frequent localization to the nucleolar periphery of Tg5S (75%) compared with empty vector controls (Tg0, 31%, p = 8×10−4, Figure 1C, Figure S1B–S1D). Strikingly, several lines showed nearly constitutive association of the Tg5S signal with the nucleolus. This was not simply a reflection of copy number, as this pattern of localization was observed in both high- and low-copy Tg5S lines (Figure S1E, R2 = 0.087). Furthermore, association of Tg5S was higher than that of the 5S rDNA array (∼40%). This could be due to a dominant localization pattern imparted by Tg5S even at low copy, or additional forces acting to constrain localization of the endogenous 5S rDNA locus. We observed very little co-localization of Tg5S arrays and the 5S rDNA cluster (<1%, data not shown), demonstrating that these loci do not occupy the same compartment in the nucleoplasm. However, we found that the structural and functional integrity of the nucleolus was essential for localization through 5S rDNA. Reorganization of nucleolar components, through pharmocological inhibition of RNA polymerase I activity, resulted in a significant decrease of both Tg5S and 5S rDNA association with the nucleolus (Figure S2).
The nucleolar periphery has typically been thought of as a transcriptionally quiescent compartment, often associated with examples of constitutive [18], [19], [20] and facultative [21], [22], [23] heterochromatin. To study the effects of nucleolar association through the 5S rDNA mechanism on pol II transcription, we quantified mRNA levels of a reporter gene present on the vector: the Thymidine kinase (Tk) gene driven by the mouse Pgk1 promoter (Figure 1B). Tk mRNA levels, when normalized for copy number, are significantly decreased in Tg5S lines compared with Tg0 lines (4.68±2.22 and 8.09±1.55 arbitrary units, respectively; p = 6×10−3, Figure 1D, Figure S3). Interestingly, Tk mRNA levels show a strong negative correlation with nucleolar association: lines with the most frequent association had the lowest normalized expression (Figure 1E, R2 = 0.664). This relationship suggests that perinucleolar targeting of transgenes via the 5S rDNA sequence has inhibitory effects on pol II transcription.
The efficiency of nucleolar localization and transcriptional repression observed by Tg5S may be related to its ability to recruit the pol III transcriptional machinery. In yeast, the regulatory capacity of tRNA and tRNA-like sequences is dependent upon the TFIIIC complex [14]. To determine whether the TFIIIC complex is associated with transgene-5S rDNA, we used chromatin immunoprecipitation (ChIP) for a subunit of TFIIIC, TFIIIC65. We observed significant levels of TFIIIC65 association with transgene-5S rDNA, relative to the negative control (the Ascl2 promoter), in three of four Tg5S lines analyzed (Figure 2A). However, TFIIIC65 enrichment showed no clear correlation with localization (Figure 2B), Tk mRNA levels (Figure 2C), or copy number (Figure 2D). These data suggest that while the TFIIIC complex may participate in the localization and transcriptional attenuation we have observed for the 5S transgenes, levels of TFIIIC65 alone are not sufficient to explain these phenomena.
To determine whether specific histone modifications characterize the presence of a 5S rDNA, we surveyed the distribution of several modifications at various positions within the transgenes (Figure 3A). We analyzed one mark of active chromatin (H3K4me2, Figure 3B, Figure S4A), one mark of constitutive heterochromatin (H3K9me3, Figure 3C, Figure S4B), and two marks of facultative heterochromatin (H3K9me2 and H3K27me3, Figure 3D, 3E, Figure S4C, S4D), in four Tg5S and two Tg0 cell lines. As expected, cell lines with higher expression of Tk (Figure 3F) had increased levels of H3K4me2 at the Tk gene. Intriguingly, all Tg5S lines were characterized by high levels of H3K9me3 near the 5S rDNA, rather than the Tk gene body or promoter. Both patterns were evident irrespective of TFIIIC65 enrichment to the transgene-5S rDNA (Figure 3F). These observations suggest an association between the 5S rDNA sequence and the H3K9me3 modification.
The frequent nucleolar association of 5S rDNA-containing transgenes suggests the capacity to direct localization of a genomic region to the nucleolar periphery. However, this observation may also reflect preferential integration of Tg5S into a chromosomal region neighboring the nucleolus in the parental cells, rather than a change in localization. To discriminate between these possibilities, we identified the insertion site for several Tg5S ES lines. We mapped the transgene insertion in Tg5S#9 to the pseudoautosomal region (PAR) of the X chromosome [24] (Figure S5A). Since these ES cells are XY, we used the X-chromosome PAR of a line without a transgene insertion in this region as a control (Tg5S#6) to assess localization changes relative to a homolgous, wild-type chromosome. The PAR with the transgene insertion was more frequently associated with the nucleolus (61%) than a wild-type PAR (43%, p = 2×10−3, Figure 4A, 4B). Although nucleolar association of the wt PAR was similar to that of the 5S rDNA locus (39%), this frequency increased significantly upon Tg5S insertion.
Tg5S line #6 (Tg5S#6), contains an integration in the first intron of the silent RAR-related orphan receptor beta (Rorb) gene (Figure S5B). The allele containing the transgene array was discernable by DNA FISH and always overlapped with the genomic probe (Figure 4C). Nucleolar association was measured for both the wild type allele (wt allele) and the allele containing the Tg5S insertion (tg allele). As a control, we measured localization of the Rorb alleles in Tg5S#9, which does not have an insertion in this region. We detected significantly more DNA FISH signals for the tg allele associated with or internal to the nucleolus (68%) than for the wt allele (52%) in Tg5S#6 (Figure 4D, p = 0.01), or either allele in the control cell line (43%, p = 4×10−4). The localization frequency of the wt allele in the Tg5S#6 was not significantly different from the alleles in the control line (p = 0.5). Interestingly, wt Rorb alleles were associated with the nucleolus significantly more frequently than the 5S rDNA locus (chi-squared test, p = 5×10−9). Together, our observations from two independent insertion events, in two very different genomic contexts, demonstrate that ectopic 5S rDNA can influence the position of a locus.
Since localization by a Tg5S was associated with decreased transcriptional output of the Tk reporter gene, we hypothesized that the transgene insertion into the Rorb locus may similarly affect transcription of this gene. Rorb is not expressed in undifferentiated ES cells, therefore we differentiated the line with the Tg5S insertion in the Rorb gene (Tg5S#6) along with Tg5S#9, where the transgene insertion is not at the Rorb locus. Although activation of Rorb was variable between biological replicates, in each case Rorb expression was significantly reduced in Tg5S#6 (Figure 4E). Intriguingly, average Rorb expression in Tg5S#6 was 60% of that in Tg5S#9, suggesting that the presence of Tg5S at the Rorb locus has detrimental effects on its transcriptional activation.
The mouse genome contains >110 5S rDNA genes annotated outside the array on chromosome 8 (NCBI m37 mouse assembly, Table S1, Figure S6A). However, these elements show low overall sequence conservation and no predicted structural similarity to bona fide 5S rDNA, and are therefore unlikely to be functional components of the large ribosomal subunit (Figure S6B, S6C). Despite acquiring numerous mutations, a high proportion of these 5S pseudogenes retain perfect, or near-perfect, internal transcription factor binding sites (Figure 1A). This conservation correlates poorly with overall similarity of the 5S pseudogenes to the 5S rDNA consensus (R2 = 0.113, Figure 5A), suggesting this is not simply due to recent insertion events, but rather indicative of differential selective pressure within the psuedogene. We found a subset 5S pseudogene loci associated with the nucleolus in E14 ES cells at a frequency comprable to that of the 5S rDNA locus (Figure 5B, Figure S7), further supporting a positional effect for this sequence. TFIIIC association with pseudogenes was not well correlated with localization: by ChIP, we observed high levels of TFIIIC65 enrichment at only one of two pseudogene loci frequently associated with the nucleolus (Figure 5C). Therefore, if nucleolar association of these regions is mediated through 5S pseudogenes, then it may not require stable association of the TFIIIC complex, or perhaps involve altogether different mechanisms. Irrespective of the putative trans-factor, frequent nucleolar association of 5S pseudogenes further support a previously uncharacterized role for for these sequences as organizational cis-elements in the mammalian genome.
The relationship between the organization of chromatin within the nucleus and the regulation of individual genes has become an intensely studied subject. However, the complex nature of mammalian genomes has largely confounded efforts to understand the nature of this relationship. Several reports have catalogued the DNA and chromatin associated with the nuclear lamina and nucleolar periphery [15], [16], [25]. These findings have identified common characteristics of each domain, yet the basis for their presence at these compartments has remained less clear. Other studies have utilized fusion proteins to artificially tether lacO arrays to the nuclear lamina and other nuclear bodies [26], [27], [28]. Conversely, we have identified an endogenous sequence element, utilizing native nuclear machinery, that is capable of influencing subnuclear position. While transgenes with binding sites for the vertebrate insulator protein CTCF [29] have been shown to associate with nucleoli in a CTCF-dependent manner [30], it is not known how frequently endogenous CTCF sites recapitulate this phenomenon. Our data demonstrate that 5S rDNA sequence can confer a positional bias in localization, and correlates with an attentuation of nearby pol II transcription (summarized in Figure 6). Importantly, the localization of 5S rDNA pseudogenes to the nucleolar periphery suggest this event is not limited to ectopic transgene integrations. Biased conservation of transcription factor binding sites within 5S pseudogenes implies a functional role in their endogenous contexts. We propose that the internal transcription factor sites of 5S rDNA represents a novel, cis- acting influence of nuclear position in mammals. This hypothesis is supported by the observed enrichment of 5S rDNA sequences in nucleolar-associated chromatin of human cells [15], [16].
Recently, genome-wide maps of pol III and associated transcription factor binding in human cells have suggested structural roles reminiscent of what has been observed in yeast. These studies identified a number of “extra-TFIIIC” (ETC) loci, TFIIIC-bound regions not associated with a pol III complex or transcription unit [31], [32]. However, unlike the ETC loci of yeast, which are associated with silencing of nearby pol II-driven promoters, human ETC loci are correlated with active pol II genes. In contrast, we observed high levels of the repressive H3K9me3 modification surrounding the 5S rDNA sequence. Thus the functional properties of ETC loci appear to be distinct from the repressive effect on pol II transcription that we have observed for 5S rDNA. Importantly, this demonstrates that presence of the TFIIIC complex alone is not sufficient to explain the effect on neighboring pol II transcription, suggesting additional or alternative factors. For example, TFIIIC recruitment to 5S rDNA first requires the binding of the TFIIIA, which specifically recognizes the A and C boxes. Alternatively, the strong nucleosome positioning properties of 5S rDNA may play a role in its localization and repressive effects on neighboring pol II transcription.
Collectively, these observations suggest broad and diverse roles for pol III genes and derived sequences in the organization of chromatin within the mammalian nucleus. Because of their number, pol III promoters may exert a stronger influence on structural organization than pol II-directed gene activity. As pol III activity is coupled with differentiation and cellular metabolism, association of pol III and transcription factors with elements such as the 5S pseudgoenes we have described, may provide the basis for global organizational and structural changes within the nucleus in response to external stimuli [33].
E14 ES cells were cultured under standard conditions. To generate stable lines, ES cells were transfected with 1 µg of linearized plasmid using lipofectamine (Invitrogen) and selected in the presence of G418 for 14 days. We verified stable neomycin resistance for most lines by culturing with G418 and noted no increased levels of cell death. To induce differentiation, 2×105 ES cells were plated on 60 mm2 dishes without LIF and in the presence of 0.1 µM retinoic acid (Sigma) then cultured for 8 days, with passaging to maintain low cell density. For immunofluoresence and DNA FISH, cells were plated at low density and grown on coverslips 18–24 hours. Coverslips were permeabilized with cytoskeletal (CSK) buffer (100 mM NaCl, 300 mM sucrose, 3 mM MgCl2, and 10 mM PIPES pH 6.8), then fixed in 4% paraformaldehyde (PFA, Electron Microscopy Sciences) for 10 minutes at room temperature, washed twice for 5 minutes in 1× PBS (Cellgro), then stored in 75% ethanol at 4°C. Coverslips were re-hydrated with several washes of 1× PBS prior to DNA FISH experiments.
To generate Tg5S, the 5S rDNA sequence (a gift of B.Solner-Webb, Johns-Hopkins University) was cloned into a vector that contains the neomycin resistance gene under the control of the HSV promoter, and the Thymidine kinase gene under the control of the mouse Pgk1 promoter (a gift of D.Ciavatta, University of North Carolina). Tg0 was the vector without the 5S rDNA insert.
RNA was isolated from cultured cells with Trizol reagent (Invitrogen), DNAsed (RQ1 DNAse, Promega) and 500 ng of total RNA was used for each reaction. Samples were reverse transcribed using random-hexamer primers, with Superscript II Reverse Transcriptase (Invitrogen). Primers are listed in Table S2. Tk mRNA levels were first normalized to Gapdh levels. Real-time quantitative PCR was carried out on 25 ng of cDNA, in triplicate for each gene, on an ABI 3700 (Applied Biosystems), using the Fast SYBR Green Master Mix (Applied Biosystems). Data was analyzed in Microsoft Excel (Microsoft), and is shown as the log2-transformation of RNA levels relative to copy number. Statistical significance was determined by two-tailed t-test.
Transgene integration sites were determined using one of two approaches. The Tg5S#6 insertion site was identified using the TAIL PCR protocol, with degenerate primers as described [34]. The insertion site for Tg5S#9 was identified using inverse-PCR. Briefly, 1 µg of DNA was digested with XbaI, then ligated overnight with T4 ligase (NEB) at 2 ng/µl. DNA was concentrated by ethanol precipitation, and 50 ng of the ligation was used in a nested PCR reaction. PCR products were purified from an agarose gel using a QIAGEN gel extraction kit (QIAGEN) and quantified on a QUBIT flourometer (QIAGEN). PCR products were directly sequenced and analyzed by BLAST searches to the reference assembly of the mouse genome. Each insertion was confirmed by PCR. Primers are listed in Table S2. sequence. Each vector was linearized with XhoI (NEB) prior to lipofection.
Coverslips were rehydrated in 1× PBS, before blocking in 10 mg/ml IgG-free BSA (Jackson Immunochemical) and 0.2% Tween-20 (Fisher) for 20–30 minutes at room temperature. Rabbit anti-Nucleolin (Bethyl Laboritories, A300-711A) was added at 1∶400 dilution into blocking buffer and incubated overnight at 4°C. Coverslips were then washed with 1× PBS, and incubated with biotinylated goat-anti-rabbit antibody, diluted in blocking buffer at 1∶400, for 2–3 hours at room temperature. Following washes with 1× PBS, cells were post-fixed with 2% PFA for 3 minutes at room temperature, washed extensively with 1× PBS, then treated with 0.01 mg/ml pepsin (Sigma) diluted in pre-warmed 0.01 N HCl for 5 minutes, and then washed extensively with 1× PBS. Following a dehydration series in ethanol, DNA was denatured in 70% formamide (Ameresco) and 2× SSC (Cellgro) for 10–20 minutes at 85°C. After several washes with cold 2× SSC, cells were incubated with prehybridized DNA FISH probes (see below) overnight at 37°C. Coverslips were washed twice with 50% formamide and 2× SSC, twice with 2× SSC (one wash had 100 ng/ml DAPI added), once with 1× SSC. To detect biotinylated secondary antibodies, coverslips were then washed once with 4× SSC for 5 minutes, incubated for 20–30 minutes in Streptavidin-647 (Invitrogen) in 2 mg/ml BSA and 4× SSC, followed by 5 minute washes of 4× SSC, 4× SSC with 0.5% Tween-20 (Fisher), and 4× SSC. All washes and incubations for biotin detection were carried out at 37°C.
In addition to the vector backbone, the following BAC and fosmid probes were used in this study: BACs: 5∶134 (RP24-193L24), 7∶30 (RP23-151J21), 8∶126 (RP24-372G15), 10∶27 (RP24-213F23), 11∶74 (RP23174M12), Mid1 (RP24-229F18); and fosmids: 6∶112 (G135P69622C7), 8∶48 (G135P60371F8, G135P60172E7), 19∶19 (G135P64778C12), PAR (G135P601180H2) (all clones were acquired from CHORI BPRC). BACs and fosmids were isolated by a standard alkaline lysis protocol. Approximately 25 ng of DNA was labeled with BioPrime DNA labeling kit (Invitrogen), using FITC-conjugated dUTP (Roche), Cy3- or Cy5-conjugated dCTP (GE Healthcare), and stored in 70% ethanol at −20°C. To prepare FISH probes for hybridization, probes were precipitated with mouse Cot-1 DNA (Invitrogen), yeast tRNA (Invitrogen), and Salmon Sperm DNA (Invitrogen). After washes with 75% and 100% ethanol, probes were air-dried and denatured for 10 minutes in 50–100 µl of 100% formamide at 85°C. An equal volume of 2× hybridization buffer (25% dextran sulfate/4× SSC) was then added, and probes were pre-hybridized for 60 to 90 minutes at 37°C. Probes were stored at −20°C until use.
IF-DNA FISH was carried out as described in Methods. For transgene-nucleolus association, cells were visualized on Leica DMLB fluorescent microscope (Leica), captured on a Retiga 2000R Fast camera (Qimaging), using QCapture software (Qimaging), and merged with Adobe Photoshop (Adobe). DNA FISH signals were considered ‘nucleolar associated’ if the FISH signals were in contact with, or within, the Nucleolin signal. For determining nucleolar association of 5S rDNA, pseudogenes, and genomic loci with transgene insertions, Z-stacks of each channel were taken on a Ziess AxioImager M2 microscope, deconvolved using the Axiovision software package (Zeiss), then rendered in 3-dimensions using the ZEN Light Edition 2009 software (Zeiss). Signals were considered ‘internal’, if the center of the FISH signal was internal to the Nucleolin signal; ‘peripheral’ if the pixels of the FISH and Nucleolin signals were overlapping, but the center of the FISH signal was outside; and ‘not associated’ if there was visible distance between the DNA FISH signal and the outside of the Nucleolin-labeled nucleolus. Statistical significance was determined by chi-squared.
Copy number was determined by quantitative PCR to determine the number of Neo gene copies relative to an endogenous locus (the Ascl2 promoter), then normalized to a genomic DNA sample containing 1 copy of Neo for each diploid genome. Primers are listed in Table S2.
ES cells were trypsinized, counted, resuspended at 107 cells/ml and fixed with 1% formaldehyde. After quenching with 0.125 M glycine, cells were pelleted, washed once with cold 1×PBS, pelleted again and used for ChIP or frozen at −80°C. Protease inhibitors (Sigma) and PMSF (Sigma) were added to all steps until washing steps. For GTF3C5 and pseudogenes, ChIP was performed as described [35]. To measure histone modificiations or GTF3C5 association with transgenes, cell pellets were resuspended in solution L1 (50 mM HEPES-KOH, pH 7.5, 140 mM NaCl, 1 mM EDTA, 10% glycerol, 0.5% NP-40, 0.25% Triton X-100) at 107 cells/ml, mixed at 15 minutes and gently pelleted at 4°C. Cell pellet was resuspended in solution L2 (10 mM Tris-HCl, pH 8.0, 200 mM NaCl, 1 mM EDTA, 0.5 mM EGTA) at 107 cells/ml, mixed at 15 minutes and gently pelleted at 4°C. Cells were lysed in solution L3 (10 mM Tris-HCl, pH 8.0, 100 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, 0.1% Na-Deoxycholate, 0.5% N-lauroylsarcosine) for 10 minutes at 4°C. Chromatin was sheared by sonication to generate fragments 2–600 bp. Before immunoprecipitation, 1/10th of each sample was removed as ‘input’. 5 µg of Antibody (rabbit anti-GTF3C5, A301-242A, Bethyl Laboratories; rabbit anti-H3K4me2, 07-030 Millipore; mouse anti-H3K9me2, ab1220, Abcam; rabbit anti-H3K9me3, ab8898, Abcam; or mouse anti-H3L27me3, ab6002, Abcam) or normal rabbit sera (Abcam) was conjugated to protein A/G beads in 0.5%BSA/1×PBS overnight at 4°C on a nutating platform. Chromatin was incubated with bead-conjugated primary antibody overnight at 4°C with gentle mixing. For GTF3C5 ChIP, beads were then washed for 5 minutes at 4°C with gentle mixing, using the following solutions: Low Salt Buffer (0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris, 150 mM NaCl), twice; High Salt Buffer(0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris, 500 mM NaCl), once; LiCl buffer (1 mM EDTA, 10 mM Tris, 250 mM LiCl, 1% NP-40, 1% Na-Deoxycholate), twice; and TE (10 mM Tris, 1 mM EDTA), twice. For histone modifications, beads were washed 4 times with RIPA buffer, and once with TE containing 50 mM NaCl. Chromatin was eluted from beads with 2, 15-minute washes at 65°C using freshly prepared Elution Buffer (1% SDS/0.1 M NaHCO3). To isolate DNA, 5 M NaCl was added to pooled eluates or input chromatin to a final concentration of 0.2 M, and incubated for at least 4 hours at 65°C, then treated with 30 µg of Proteinase K (Roche) for 2 hours at 55°C. After addition of 10 µg linear acrylamide as a carrier (Ambion), DNA was extracted with 25∶24∶1 phenol∶choloform∶isoamyl alcohol (Sigma), precipitated with 100% ethanol, and resuspended in nuclease-free ddH20 (Promega). For psuedogenes, three replicates of quantitative PCR were carried out on an ABI 3700 (Applied Biosystems), using the Fast SYBR Green Master Mix (Applied Biosystems). For transgene and 5S rDNA enrichment, 2–5 replicates were performed on Bio-Rad CDX96 instrument, a using SsoFast EvaGreen Supermix (Bio-Rad). PCR primers are listed in Table S2. Data are displayed as enrichment of amplicon relative to a negative control region in each ChIP. Data was analyzed in Microsoft Excel (Microsoft); statistical significance was determined by two-tailed t-test.
|
10.1371/journal.ppat.1000425 | Mutation of the Protein Kinase C Site in Borna Disease Virus Phosphoprotein Abrogates Viral Interference with Neuronal Signaling and Restores Normal Synaptic Activity | Understanding the pathogenesis of infection by neurotropic viruses represents a major challenge and may improve our knowledge of many human neurological diseases for which viruses are thought to play a role. Borna disease virus (BDV) represents an attractive model system to analyze the molecular mechanisms whereby a virus can persist in the central nervous system (CNS) and lead to altered brain function, in the absence of overt cytolysis or inflammation. Recently, we showed that BDV selectively impairs neuronal plasticity through interfering with protein kinase C (PKC)–dependent signaling in neurons. Here, we tested the hypothesis that BDV phosphoprotein (P) may serve as a PKC decoy substrate when expressed in neurons, resulting in an interference with PKC-dependent signaling and impaired neuronal activity. By using a recombinant BDV with mutated PKC phosphorylation site on P, we demonstrate the central role of this protein in BDV pathogenesis. We first showed that the kinetics of dissemination of this recombinant virus was strongly delayed, suggesting that phosphorylation of P by PKC is required for optimal viral spread in neurons. Moreover, neurons infected with this mutant virus exhibited a normal pattern of phosphorylation of the PKC endogenous substrates MARCKS and SNAP-25. Finally, activity-dependent modulation of synaptic activity was restored, as assessed by measuring calcium dynamics in response to depolarization and the electrical properties of neuronal networks grown on microelectrode arrays. Therefore, preventing P phosphorylation by PKC abolishes viral interference with neuronal activity in response to stimulation. Our findings illustrate a novel example of viral interference with a differentiated neuronal function, mainly through competition with the PKC signaling pathway. In addition, we provide the first evidence that a viral protein can specifically interfere with stimulus-induced synaptic plasticity in neurons.
| Neurotropic viruses have evolved diverse strategies to persist in their host, with variable consequences for brain function. The investigation of these mechanisms of persistence and associated disease represent a major issue in viral pathogenesis, as it may also improve our understanding of human neurological diseases of unclear etiology for which viruses are thought to play a role. In this study, we have examined the mechanisms whereby the neurotropic Borna disease virus (BDV) can selectively interfere with synaptic plasticity upon infection of neurons. Using genetically engineered recombinant viruses, we show that the phosphorylation of BDV phosphoprotein (P) by the cellular protein kinase C (PKC) is the main determinant for this interference, mainly by competing with the phosphorylation of the natural PKC substrates in neurons. A mutant virus in which the PKC phosphorylation site of P has been destroyed no longer interferes with this signaling pathway. As a result, the calcium dynamics and electrical activity in response to stimulation of neurons infected with this mutant virus are completely corrected and become similar to that of non-infected neurons. Thus, our findings uncover a previously undescribed mechanism whereby a viral protein interferes with neuronal response to stimulation.
| The finding that persistent viruses could selectively affect differentiated functions of their target cell without causing cell lysis or widespread inflammation was first demonstrated more than 25 years ago [1]. This type of viral persistence, characterized by minimal cell damage, seems particularly well suited for the central nervous system (CNS) given the limited capacity of renewal of CNS resident cells, in particular of neurons. Viral interference with selected signaling pathways will nevertheless disrupt cellular homeostasis and cause disease [2]. As viral impairment of neurons may lead to behavioral or cognitive impairment, it was therefore hypothesized that persistent viruses could play a role in human mental disorders of unclear etiology [3],[4]. To date, the mechanisms whereby viruses can interfere with brain function are not well understood and are strongly dependent on the strategy that a given virus has developed to persist in the CNS [5],[6]. For viruses actively replicating in neuronal cells, one hypothesis is that the expression and/or accumulation of viral products in the cell may affect neuronal activity and cause disease. To date, it is clear that much is needed for a better understanding of the pathogenesis of persistent viral infections of the CNS and for the identification of the viral determinants responsible for the associated diseases.
Borna disease virus (BDV) is a highly neurotropic, non-cytolytic virus that provides an ideal paradigm for studying the behavioral correlates of CNS viral infections. BDV is an enveloped virus with a non-segmented, negative strand RNA genome [7],[8]. In contrast to other Mononegavirales, BDV replicates in the nucleus of infected cells [9] and uses the host cell splicing machinery for maturation of viral transcripts [10],[11]. The BDV compact genome encodes for six proteins, namely, the nucleoprotein (N), phosphoprotein (P), protein X, matrix protein (M), glycoprotein (G), and polymerase (L). Whereas M and G are involved in particle formation, P, N, L, and X are components of the polymerase complex. BDV infects a wide variety of mammals [12],[13] and is associated with a large spectrum of neurological disorders, ranging from immune-mediated diseases to behavioral alterations without inflammation [12],[14],[15]. These disorders are reminiscent of symptoms observed in certain human neuropsychiatric diseases [16]. Evidence suggest that BDV infections may also occur in humans, although a link between BDV infection and any human neurological disease has not been firmly established yet [17]–[19]. The neurobehavioral manifestations associated with BDV infections in animals are partly due to the selective tropism of BDV in the CNS for neurons of the cortex and hippocampus [15],[20],[21], which govern many cognitive and behavioral functions [22].
In an effort to better characterize the impact of BDV persistence on neuronal function, we recently analyzed the neuronal activity of primary cultures of neurons infected with BDV, using both functional imaging and electrophysiological approaches [23],[24]. These studies clearly showed that BDV interferes with activity-dependent plasticity, while leaving the basal properties of neuronal activity unaffected. Moreover, the selective impairment of neuronal plasticity due to BDV infection was correlated to a reduced phosphorylation of the neuronal targets of Protein Kinase C (PKC), a kinase that plays important roles in the regulation of neuronal activity [25]. Amongst the different viral proteins, BDV P appeared as the most plausible candidate for mediating this interference. Similar to the phosphoproteins of other Mononegavirales, BDV P is a component of the viral polymerase complex, which serves several functions in the viral life cycle. These functions are thought to be regulated, at least in part, by its phosphorylation by cellular kinases [26]. BDV P is preferentially phosphorylated at serine residues 26 and 28 by PKC and, to a lesser extent, at serine residues 70 and 86 by casein kinase II (CKII) [27]. Taken together, these observations led us to postulate that BDV P may serve as a PKC kinase decoy substrate when expressed in neurons, resulting in the decreased phosphorylation of other PKC neuronal targets. Although transfection experiments using a BDV P expressing plasmid provided the first evidence that this could indeed be the case [24], a formal demonstration of the role of BDV P in PKC-dependent signaling and on neuronal activity was needed.
The newly established reverse genetics technique allowing the generation of recombinant BDV (rBDV) entirely from cDNA has provided means to test this hypothesis directly [28]. Recently, we characterized rBDV expressing P mutants lacking either the PKC or the CKII phosphorylation sites, upon replacement of the corresponding serine residues with alanines to abrogate phosphorylation [29]. We showed that phosphorylation of BDV P acts as a negative regulator of the viral polymerase complex activity, in contrast to what has been shown for other Mononegavirales. Here, we infected primary cultures of rat hippocampal and cortical neurons with these different recombinant viruses and analyzed their phenotype and responses to stimulation. Using a rBDV where P can no longer be phosphorylated by PKC, we demonstrate a complete suppression of viral interference with neuronal activity. This was shown not only at the molecular level in terms of PKC-dependent signaling, but also at the functional level by studying calcium responsiveness to depolarization and the electrophysiological network properties of cultured neurons.
Thus, our findings illustrate a novel example of viral interference with a differentiated neuronal function, mainly through competition with the PKC signaling pathway. In addition, we provide the first evidence that a viral protein can specifically interfere with stimulus-induced synaptic plasticity in neurons.
To study the impact of BDV P phosphorylation on viral neuropathogenesis, we used a recombinant BDV in which the two serine (S) residues in position 26 and 28 of P have been replaced by alanine (A) residues (rBDV-AASS). We have previously shown that these mutations, when introduced using the recently established technique for generating BDV from cDNA [28], completely abolishes BDV P phosphorylation by PKC, while still supporting full polymerase activity and viral growth [27],[29]. As a control, we also used a recombinant wild-type BDV (rBDV-wt), composed of the canonical sequence of wild-type BDV strain He/80 [30].
We first analyzed the impact of these mutations on the kinetics of viral spread in primary hippocampal rat neuronal cultures. To this end, we infected neuron-rich cultures (>80% neurons) one day after their plating with 300 focus forming units (FFU) per well of either rBDV-wt or rBDV-AASS cell-released virus (prepared from persistently infected Vero cells). We studied viral dissemination occurring after primary infection by assessing expression of the BDV nucleoprotein at different times post-infection, using immunofluorescence microscopy to quantify viral spread. Consistent with our previous reports using wild-type BDV [31], infection with rBDV-wt was apparent by 5 to 7 days post-infection, with a low percentage (<10%) of all neurons being positive for BDV nucleoprotein. The virus spread rapidly increased thereafter and by day 20, more than 95% of neurons were infected (Figure 1A). In contrast, spread of the rBDV-AASS virus was significantly delayed, in particular between days 10 and 17 post-infection, a time when viral dissemination speed is usually maximal due to the increased density of the neuronal network [31]. Thus, the phosphorylation of BDV P seems to be required for optimal transneuronal virus spread in primary neurons.
Importantly, despite the delayed kinetics for rBDV-AASS, infection ultimately disseminated to the whole neuronal cultures. By 20–21 days post-infection, the large majority of neurons was positive for BDV antigens, with no significant differences between cultures infected with either rBDV-wt or rBDV-AASS, as assessed by immunofluorescence analysis (Figure 1B). At this stage, quantitative Western blot analysis revealed that comparable amounts of N, P and X viral proteins were present in neurons infected with both rBDV-wt and rBDV-AASS viruses. Consistent with our previous characterization of rBDV-AASS, we observed a delayed migration for the X protein [29], presumably resulting from the two amino acids substitutions introduced in X when generating the rBDV-AASS mutant.
In conclusion, both recombinant viruses were able to infect the totality of our neuronal cultures, albeit with a delayed kinetics for rBDV-AASS. For our subsequent signaling and functional studies, we therefore used neurons that had been infected for at least 21 days and verified for each experiment that infection was indeed complete prior to any subsequent analysis.
Recently, we have shown that BDV-induced impairment of potentiation of synaptic activity was due to an interference with the PKC-dependent phosphorylation of synaptic proteins that modulate neuronal activity [24]. To determine the impact of the destruction of the PKC phosphorylation sites on BDV P, we directly stimulated the PKC pathway of neurons using the phorbol ester PMA and analyzed the phosphorylation status of two major PKC targets in neurons, myristoylated alanine-rich C kinase substrate (MARCKS) and synaptosomal-associated protein of 25 kDa (SNAP25) [32],[33]. Consistent with our previous reports, we showed by quantitative Western blot analysis that the phosphorylation of these two neuronal PKC targets was significantly impaired in neurons infected with rBDV-wt (Figure 2A and 2B). In contrast, there was a complete restoration of phospho-MARCKS and phospho-SNAP25 levels in neurons infected with rBDV-AASS, with levels being comparable to those observed in non-infected neurons following PKC stimulation.
SNAP25 is a synaptic protein that plays an essential role in neurotransmitter release through regulation of synaptic vesicle exocytosis [34]. In addition, SNAP-25 also modulates calcium dynamics in response to depolarization by acting on Voltage-Gated Calcium Channels (VGCC). It has recently been shown that activity-dependent phosphorylation of SNAP-25 is mediated by PKC and is required for negative regulation of VGCCs [35]. Indeed, PKC phosphorylation of SNAP-25, by promoting inhibition of VGCCs decreases calcium signaling and controls neuronal excitability. This led us to investigate whether the differences observed between neurons infected with rBDV-wt and rBDV-AASS in the phosphorylation of SNAP25 would have an impact on their calcium responsiveness upon depolarization. We therefore analyzed the kinetics of calcium changes in response to depolarization after exposure of infected and control neurons to 50 mM KCl. Infection of neurons with rBDV-wt significantly enhanced their response to depolarization, with a peak calcium response being about fifty percent stronger than control non-infected neurons. In sharp contrast, the calcium response of neurons infected with rBDV-AASS was similar to that of non-infected neurons (Figure 3A and 3B). Together, these findings suggest that BDV mediated interference with PKC-dependent phosphorylation, by reducing phospho-SNAP-25 levels, leads to calcium hyper-responsiveness and that mutation of the BDV P phosphorylation sites restores normal calcium responses.
Since P is expressed from a bicistronic mRNA encoding P and X, the introduced mutation into the open reading frame of P in the rBDV-AASS mutant also resulted in two amino acid substitutions in the X protein (Figure 4A). We previously showed that these mutations had no impact on X binding efficiency to P or its ability to interfere with the polymerase activity [29]. However, we wanted to exclude formally the possible contribution of the X mutations in the phenotype of the rBDV-AASS mutant. To explore the importance of BDV-P phosphorylation in the presence of wild-type X, we generated a new recombinant BDV, in which S26 and S28 of BDV P were substituted with leucine residues (rBDV-LLSS, Figure 4A). Similar to rBDV-AASS, viral spread of rBDV-LLSS was delayed upon infection of hippocampal neurons (Figure 4B). Likewise, we also demonstrated the restoration of phospho-MARCKS and phospho-SNAP25 levels in neurons infected with rBDV-LLSS, with levels being comparable to those observed in non-infected neurons following PKC stimulation (Figure 4C). Finally, the calcium hyper-responsiveness observed in rBDV-wt infected neurons was also corrected in neurons infected with rBDV-LLSS (Figure 4D). Thus, the phenotype observed with rBDV-LLSS is very similar to rBDV-AASS, providing additional evidence that our findings are indeed due to phosphorylation of BDV-P by PKC and not to the mutations present in X.
The S26/S28A mutation of BDV P corrects the defects in electrical activity observed in rBDV-wt infected neurons. Given the dramatic consequences of the S26/28A mutation on neuronal signaling and calcium responsiveness, we next studied its impact on the electrophysiological properties of infected neurons. Recently, we described a cell culture system using microelectrode arrays (MEA), which allows to monitor the firing pattern of a neuronal network grown on a grid of sixty electrodes embedded in a culture dish (Figure 5A) [36],[37]. Using this system, we showed that BDV selectively blocks activity-dependent enhancement of neuronal network activity, one form of synaptic plasticity thought to be important for learning and memory [23],[38]. Given the central role of PKC in synaptic plasticity, we hypothesized that this defect could be linked to BDV interference with this signaling pathway. To test this hypothesis, we compared the electrophysiological properties of cultures of cortical neurons infected with rBDV-wt or rBDV-AASS viruses, using the MEA culture system. All experiments were again performed at day 21, to allow spreading of both viruses to the totality of the MEA cultures. At this time point, neurons have developed a rich network of processes and form numerous functional synaptic contacts [31],[39]. In agreement with our previous results [23], we did not observe any significant difference in the spontaneous network firing activity between neurons infected with either rBDV-wt or rBDV-AASS viruses (Figure 5B). This firing pattern was also indistinguishable from that of control non-infected neurons. Next, we induced increased synaptic efficacy by exposing neuronal cultures for 15 min to 50 µM of bicuculline, a GABAA receptor antagonist. Treatment with this antagonist leads to the removal of the tonic inhibition imposed by GABAergic interneurons on the network [40]. As a result, we observed a significant increase of the mean burst frequency, which shifted from 0.175 Hz to 0.285–0.31 Hz (Figure 5B). Here again, the behavior of all neuronal cultures was remarkably similar, regardless of their infection status. Interestingly, the increase in the strength of the synaptic connections triggered by bicuculline lasts for several hours upon removal of the drug [40],[41] and is thought to represent the cellular basis of learning and memory [38]. In non-infected neurons, we indeed observed this maintenance of a high level of network activity, lasting more than two hours after washout of the drug. In contrast, neurons infected with rBDV-wt had returned to basal levels of network activity already one hour after bicuculline washout, confirming our previous results using wild-type BDV [23]. Very strikingly, the network properties of neurons infected with rBDV-AASS were completely different, as we observed the maintenance of high levels of synaptic activity persisting up to two hours after bicuculline washout, similarly to non-infected neurons. Therefore, a recombinant BDV which can no longer be phosphorylated by PKC on its mutated P protein has lost its capacity to block activity-induced synaptic potentiation.
The goal of our study was to provide further information about the mechanisms whereby BDV infection of neurons selectively interferes with synaptic plasticity and to identify the viral determinant responsible for this interference. Using a recombinant virus in which the PKC phosphorylation sites of BDV P protein have been destroyed, we demonstrate that primary cultures of neurons infected with this recombinant virus exhibit a behavior that becomes indistinguishable in many aspects to that of control, non-infected neurons. Therefore, our results clearly establish that BDV interference with PKC signaling, but also with calcium responses to depolarization and with network electrical properties, all result from the competition mediated by P with the phosphorylation of endogenous PKC substrates in neurons. We therefore propose a pathogenesis mechanism by which BDV would use PKC-dependent phosphorylation of P for its optimal spread in neurons, at the expense of an impaired response to potentiation stimuli of the infected neurons (Figure 6). Our findings provide strong evidence for a novel mechanism whereby a viral protein selectively blocks neuronal plasticity, representing a fascinating aspect of viral interference with neuronal functioning.
Interestingly, both mutant viruses rBDV-AASS and rBDV-LLSS were strongly delayed in its capacity to spread within the neuronal cultures. These findings suggest that PKC-dependent phosphorylation of P is an important parameter for efficient transmission of BDV from neuron to neuron. This may also explain the preferential BDV infection of CNS regions where PKC activity is high, such as hippocampus [42]. The underlying mechanism for this delayed spread is unclear and it is presently not possible to discriminate whether it is due to reduced cell-to-cell transmission or to a reduced capacity of the virus to be transported along neuronal processes. It does not seem to result from a decreased efficacy of viral replication, since analysis of the steady state levels of viral transcripts in rBDV-AASS- as well as in rBDV-LLSS-infected cells appeared to be normal [29]. It could be due to a less efficient transport of BDV ribonucleoparticles along the neuronal processes, resulting from impaired interaction of non-phosphorylated P with yet unidentified neuronal motor proteins involved in BDV trans-neuronal spread. Alternatively, the delayed spread may be a consequence of impaired virus assembly or release, similar to human respiratory syncytial virus, where phosphorylation of P regulates the viral budding process by blocking the interaction of P with the viral matrix protein [43]. Finally, mutations present in the X protein should also be taken into account, as we cannot exclude the possibility that these mutations may affect unknown functions of X, leading to impaired viral spread. However, this latter possibility seems unlikely as the rBDV-LLSS mutant, which has no mutation in X, also exhibits delayed spread kinetics in neurons.
Infection with the rBDV-AASS virus led to a complete restoration of the PKC-dependent phosphorylation of two neuronal targets that play crucial roles in modulating neuronal plasticity. Indeed, phosphorylation of MARCKS by PKC is implicated in actin-dependent cytoskeletal plasticity [44] and in the maintenance of long-term potentiation in vivo [45]. SNAP-25 is not only central for neuronal exocytosis but also for the regulation of calcium responsiveness. Modulation of neuronal excitability by SNAP-25, which is dependent on PKC, is thought to have crucial consequence for brain functioning [34]. The calcium hyper-excitability in response to depolarization that we observed after infection with rBDV-wt is consistent with a recent study showing that non-phosphorylable (S187A) SNAP-25 mutants also display increased calcium responsiveness [35]. Moreover, SNAP-25 S187A mutant mice exhibit behavioral abnormalities and hyperlocomotor activity, similar to what has been described following BDV infection [46]. Finally, genetic studies have demonstrated an association of polymorphisms in the human SNAP-25 gene with attention deficit hyperactivity disorder or cognitive performance [47],[48]. Nevertheless, as BDV is likely to affect all PKC neuronal substrates in the infected neurons, the relative contribution of the decreased phosphorylation of each of these substrates, including SNAP-25, is difficult to appreciate.
Very striking was the impact on the neuronal electrical activity measured using the MEA system. Since neurons infected with the rBDV-AASS mutant displayed a response to bicuculline stimulation that became indistinguishable from that of non-infected neurons, it is likely that interference with PKC signaling is indeed a main determinant for BDV impairment of neuronal functioning. The question remains open of whether the impairment of neuronal activity due to BDV infection is solely dependent on PKC inhibition. Using a global proteomic approach, we recently demonstrated that other neuronal pathways were altered by BDV infection, even prior to any stimulation [49]. At present, the link between some of these pathways, such as chromatin dynamics or transcriptional regulation and PKC-dependent signaling is not clear. Using infection with rBDV-AASS, in which PKC interference is suppressed, will be instrumental to test whether these other pathways are still affected, allowing to gain further insight on the relationship between the different BDV targets in neurons.
Phosphoproteins of non-segmented negative-strand RNA viruses are subunits of the viral polymerase complex and are all phosphorylated by host kinases [26]. Although many studies have addressed the role of P phosphorylation on the virus life cycle, very few have tested its consequences on the physiology of the target cell. In theory, other viral phosphoproteins could also block endogenous phosphorylation, including that of PKC. For example, it has been shown that rabies P can be phosphorylated by PKC [50]. However, BDV represents a unique example where neurons can be persistently infected with strong antigenic load and no widespread cytolysis. Thus, a possible interference with PKC-dependent phosphorylation in neurons becomes apparent following BDV infection due to its outstanding non-cytolytic replication strategy, whereas it would not be detected with other viral systems that kill their target cell within a few days.
As for the other Mononegavirales, BDV P is a multifunctional protein, which plays many roles in the virus life cycle. Besides its interference with PKC signaling, P has been reported to influence cellular functions at different levels. In particular, recent studies revealed interactions of P with a neurite outgrowth factor, amphoterin/HMG-1 [51], with the Traf family member-associated NF-κB (TANK)-binding kinase-1 (TBK-1) [52] and the gamma-aminobutyric acid receptor-associated protein [53]. Since these studies were based on transient transfection using non-neuronal cells, the relevance of these findings for BDV pathogenesis remains elusive and awaits further confirmation. It has also been shown that the expression of P in glial cells of transgenic mice leads to behavioral abnormalities [54], although the underlying mechanism was not identified. Given the important role of astrocytes in the regulation of neuronal activity [55], one hypothesis could be that P could also interfere with PKC-signaling in astrocytes and thereby disrupt glia-neuron communication. Interestingly, MARCKS, the main PKC substrate, is also expressed in astrocytes [56]. Finally, the two phospho-serine residues of P in themselves could have other unknown effects that could contribute to the phenotype observed, besides acting as PKC decoy substrates. Although none of the other known functions of P were affected, there may still be other uncharacterized functions of P that may be involved in the regulation of neuronal activity.
To date, the analysis of the mechanisms underlying viral interference with neuron-specific differentiated functions, particularly those that support synaptic activity has been hampered by the lack of suitable model systems and easily testable hypotheses. BDV infection has provided considerable new insight on these issues, and the recent availability of a reverse genetics system has offered a powerful tool to assess directly the role of individual viral proteins in virus-host interplay and pathogenicity. Our data unambiguously demonstrate the role of P as a decoy substrate interfering with PKC signaling pathway, a kinase which plays important roles in learning and behavior [25]. Moreover, they reveal an original strategy for a neurotropic persistent virus and provide clues to better understand the basis of neuronal impairment caused by BDV. It will be important in the future to test the impact of BDV P in vivo, either using animal models for BDV infection or expressing the wild-type and mutant forms of P in selected brain areas.
Hippocampal neurons were prepared from newborn Sprague-Dawley rats and maintained in Neurobasal medium (Invitrogen, Cergy-Pontoise, France) supplemented with 0.5 mM glutamine, 1% fetal calf serum, 1% Penicillin/Streptomycin and 2% B-27 supplement (Invitrogen), as described [31],[57]. Neuronal cultures contained more than 80% neurons, as assessed by staining with the neuron-specific markers MAP-2 or ß-III Tubulin (data not shown). Neurons were infected one day after plating with cell-free BDV. Cell-released virus stocks were prepared as described [31],[57], using Vero cells persistently infected with the different recombinant viruses. BDV infection of neurons was verified by immunofluorescence for each experiment.
We used: mouse monoclonal antibodies to SNAP25 (Synaptic Systems, Goettingen, Germany), ß-tubulin (Sigma-Aldrich, Lyon, France), rabbit polyclonals to phospho-MARCKS (Ser152/156, a site specifically phosphorylated by PKC; Cell Signaling Technology, Danvers, Massachusetts, USA), MARCKS (Chemicon-Millipore, Saint-Quentin-en-Yvelines, France). Phospho-SNAP25 (Ser187) antibody, a site specific for PKC phosphorylation [33] was kindly provided by Pr. M. Takahashi (Kitasato-University School of Medicine, Kitasato, Japan). All other antibodies have been described elsewhere [57]. Pharmacological agents were used at the following final concentrations: 1 µM PMA, 1 mM tetrodotoxin (TTX; Sigma-Aldrich), 100 mM of NMDA receptor blocker D-(-)-2-amino-5-phosphopentanoic acid (APV) and 40 mM of AMPA/kainate receptor blocker 6-cyano-7-nitroquinoxaline-2,3-dione disodium (CNQX; Tocris Bioscience, Bristol, United Kingdom). Bicuculline (Bicuculline methiodide, Tocris Biosciences) was used at a final concentration of 50 µM.
To introduce point mutations into the P gene of the full-length BDV genome (pBDV-LLSS), assembly PCR using the plasmid pBRPolII-HrBDVc as a template was carried out as described [29].
Standard immunofluorescence was performed as described previously [49]. Briefly, cells grown on glass coverslips were fixed for 20 min at room temperature with 4% paraformaldehyde, permeabilized using PBS+0.1% Triton-X100 during 4 min, rinsed with PBS, and blocked overnight at 4°C with PBS+2% normal goat serum. Incubation for 1 h at room temperature or overnight at 4°C with primary antibodies was followed, after several washes in PBS, by a 1-h incubation at room temperature with secondary antibodies. After extensive washing, coverslips were mounted by using Vectashield containing DAPI to stain nuclei (Vector Laboratories, Burlingame, CA).
Neurons were incubated for 60 min at 37°C in culture medium containing the blockers of neuronal activity TTX, APV and CNQX. Following this resting period, the medium was replaced by medium containing 1 µM PMA and neurons were stimulated for 10 min. Neurons were then rapidly washed in ice-cold PBS and harvested in lysis buffer containing phosphatase inhibitors [57]. The rest of the procedure was performed as described [49]. Briefly, equivalent amounts of cell lysates were separated by electrophoresis using 10% Bis-Tris Nu-PAGE gels (Invitrogen) and then transferred onto nitrocellulose membranes (Hybond-C extra, Amersham Biosciences, Orsay, France). After blocking (Li-Cor blocking buffer, ScienceTec, Les Ulis, France, or Tris buffer saline containing 5% non-fat dry milk), membranes were incubated with primary antibodies. Secondary fluorescent antibodies used were the following: IRDye 800CW goat anti-Mouse IgG (Li-Cor) or Alexa Fluor 680 goat anti-rabbit IgG (Invitrogen). Laser scanning and quantitative analyses of the blots were performed using the Odyssey Infrared Imaging System (Li-Cor). Quantification of protein phosphorylation was carried out by measuring the intensity of fluorescence of the band corresponding to the phosphorylated protein normalized by ß-tubulin expression, due to inefficient stripping of phospho-MARCKS and phospho-SNAP-25 antibodies binding. In parallel, total levels for each protein was verified on a separate blot. Results are expressed as percentage of increase over the mean of unstimulated controls, which was set to 100%.
Neurons grown on flat-bottom 96-well plates were incubated for 30 min at 37°C in HBSS (Invitrogen) –BSA (Sigma) solution containing 2.5 mM probenecid (Sigma), pH 7.45 supplemented with fluo-3 acetoxymethyl (AM) 1 mM (Invitrogen) and 20% pluronic F-17 (Sigma). After this incubation period, neurons were washed twice with HBSS-BSA-probenecid solution and placed into a 37°C incubator in the dark for 30 min. Fluorescence was measured at 460–490 nm excitation and 515 nm emission in each well, using a Novostar plate reader (BMG Labtech, Champigny s/Marne, France). For each well, a series of 250 recordings (one per second) was performed; neurons were first exposed to 50 mM KCl (Sigma) from 10 to 90 s, then to HBSS-BSA-probenecid solution containing 50 µM thapsigargin (Calbiochem, Fontenay-sous-Bois, France), 5 µM ionomycin (Calbiochem) and 10 µM EGTA (Sigma) from 90 to 177 s and finally to HBSS-BSA-probenecid solution containing 120 mM CaCl2 from 177 to 250s. [Ca2+]i was calculated using the equation [Ca2+]i = Kd (F−Fmin)/(Fmax−F), where Kd is the dissociation constant of the Ca2+-fluo-3 complex (390 nM), and F represents the fluorescence intensity of the cells expressed as the ratio between the highest fluorescence measurement between 10 and 90 s and the baseline. Fmin corresponds to the minimum fluorescence between 90 and 177 s. Fmax represents the maximum fluorescence between 177 and 250 s (see Figure 3A).
Neuronal cortical cultures were prepared from embryonic Sprague Dawley rats at gestational day 18, according to a previously described protocol [37]. Neurons were seeded at a density of 105 cells per MEA and half of the MEA dishes were infected with BDV on day 1. All experiments were made on day 21, to allow spreading of the different recombinant viruses to the totality of the MEA dishes. Signals corresponding to the electrical activity from the 60 electrodes of the MEA were recorded using the MC Rack Software (Multi Channel Systems GmbH, Reutlingen, Germany) for online visualization and raw data storage. The signal corresponding to the firing of a single action potential by a neuron in the vicinity of an electrode was identified as a spike. We also detected high frequency grouped spikes trains, known as bursts, which represent an important parameter of the analysis of neuronal network activity [58]. Spikes and bursts were detected by a dedicated analysis software developed at INSERM U862 (Bordeaux, France) [37], which computes the signal obtained from the electrodes, calculates a threshold and detects a spike every time the signal crosses this threshold with a negative slope. The threshold was set to a minimum of three standard deviations of the average noise amplitude computed over the whole recording and applied from the signal averaged value as a baseline [59]. Bursts were defined as a series of ≥3 spikes occurring in less than 100 ms. Measures were performed under spontaneous conditions, during a 15 min stimulation period using 50 µM Bicuculline, and after washout of bicuculline upon perfusing the MEA dish with 2.5 ml medium. After each manipulation, neurons were allowed to rest for 2 min before recordings were taken, to avoid vehicle effects. For each condition, recordings were performed over a 3 min period, and the mean burst frequency was calculated by averaging the results obtained for all electrodes.
Data are presented as mean±standard error of the mean (s.e.m). Statistical significance was determined using Student's unpaired t-test.
The SwissProt (http://www.expasy.org/sprot/) accession numbers for proteins mentioned in the text are BDV Nucleoprotein (Q01552), BDV Phosphoprotein (P26668), BDV X protein (Q912Z9), BDV matrix protein (P52637), BDV polymerase (P52639), MARCKS (P30009), SNAP-25 (P60881), PKC epsilon (P09216), CKII (P19139), TBK-1 (Q9WUN2). The GenBank (http://www.ncbi.nlm.nih.gov/Genbank) accession number for the strain discussed in this paper is BDV strain He/80/FR (AJ311522).
|
10.1371/journal.ppat.1002774 | A Highly Intensified ART Regimen Induces Long-Term Viral Suppression and Restriction of the Viral Reservoir in a Simian AIDS Model | Stably suppressed viremia during ART is essential for establishing reliable simian models for HIV/AIDS. We tested the efficacy of a multidrug ART (highly intensified ART) in a wide range of viremic conditions (103–107 viral RNA copies/mL) in SIVmac251-infected rhesus macaques, and its impact on the viral reservoir. Eleven macaques in the pre-AIDS stage of the disease were treated with a multidrug combination (highly intensified ART) consisting of two nucleosidic/nucleotidic reverse transcriptase inhibitors (emtricitabine and tenofovir), an integrase inhibitor (raltegravir), a protease inhibitor (ritonavir-boosted darunavir) and the CCR5 blocker maraviroc. All animals stably displayed viral loads below the limit of detection of the assay (i.e. <40 RNA copies/mL) after starting highly intensified ART. By increasing the sensitivity of the assay to 3 RNA copies/mL, viral load was still below the limit of detection in all subjects tested. Importantly, viral DNA resulted below the assay detection limit (<2 copies of DNA/5*105 cells) in PBMCs and rectal biopsies of all animals at the end of the follow-up, and in lymph node biopsies from the majority of the study subjects. Moreover, highly intensified ART decreased central/transitional memory, effector memory and activated (HLA-DR+) effector memory CD4+ T-cells in vivo, in line with the role of these subsets as the main cell subpopulations harbouring the virus. Finally, treatment with highly intensified ART at viral load rebound following suspension of a previous anti-reservoir therapy eventually improved the spontaneous containment of viral load following suspension of the second therapeutic cycle, thus leading to a persistent suppression of viremia in the absence of ART. In conclusion, we show, for the first time, complete suppression of viral load by highly intensified ART and a likely associated restriction of the viral reservoir in the macaque AIDS model, making it a useful platform for testing potential cures for AIDS.
| Novel research aimed at finding a cure for AIDS requires animal models responding to human antiretroviral drugs. However, there have been few antiretrovirals cross-active against the simian viruses. In this study, we expanded the arsenal of drugs active against the simian retrovirus SIVmac251 and showed that this virus is inhibited by the protease inhibitor, darunavir, and the CCR5 blocker, maraviroc. Administration of these two drugs in combination with the reverse transcriptase inhibitors, tenofovir and emtricitabine, and the integrase inhibitor, raltegravir, resulted in prolonged plasma viral loads below assay detection limits, and, surprisingly, restricted the viral reservoir, a marker of which is viral DNA. We then decided to employ this multidrug regimen (termed “highly intensified ART”) in order to increase the potency of a previous strategy based on the gold drug auranofin, which recently proved able to restrict the viral reservoir in vivo. A short course of highly intensified ART following the previous treatment resulted, upon therapy suspension, in a remarkably spontaneous control of the infection, that may pave the way to a persistent suppression of viremia in the absence of ART. These results corroborate the robustness of the macaque AIDS model as a vanguard for potentially future treatments for HIV in humans.
| The study of persistence of viral sanctuaries during antiretroviral therapy (ART) and the possibility for their therapeutic targeting is crucial for eradication of HIV-1. Animal models for lentiviral persistence during therapy are therefore needed. The creation of such animal models requires knowledge of the response of animal lentiviruses to antiretroviral drugs adopted in treatment of humans with HIV-1. Finding cross-active drugs has been a difficult task because non-HIV-1 lentiviruses often mimic drug resistance mutations found in HIV-1. This mimicry has been shown for the viral protease [1] and for the portion of reverse transcriptase (RT) that binds the non-nucleosidic reverse transcriptase inhibitors (NNRTIs) [2].
One of the current models is based on macaques infected with a molecularly engineered simian immunodeficiency virus (SIVmac239) expressing HIV-1 RT, in order to overcome drug resistance mimicry of the primate lentiviruses to NNRTIs [3]. Another model (SIV-based) has been developed for neurotropic infection, a condition often occurring in late-stage AIDS [4]. In this case, in order to by-pass the different response to antiretrovirals, the authors used a drug combination which is not adopted in humans. However, in both of these animal models, low-level viremia persisted and viral RNA was consistently detectable in anatomical sanctuaries [3], [4].
A model recently developed by our group is based on a polyclonal virus, such as SIVmac251, mimicking, at least in part, the genetic diversity of HIV-1 naturally inoculated in human subjects [5]. It was recently shown that SIVmac251 responds to combined ART consisting of two nucleosidic/nucleotidic reverse transcriptase inhibitors (NRTIs), i.e. tenofovir and emtricitabine, and the integrase inhibitor raltegravir [5], [6]. In this treatment model, the virus persists during ART, and viral load rebounds following treatment suspension in a time frame remarkably similar to that observed in humans after treatment interruption [7].
Recent research has added more credit to the macaque AIDS model, showing that, similarly to humans [8], [9], rhesus macaques (Macaca mulatta) harbour a central memory CD4+ T-cell reservoir, which plays a pivotal role in AIDS pathogenesis [7], [10]. Important insight has been derived from comparisons between rhesus macaques and sooty mangabays (Cercocebus atys) which, unlike M. mulatta, do not progress to AIDS [11]. M. mulatta, but not C. atys, shows up-regulation of the lentiviral co-receptor CCR5 in activated central memory T-cells, thus rendering this T-cell pool highly permissive to infection [10]. Conversely, the reduction of the long-lived memory T-cells (CD95+CD28+), including central memory T-cells, by the gold-based compound auranofin in intensified ART (iART)-treated rhesus macaques resulted in decreased levels of viral DNA and delayed progression of the infection upon therapy suspension [7]. Therefore, a model mimicking the effects of suppressive ART in humans is of fundamental importance also for the study of the dynamics of this viral reservoir.
One major limitation of current models for HIV persistence during therapy is their large discrepancy from conditions observed in humans. So far, due to financial and temporal constraints, animals have been chosen from homogeneous cohorts in terms of timing, type and route of the inocula, and have been treated in the early phases of chronic infection [3]–[6] or during acute infection [12]. Instead, at therapy initiation, HIV-infected humans are usually characterized by different timings and routes of disease acquisition and different levels of progression of the infection. In order to obtain a robust animal model for HIV persistence during therapy, the drug regimens should display similar efficacies as compared to those employed for human treatment, and reproducible control of heterogeneous viral loads in wide cohorts of subjects with different characteristics and previous treatment histories.
Here, we report a highly intensified ART (H-iART) regimen for the simian model, reproducibly capable of decreasing viral load to levels below assay detection limits in SIVmac251-infected macaques starting from a wide range of baseline viremic conditions, and overcoming previous treatment failures. We also report an unexpectedly impressive restriction of viral DNA in peripheral blood mononuclear cells, obtained by means of a pharmacological strategy entirely based on antiretroviral drugs.
CEMx174 and HTLV-I-transformed MT-4 cells were grown in RPMI-1640 medium supplemented with glutamine (200 µg/mL) (Invitrogen Life Technologies, Inc. Carlsbad, California), 10% heat-inactivated fetal bovine serum (FBS; Invitrogen Life Technologies), penicillin (500 U/mL; Pharmacia Italia SPA) and streptomycin (66.6 U/mL; Bristol-Myers, Sermoneta, LT).
Peripheral blood from uninfected nonhuman primates was diluted 1∶2 with PBS 1x-NaCl, and peripheral blood mononuclear cells (PBMCs) were Ficoll-separated, resuspended at a concentration of 2×106/mL and stimulated for 3 days with 5 µg/mL phytohaemoagglutinin (PHA) (Difco Laboratories, Detroit, MI, USA) and 100 units/mL of human recombinant IL-2 (Roche Diagnostics, Indianapolis, IN, USA).
CEMx174, MT-4 cells, and three-day old PBMCs were challenged with standard viral stock preparations for 2 hours in an incubator at 37°C with 5% CO2, washed and incubated with increasing drug concentrations (0.0001–1 µM), according to a previously published protocol [5]. The assays on virus entry inhibitors such as maraviroc (MRV), were conducted as in [13]. Briefly, the drug was first added during incubation with the virus and the same drug concentrations were then re-added upon cell washing. In MT-4 cells, through the MTT assay (MT4-MTT), we measured inhibition of the cytopathic effect of the two viruses. The assay was performed when the majority of control infected cells were dead. At different intervals post-infection, the viral core antigen p27 was measured in supernatants by antigen-capture ELISA assays (Advanced BioScience lab., Inc.).
The Indian rhesus macaques used in this study were housed at Bioqual, Inc., according to standards and guidelines as set forth in the Animal Welfare Act, the Guide for the Care and Use of Laboratory Animals, and the Association for the Assessment and Accreditation of Laboratory Animal Care (AAALAC), following approval by the Institutional Animal Care and Use Committee (IACUC). A total of eleven macaques have been enrolled for this study, while five previously enrolled macaques have been employed as historical controls.
For the pilot study, four SIVmac251-infected non-human primates (M. mulatta) that had been stably viremic at least for the last 3.3 months were put under a regimen (i.e. ART) consisting of tenofovir (PMPA), emtricitabine (FTC) and raltegravir [5], for 1.5 months. To improve control of viral load, this regimen was intensified by the addition of darunavir (DRV) boosted with ritonavir (/r) [intensified ART (iART)]. After 80 days, the treatment was further reinforced [highly intensified ART (H-iART)] by the addition of maraviroc.
For the second part of the study, eight additional SIVmac251 infected animals were used. These animals were divided into three treatment groups. One group (n = 2) was treated with MRV/r alone for three weeks, followed by addition of tenofovir/emtricitabine/raltegravir/DRV. A second group (n = 4) was treated with all H-iART drugs administered simultaneously. A third group (n = 2) was treated with iART to serve as controls.
For the combined antireservoir/antiretroviral therapeutic protocols, macaque P252, previously treated with iART plus the anti-reservoir drug auranofin (for detail, see Ref. 7), was put under a H-iART regimen for one month when viral load rebounded after suspension of the previous treatment. Another macaque, P177 of the pilot study, was treated (after the end of the follow-up aimed at monitoring the effects of H-iART alone) with auranofin in addition to H-iART. This macaque was then subjected, similarly to P252, to a further cycle of H-iART at viral load rebound.
More detailed information on the macaques enrolled, their viro-immunological background and the therapeutic regimens adopted for each animal can be found in Table S1.
All animals were dosed subcutaneously with tenofovir, and emtricitabine, and orally (with food) with raltegravir, DRV/r, and MRV. Initial drug dosages were: tenofovir, 30 mg/kg/day; emtricitabine, 50 mg/kg/day; raltegravir, 100 mg bid; DRV, 375 mg bid (for macaques starting from viral loads lower than 105 viral RNA copies/mL) or 700 mg bid (for macaques starting from viral loads higher than 105 viral RNA copies/mL); ritonavir 50 mg bid; MRV 100 mg bid.
Tenofovir and emtricitabine were kindly provided by Gilead Sciences (Foster City, CA). Raltegravir, DRV/r and MRV were purchased from the manufacturers.
For measurement of plasma SIVmac251 RNA levels, a quantitative TaqMan RNA reverse transcription-PCR (RT-PCR) assay (Applied Biosystems, Foster City, Calif.) was used, which targets a conserved region of the gag transcripts. The samples were then amplified according to a method previously validated in our hands [see ref 5 and Fig. S1]. The sensitivity of the method is two copies per run, which results in a detection limit as low as 40 RNA copies/mL in our routine analyses. Briefly, a 500-µL aliquot of plasma was spun down at 13,000× g for 1 h. The liquid was poured off and 1 mL of RNA-STAT 60 was added. After 5 min., 250 µL of chloroform was added and vortexed. The samples were spun at the same speed for 1 h. The clear aqueous layer on top was removed, and added to 500 µL of isopropanol. Then, 10 µl of 10 µg/mL tRNA was added and precipitated overnight at −20°C. The samples were spun for 1 hour, washed with a cold (−20°C) 75% ethanol solution, and re-spun for 1 h. The RNA was resuspended in 30 µL of RNAse-free water. 10% of the resuspended RNA was added to Taqman reagents (Applied Biosystems), plus primers and probe, and amplified in a 7700 Sequence Detection System by Applied Biosystems. Briefly, the sample was reverse transcribed at 48°C for 30 min. using One-Step RT-PCR Master Mix (Applied Biosystems), then held at 95°C for 10 min., and run for 40 cycles at 95°C for 15 sec. and 60°C for 1 min. The following PCR primer/probes were used: SIV2-U 5′ AGTATGGGCAGCAAATGAAT 3′ (forward primer), SIV2-D 5′ GGCACTATTGGAGCTAAGAC 3′ (reverse primer), SIV-P 6FAM-AGATTTGGATTAGCAGAAAGCCTGTTGGA-TAMRA (TaqMan probe). The signal was finally compared to a standard curve of known concentrations from 107 down to 1 copy (the linear range of concentration/signal relation spans eight Logs). All samples were done in triplicate for consistency and accuracy. In our increased sensitivity analyses, RNA was extracted from 6 mL of starting plasma, leading to a sensitivity threshold of 3 copies/mL. The inter-assay variability of the assay is 23.4%; The intra-assay variability is 20.6%.
For proviral DNA detection, cells were spun down to a pellet, and the supernatant was poured off. The cell pellet was lysed with 1 mL of DNASTAT for 10 min. 250 µL of chloroform was added and the mixture was vortexed. The samples were spun at 13,000 for 1 h. and the aqueous layer was removed and added to another tube. To this, 500 µL of isopropanol was added, and the mixture was precipitated overnight at −20°C. The samples were then spun for 1 h and the precipitate was washed with a −20°C-cold, 75% ethanol solution, and re-spun for 1 h. The DNA pellet was resuspended in 30 µL of water and 10% of the resulting solution was added to Taqman reagents (Applied Biosystems) plus primers and probe (the same as in previous paragraph) and amplified in a 7700 Sequence Detection System by Applied Biosystems. The signal was finally compared to a standard curve of known concentrations from 106 down to 1 copy (the linear range of concentration/signal relation spans seven Logs). The detection limit of this assay is two copies of proviral DNA/5×105 cells. The inter-assay variability is 28.3%; the intra-assay variability is 9.9%.
The presence of PCR inhibitors in both the quantitative assays (viral RNA and proviral DNA) has been ruled out by spiking the samples with known amounts of viral RNA and proviral DNA respectively (see Table 1 and Table S2).
Animals were bled before feeding in the morning, in order to obtain reliable measurements of trough drug levels. Plasma was obtained from supernatants of ficoll-centrifuged blood.
For DRV, sample preparation involved addition of an internal standard and liquid-liquid extraction with 2 mL tert-butylmethylether (tBME) at basic pH, and reconstitution in 100 µL of mobile phase to concentrate the sample. Reversed phase chromatographic separation of the drugs and internal standard was performed on a YMC, C8 analytical column under isocratic conditions. A binary mobile phase was used consisting of 55% 20 mM sodium acetate buffer (pH 4.88) and 45% acetonitrile. The UV detector set to monitor the 212 nm wavelength provided adequate sensitivity with minimal interference from endogenous matrix components. Calibration curves are linear over the range of 50 to 20,000 ng/mL. Inter- and intraday variability was less than 10%.
For MRV, a protein precipitation method using acetonitrile (AcN) containing internal standard (MVC-d6) was employed to extract the drug from macaques' plasma. An aliquot of the supernatant was further diluted with 0.5% tirfluoroacetic acid to maintain signal intensity within the linear range of the instrument. Reversed phase chromatographic separation was performed on an XBridge C18 analytical column under isocratic conditions. A binary mobile phase consisting of 0.1% formic acid in water and 0.1% formic acid in acetonitrile (72∶28) was used and provided adequate separation from other analytes. Detection and quantitation was achieved by multiple reaction monitoring (MRM), and MVC and internal standard were detected using the following transitions for protonated molecular products [M+H]+: m/z MVC 514.2>106.0; m/z MVC-d6 520.3>115.0. The assay has a dynamic range of 5 to 5,000 ng/mL using 20 µL plasma.
For both DRV and MRV total drug concentrations were measured (i.e free and protein bound).
Hematological analyses were performed by IDEXX (IDEXX Preclinical Research, North Grafton, MA). For calculation of absolute CD4+ and CD8+ T-cell numbers, whole blood was stained with anti-CD3-fluorescein isothiocyanate (FITC)/anti-CD4-phycoerythrin (PE)/anti-CD8-peridinin chlorophyll α protein (PerCP)/anti-CD28-allophycocyanin (APC), and anti-CD2-FITC/anti-CD20-PE, and red blood cells were lysed using lysing reagent (Beckman Coulter, Inc., Fullerton, Calif.). Samples were run on a FACSCanto II (BD Biosciences, San Jose, CA).
Staining for naïve (TN: CD28+CD95−), central and transitional memory (TTCM/TM: CD28+CD95+), and effector memory (TEM: CD28−CD95+) T-cells was performed on PBMCs isolated from total blood of three rhesus macaques treated with H-iART. For each animal, the blood was collected monthly from 0 to 4 months from the addition of MRV to the drug regimen. The cells (3×105 per sample) were surface stained by incubation with six appropriately titrated monoclonal antibodies (mAbs) for 20′ at 4°C, washed with PBS and resuspended in 1% paraformaldehyde in PBS. The following mAbs were used: anti-CD3 (APC-Cy7), anti-CD4 (Per-CP), anti-CD8 (Pe-Cy7), anti-CD20 (APC), anti-CD28 (FITC) and anti-CD95 (PE). Six-parameter flow-cytometric analysis was performed on a FACS Canto II instrument (BD Biosciences) [7]. Staining for HLA DR+ T-cells was performed with the same procedure described above, but with the substitution of an anti-HLA-DR antibody (APC, clone G46-6) to the aforementioned anti-CD20 antibody. The absolute numbers of naïve (CD95−CD28+), long-lived (CD95+CD28+) and short-lived (CD95+CD28−) memory CD4+ T-cells and the numbers of HLA-DR+ cells were deduced from percentage values of parent cells.
Specific immune responses were detected by measuring gamma interferon (IFN-γ) secretion of macaque PBMCs stimulated with a SIVmac239 Gag peptide (15-mer, obtained through the AIDS Research and Reference Reagent Program, National Institutes of Health [NIH], catalogue no. 6204, peptide 64) in an enzyme-linked immunospot (ELISPOT) assay. The assay was performed with the ELISpotPRO for monkey interferon-γ kit (Mabtech AB, Nacka Strand, Sweden) according to the manufacturer's instructions. Briefly, 1.5×105 Ficoll isolated macaque PBMCs were added to 96 well plates pre-coated with an anti-human/monkey IFN-γ antibody (MAb GZ-4). Cells were resuspended in RPMI 1640+10% FBS with 2 µg/mL of the peptide. After 48 hours incubation at 37°C with 5% CO2, the cells were rinsed from the plates, and a biotinylated anti-human/monkey IFN-γ antibody (MAb 7-B6-1; Mabtech) was added to the wells. The plates were then washed with PBS and incubated with the substrate solution (BCIP/NBT-plus). Spots were counted by using an automated reader (Immunospot Reader, CTL analyzers, LLC, Cleveland, OH). Numbers of spot-forming cells (SFC)/106 cells for each set of wells were averaged. A response was considered positive if the number of SFC/106 cells was at least four times the background value.
Data were analyzed using the software GraphPad Prism 5.00.288 (GraphPad Software, Inc., San Diego, CA). For calculation of the EC50 and EC90 values, data were transformed into percentage-of-inhibition values, plotted on x,y graphs, and subjected to linear or non-linear regression, depending on the best-fitting equation. Response to drugs in vivo was evaluated by repeated-measures ANOVA, followed by an appropriate post-test to analyze differences between time points. An appropriate transformation was employed to restore normality, where necessary. Logit analysis was adopted to investigate the influence of variables on binary outcomes, using an online calculator (http://statpages.org/logistic.html).
Trends in time were analyzed by regression analysis (GraphPad Prism), using the most appropriate equations. Akaike's information criteria (AICc) were used to select the model that was most likely to have generated the data and to compare the differences between equation parameters.
The inter-assay variability of quantitative real time PCR was estimated as an average of the coefficients of variation (CV) of matched measurements in two assays conducted on different occasions; the intra-assay variability was estimated as the coefficient of variation of multiple replicates (at least five) within the same assay.
Numerical simulations were performed with the ordinary differential equations solver ODEPACK of the Scilab 5.3.3 software (http://www.scilab.org/). The solver is based on finite difference methods for non-stiff problems, but it dynamically monitors the data in order to decide whether the stiffness of the problem requires a Backward Differentiation Formula method. The values of the discrete five-dimensional vector function of the solution were computed every 0.01 days. Details on mathematical modeling are given in Text S1.
The first part of this study was aimed at obtaining long-term viral suppression in a group of macaques (n = 4) in order to develop a suitable platform for testing experimental eradication strategies. We first analyzed the susceptibility of SIVmac251 to the protease inhibitor darunavir (DRV) and the CCR5 blocker maraviroc (MRV) in order to expand the arsenal of antiretroviral options available for the macaque AIDS model. DRV was chosen because of its well documented ability to inhibit several drug-resistant HIV-1 isolates as well as HIV-2, a virus closely related to SIVmac251 [1], [14], [15]. Moreover, the choice of this drug was supported by preliminary bioinformatic and molecular modeling analyses showing the potential interactions of DRV with the SIVmac251 protease [Text S2 and Fig. S2]. MRV, a CCR5 antagonist, was chosen on the basis of the important role of CCR5 as a SIVmac251 co-receptor [16] and due to the antilentiviral activity previously demonstrated by one experimental CCR5 blocker in macaques [17]. Moreover, retrospective analysis of one previous in-vivo experiment supported the anti-SIVmac251 effect of this drug [Text S3 and Fig. S3]. Results from tissue culture experiments showed that both DRV and MRV inhibited SIVmac251 replication in the nanomolar range, with EC50 values well below the trough concentrations detected in macaques treated with these drugs and described below in the text. (Fig. 1).
A group of macaques [n = 4] displaying signs of immune deterioration (eighteen months post-inoculation) was treated with a regimen of tenofovir, emtricitabine and raltegravir (Fig. 2). These macaques were derived from viral titration experiments and selected among those maintaining stable plasma viral loads (Fig. 2A). The selected animals displayed viral load set points between 103 and 105 viral RNA copies/mL. As our study was aimed at obtaining a model mimicking the conditions found in HIV-1-infected individuals under ART, such baseline values were chosen in order to reflect the average viral loads at which treatment is started in humans. The CD4 counts displayed by the macaques enrolled in this “pilot” study were significantly lower than values observed in uninfected controls (Fig. S4), suggesting that they were unlikely to be long-term non-progressors or élite controllers.
The three-drug regimen proved insufficient to maintain control of viral load in three of the four animals treated (Fig. 2A). DRV (375 mg bid), boosted with ritonavir (50 mg bid), henceforth referred to as DRV/r, was added to the treatment in an attempt to obtain a more stable control of viral load. DRV/r significantly improved control of viral load, inasmuch as viral RNA in plasma was maintained at a significantly lower level as compared to the pre-therapy values (Fig. 2A). No similarly decreasing trend of viral load was observed in an untreated control group of macaques [n = 2] showing non-significant differences in baseline viral loads as compared to the treatment group (two tailed t-test: P = 0.803; Fig. 2A). We conclude that the iART regimen adopted improves control of viral load in SIVmac251-infected macaques.
To increase the chances for long-term control of SIVmac251 replication, we explored the in-vivo efficacy of the CCR5 inhibitor MRV. This drug (100 mg BID) was eventually added to the drug cocktail in the aforementioned group of macaques (Fig. 2). After MRV was started, all macaques stably maintained viral loads below the limit of detection of the assay (i.e. 40 copies RNA/mL; Fig. 2A). There were also significant increases in the absolute numbers of CD4+ T-lymphocytes (Fig. 2B). Henceforth, this multidrug combination will be referred to as highly intensified ART (H-iART).
In order to further support the contribution of MRV to the antiretroviral effects observed, we treated two macaques with MRV (ritonavir boosted, MRV/r) in monotherapy (Fig. 3). In line with its CCR5-blocking ability, MRV decreased the viral loads in two drug-naïve macaques with dynamics similar to those previously shown by an investigational CCR5 blocker [17]. When the other H-iART drugs were added to MRV, a quick abatement of viral load to levels below the assay detection limit could be demonstrated (Fig. 3).
Prior to treatment with antiretrovirals, approximately one third of the experimental infections of macaques with SIVmac251 results in viral set points comparable to those displayed by the macaques described in the previous paragraphs (Fig. S5). To check whether H-iART might reproducibly control viral replication in SIVmac251 infected macaques characterized by higher viral loads, five animals with viral set points ranging from 103 to 107 viral RNA copies/mL of plasma were treated with H-iART, and the viral decay dynamics were compared with those of macaques treated with iART. Results clearly showed that H-iART induced a significantly more rapid decay in viral load than did iART (Fig. 4A). In line with the efficacy of H-iART, CD4+ T-cells increased in all study macaques (Fig. S6). We conclude that MRV-containing H-iART is superior to iART in abating viremia load in a group of macaques with a wide array of baseline viral loads.
We then analyzed the viral load decay dynamics in macaques treated with H-iART ab-initio. SIVmac251-infected macaques responded to administration of H-iART with a two phase exponential decay, as described in humans treated with suppressive ART [18] (Fig. 4). Similarly to the average treatment outcomes in humans [19], the level of viral load suppression depended on the baseline viral loads, with macaques starting from higher viral loads showing viral blips or residual, though markedly decreased (>3 Logs), viral replication (Fig. 4D–F).
We increased the DRV and MRV dosage in macaques 4887, BD64 and BD69, i.e. those starting from higher baseline viral loads (>105) and showing incomplete control of viral replication or major blips. Results showed that the improved drug regimen led to viral loads consistently below the assay detection limit in animals BD64 and BD69 (Fig. 4D,E). The increased drug dosage was also able to decrease the amplitude of the remaining sporadic blips (Fig. 4E). The resulting blips were lower than 103 copies of viral RNA/mL, thus mimicking those observed in humans under ART [20]. Nevertheless, one animal (4887) experienced a further viremic episode. Analysis of the cerebrospinal fluid (CSF) of this animal showed a viral load approximately one order of magnitude higher than that in plasma, while CSF samples were below the assay detection limit (i.e. 40 copies/mL) in the macaques showing stable control of viral replication (data not shown). This evidence suggested that the central nervous system (CNS) was a likely major source for the rebounding virus in macaque 4887. According to previously published studies: 1) virus levels in the CSF during the advanced stages of the disease are mostly due to CNS sources [21], and 2) the protease inhibitors (i.e., the only drug class in our cocktail acting at a post-translational level, and hence on chronically infected cells) are extruded from the CNS by P-glycoprotein (P-gp) molecules in the blood-brain barrier [22]. We thus intensified the P-gp blockade by increasing, from 50 to 100 mg bid, the dosage of ritonavir, which is a well-known P-gp inhibitor [23]. The viral load decreased in both plasma and CSF, with a more rapid decay kinetic in plasma, in which viral RNA eventually fell to levels below the assay detection limit (Fig. 4F). This result is in good agreement with the hypothesis of the CNS as a major source for the rebounding virus.
We conclude that macaques starting from high viral loads respond to H-iART similarly to HIV-infected humans and that viral loads can be abated to levels below the assay detection limit by adjusting the drug dosages and boosting procedures.
To check the presence of low-level viremia in SIVmac251-infected macaques under H-iART, we lowered the detection limit to 3 copies of viral RNA/mL and re-measured viral loads in some selected pooled serum samples. We found no evidence for low-level viral replication in plasma of all of the macaques tested (Table 1). Of note, viral RNA was below the assay detection limit in the plasma samples taken from macaque 4887 before its last viremic episode, supporting the hypothesis that H-iART was able to completely control viral replication in the periphery, despite the presence of a major CNS reservoir (Fig. 4F). Analyses conducted on lymph node biopsies (inguinal) showed that four out of six macaques analyzed had levels of cell-associated RNA below the limit of detection of the assay (i.e. 2 copies/5*105 cells/mL) (Table 2). The presence of cell-associated RNA in lymph nodes was independent of baseline viremia at treatment initiation (Logit analysis P = 0.801), thus supporting the idea that the suppressive efficacy of H-iART is not confined only to those macaques starting from moderate viral loads. In addition, cell associated RNA measured in samples taken from rectal biopsies was below the assay detection limit in all animals analyzed, supporting the idea of full suppression of peripheral viral replication (Table 2). This was rather surprising, because other antiretroviral regimens adopted in macaques proved unable to completely control viral RNA in anatomical sanctuaries [3], [24].
In the pilot study presented above, we unexpectedly found that H-iART profoundly impacted on viral DNA. First, there was a late viral DNA decay to levels below the assay detection limit which was associated with the addition of MRV to the drug cocktail (Fig. 5A). In addition, the CD4/CD8 ratio, the decrease of which is a marker of the viral reservoir and/or ongoing viral replication [7], [8], significantly increased during treatment (Fig. 5B). Of note, viral DNA in PBMCs also fell below the assay detection limit in all macaques included in the group treated with H-iART ab-initio (median treatment duration = 125 days, range from 45 to 174 days), i.e. no viral DNA copies were detectable in six out of six repeats with a threshold sensitivity of 2 copies/5*105 cells.
Moreover, we could not detect viral DNA in lymph node and rectal tissue biopsies (detection limit: 2 copies/5*105 cells, three repeats per sample) in all the macaques of the pilot study tested (Table 2). Lymph node viral DNA was also below the assay detection limit in one of three macaques from those treated with H-iART ab-initio, while viral DNA was below the limit of detection in rectal biopsies of all the macaques of the same group (Table 2). The results were further validated by excluding the presence of PCR inhibitors using spiked DNA for selected samples (Table S2).
The dynamics of the viral DNA decay during H-iART were studied in those animals to which all H-iART drugs were administered simultaneously and for which viral DNA measurements were available.
The levels of viral DNA in PBMCs during time were consistent with a three-phase decay, with the first two phases paralleling the two-phase decay of viremia, and a third, slower phase occurring after viremia had fallen to levels below the assay detection limit (Fig. 5C). This last phase of the viral decay has been ascribed to the latently infected T-cell numbers [18]. This result was noteworthy, because no such decreasing trends in viral DNA had been observed in animals treated with iART (i.e. without MRV) [7].
In line with the reportedly stimulating effect of the major CCR5 ligand RANTES on T-cell proliferation [25] some studies suggested that MRV, by acting as an antagonist of this cytokine, might alter the T-cell dynamics in vivo [26]. To study these phenomena, the CD4+ T-cell subpopulations were analyzed by six-color flow-cytometry at different time points following addition of MRV to the therapeutic regimen (Fig. 6). To avoid biasing the result with the possible effects of a detectable viral load on the T-cell subpopulations, these tests were conducted on PBMCs from macaques P157, P185 and P188 which already displayed a viral load below the assay detection limit when MRV was added (Fig. 2). Results showed that H-iART decreased the memory CD4+ T-cell numbers over time (Fig. 6A,B), while it carried out no significant effect on the naïve T-cell subpopulation (Fig. 6C). This result is in accordance with the in-vitro inhibitory effect of MRV on the proliferation of sorted memory T-cell subpopulations (Fig. S7). MRV significantly decreased the numbers of activated (HLA-DR+) CD4+ TEM cells (Fig. 6D). This effect is in line with decreased levels of immune activation already observed in humans treated with this drug [26], [27]. In conclusion, MRV decreased the number of memory T-cells as well as TEM cell-activation. Since these two parameters are linked to the magnitude of the viral reservoir and ongoing viral replication [9], [28], this effect is in good agreement with the aforementioned three-phase decay of viral DNA induced by MRV (Fig. 5C).
The results so far obtained were in line with a recently issued report which suggested that MRV decreased the magnitude of the viral reservoir in HIV-1-infected individuals [26]. This study, which was unable to provide conclusive evidence, did not show an impact of MRV on the viral set point following therapy suspension, a parameter stringently associated with the extent of the viral reservoir [7], [29], [30]. To test this hypothesis, we analyzed the difference in the pre and post-therapy viral set points in those macaques from our cohort that had received MRV and that had undergone therapy suspension (for treatment details see Figs. 2, 4 and Text S3). Results show that treatment with MRV is associated with a reduction of the viral set point post-therapy (Fig. 7A), and that the extent in the viral set point decrease depends on the total exposure to the drug (Fig. 7B). These results are suggestive of an independent effect of MRV on the viral set point following therapy suspension and add credit to the hypothesis that MRV may contribute to an anti-reservoir effect of H-iART.
Finally, given the aforementioned effects of H-iART, we tested whether this therapeutic regimen might be adopted to improve the effect of a previous anti-reservoir strategy based on the anti-memory drug auranofin in combination with antiretrovirals [7]. Upon interruption of this anti-reservoir treatment, SIVmac251-infected macaques experience an acute infection-like condition, i.e. an initial viral load peak followed by rapid containment of viral load [7]. The peak, which is rapidly reached upon virus re-appearance in plasma, is associated with the reconstitution of the viral reservoir, as shown by the previously published independent association between the area under the curve (AUC) describing the initial peak of viral load and the eventual viral load set point ([7] see also Fig. 8A). From this association, it follows that decreasing the AUC at peak artificially through a cycle of H-iART should limit the reconstitution of the viral reservoir and may result in spontaneous control of viral load following H-iART suspension.
The experiment was attempted in two macaques. A first macaque (P252) was treated with a one-month cycle of H-iART at viral load rebound, after the suspension of the aforementioned auranofin/antiretroviral regimen. Another macaque (P177) was treated with auranofin in addition to H-iART as a follow-up of the treatment presented in the pilot study. Eventually, following therapy suspension, P177 was subjected to a short H-iART cycle at viral rebound, similar to that administered to P252. In both cases, the short H-iART cycle promptly abated viral load to levels below the assay detection limit, thus efficiently decreasing the initial AUC (Fig. 8 A–C).
The macaques showed exceptionally low viral set points after the short cycle of H-iART was suspended, in line with the expected values calculated on the basis of our AUC/viral set point correlation curve (Fig. 8A). Both macaques periodically displayed viral load peaks that subsequently decreased to low-level viremia (<500 copies of viral RNA/mL) or to levels below the assay detection limits. The CD4 slope was non-significant during the follow-up period (P = 0.7079 for P252 and P = 0.2319 for P177; Fig. 8D,E), in line with the previous observation that the CD4 slope following therapy suspension identifies the impact of a treatment on the viral reservoir [7]. Conversely, CD4 counts had shown significantly decreasing trends in both macaques before all treatments were started (P<0.0001 for P252 and P = 0.0039 for P177; Fig. 8D,E), thus supporting the concept that the therapies adopted significantly impacted on the natural course of the disease.
Consistently with its exceptional reduction of the AUC at peak, macaque P177 showed a remarkable degree of spontaneous control of viral load during six months of follow-up, which was not yet considerable as, but seemingly close to a drug-free remission of the disease (Fig. 8C). In this macaque, viral load was maintained at levels below the assay detection limit during the periods between peaks (detection limit: 40 RNA copies/mL) and, when the RNA detection limit was further lowered to 3 copies/mL, no evidence of residual viremia was found (see Table 1). This control of viral replication could hardly be ascribed to cell-mediated responses, in that a moderate increase in the number of IFN-γ positive spots could be detected only at viral rebound but not during the viral set point (Fig. S8), thus suggesting that H-iART induced a true containment of the viral reservoir reconstitution, similarly to other experimental strategies restricting the formation of the viral reservoir during acute infection [29], [30], [31]. We conclude that a short course of H-iART, in line with the highly suppressive effect of this therapeutic regimen on SIVmac251, may prevent the viral reservoir reconstitution following suspension of a previous anti-reservoir therapy and result in a drug-free spontaneous control of viral load.
Some investigators recently questioned the robustness of primate models, citing the difficulty of obtaining, with the cross-active drug options available, full viral suppression in sanctuaries and viral loads below the assay detection limits for prolonged periods [32], [33]. The results reported in the present article do not support this argument.
1) Since a good animal model should mirror full viral suppression in humans, we checked viral loads in plasma for prolonged periods and analyzed the presence of viral nucleic acids in anatomical sanctuaries. The level of abatement of viral nucleic acids that we found in the present study in peripheral blood and anatomical sanctuaries of the majority of the macaques tested provide the maximum degree of viral suppression so far observed in antiretroviral treated primates. The level of reproducibility of these results is shown by the fact that they were obtained in a heterogeneous group of macaques, likely mirroring a wide number of possible disease conditions in humans. This is the first report, to our knowledge, of a therapy capable of stably controlling viral replication to levels below the assay detection limits also in macaques in the advanced stage of the disease, since the studies so far published have been able to report control of SIV replication only during acute infection [12] or in the early chronic phase of the disease [3]–[6]. Apart from mimicking the clinical conditions of a significant portion of HIV-infected individuals who are diagnosed in the chronic or pre-AIDS stages of the disease, this ‘late’ treatment allows excluding those macaques able to spontaneously control the infection, a phenomenon which usually occurs soon after the acute infection phase [34]. For the macaques enrolled in this study, the average plasma viral load at the time of therapy initiation was of 4.8±1.1 Log10 RNA copies/mL (mean ± SD). This value is lower than those reported in some articles during chronic SIVmac infection of macaques [35], [36], but similar to those published in other articles [37], [38]. As in this study we have not included macaques with viral loads during chronic infection higher than 6.8 Log10 RNA copies/mL or with the rapid progressor phenotype, the effect of our H-iART regimen on this more aggressive course of SIV infections remain to be ascertained.
Of note, persistence of the virus at low level in the lymph nodes of a minority of H-iART treated macaques provides another similarity of our macaque model with clinical conditions observed in humans infected with HIV-1, as this anatomical sanctuary has recently been shown to be a major site for ongoing viral replication in humans [39]. Studies of drug penetration in this anatomical compartment will be necessary to overcome this limitation in both macaques and humans.
2) As in any well respected science, the results are in good agreement with mathematical models (Fig. 9), and are mathematically predictable (as an example, see Fig. 8A). In this regard, important insight into the necessity for a multidrug regimen to control viral loads in macaques can be derived from a mathematical model developed by Rong and Perelson [40] and based on experimental observations [8]. This model suggests that a superior drug efficacy is required in simian AIDS to control viral replication (Fig. 9) because of the viral burst size, (i.e. the average number of virions produced by a single productively infected cell in a day). The viral burst size was shown to be higher in SIV infection as compared to HIV-1 infection [41], where a lower drug efficacy is expected to be sufficient to maintain viral control (Fig. 9A–C). Also a drug acting on the proliferation rate of activated T-cells, such as MRV (which antagonizes the proliferative effect of RANTES, see ref 25 and Fig. S7), appears to be important for containment of the viral blips (Fig. 9D). These simulations also show that the decreased proliferation rates may impact on the viral reservoir size (half-life: ≈200 days, see Text S4, S3, S2, S1 and Fig. 9D), which shows a half-life of the same order of magnitude as that calculated by analyzing the dynamics of the viral DNA decay during H-iART (Fig. 5).
3) According to the idea that a good animal model should represent a vanguard for future treatments to be tested in humans, our quest for increased drug efficacy in the macaque AIDS model allowed identifying unexpected benefits of H-iART on the immune system. Apart from the possible impact of H-iART on the viral reservoir (a concept supported by recent data in humans [42]), reduction by MRV of the memory T-cell subpopulation may restrict one major source for viral spread and ongoing viral replication. A decrease in the memory T-cell size is a logical expectation of the anti-proliferative effect exerted by MRV through CCR5 inhibition (Fig. S7), as antigen-driven proliferation contributes to maintenance of the size of this T-cell subpopulation [8]. It is well known that memory T-cells are a preferential target of HIV-1 replication [43], and that their decrease may affect the overall viral dynamics in vivo. In this regard, the MRV-induced decrease in the memory T-cell size is not only unlikely to be dangerous but, rather, likely to be beneficial. This hypothesis is supported by results showing that the pool of TCM cells is a correlate of anergy towards the viral antigens in Macaca mulatta but not in Cercocebus atys, which is naturally resistant to CD4+ T-cell loss and full-blown AIDS [44]. In addition, the results obtained with the present macaque model suggest that a short cycle of H-iART could be used for improving the efficacy of our previous anti-reservoir treatment based on auranofin and strengthen the idea that an arrest in disease progression may be obtained during the chronic phase of the disease. Although the data on the combined effect of the two subsequent treatment cycles are derived from a limited number of macaques, the result obtained is corroborated by the fact that no similar trend was observed in the same animals prior to starting therapy [5], [7] or in historical controls that had not received H-iART at rebound [7]. Of note, although certain major histocompatibility complex (MHC) class I alleles, including Mamu-A*01 and Mamu-B17* are associated with slow disease progression in SIV infected macaques [45], [46], independently, the presence of these alleles is not predictive for disease outcome [47], and none of our macaques presented the protective alleles in association (Table S1). Instead, P177, which, following our therapies, remarkably controlled viral load, presented the HLA Mamu-B*01 allele, that is associated with aggressive simian lentivirus infection [48]. In line with this genotype, P177 showed a significant immune deterioration before our treatments were initiated (Fig. 8C).
Finally, recent analyses [reviewed in 49] re-evaluated the necessity of wide numbers of subjects as a support for breakthrough findings, such as, in this case, the obtainment of a condition close to a persistent suppression of viremia in the absence of ART.
If the results of the present study should prove reproducible in humans, H-iART could represent a useful tool for improving the viro-immunological conditions of HIV-infected individuals and a useful addition to experimental anti-reservoir strategies.
|
10.1371/journal.pntd.0004951 | Differential Gel Electrophoresis (DIGE) Evaluation of Naphthoimidazoles Mode of Action: A Study in Trypanosoma cruzi Bloodstream Trypomastigotes | The obligate intracellular protozoan Trypanosoma cruzi is the causative agent of Chagas disease, a neglected illness affecting millions of people in Latin America that recently entered non-endemic countries through immigration, as a consequence of globalization. The chemotherapy for this disease is based mainly on benznidazole and nifurtimox, which are very efficient nitroderivatives against the acute stage but present limited efficacy during the chronic phase. Our group has been studying the trypanocidal effects of naturally occurring quinones and their derivatives, and naphthoimidazoles derived from β-lapachone N1, N2 and N3 were the most active. To assess the molecular mechanisms of action of these compounds, we applied proteomic techniques to analyze treated bloodstream trypomastigotes, which are the clinically relevant stage of the parasite.
The approach consisted of quantification by 2D-DIGE followed by MALDI-TOF/TOF protein identification. A total of 61 differentially abundant protein spots were detected when comparing the control with each N1, N2 or N3 treatment, for 34 identified spots. Among the differentially abundant proteins were activated protein kinase C receptor, tubulin isoforms, asparagine synthetase, arginine kinase, elongation factor 2, enolase, guanine deaminase, heat shock proteins, hypothetical proteins, paraflagellar rod components, RAB GDP dissociation inhibitor, succinyl-CoA ligase, ATP synthase subunit B and methionine sulfoxide reductase.
Our results point to different modes of action for N1, N2 and N3, which indicate a great variety of metabolic pathways involved and allow for novel perspectives on the development of trypanocidal agents.
| Trypanosoma cruzi is the etiological agent of Chagas disease, an important illness for Latin American countries that is now afflicting other continents due to the immigration of infected people. The available chemotherapy is limited to the chronic phase of the disease, being the development of novel active compounds essential, and the search for specific molecular targets for drugs in T. cruzi is necessary. In this context, our group has synthesized and screened many compounds ranging from natural to semi-synthetic naphthoquinones and derivatives on T. cruzi, displaying naphthoimidazoles N1, N2 and N3 the highest activity. Previous studies correlated phenotypic alterations by cell biology techniques as well as investigated mode of action by proteomic approaches in insect stage epimastigotes as a model. However, T. cruzi presents three morphologically distinct life stages with their own specific biological peculiarities and requirements that could be potential targets to drug intervention. Here, we evaluated the mechanism of action of N1, N2 and N3 in clinical relevant form of the parasite, bloodstream trypomastigotes, by proteomics. Our data pointed to 61 differentially abundant protein spots, being these proteins involved with cellular trafficking, protein synthesis, transduction signaling and energetic metabolism, among others, open interesting perspectives for trypanocidal strategies.
| Trypanosoma cruzi is an obligate intracellular protozoan and the causative agent of Chagas disease, a neglected illness that affects millions of people in Latin America that has recently been found in non-endemic countries because of immigration related to globalization [1]. Currently, the transmission of this disease primarily depends on the ingestion of food contaminated with the feces of sucking Triatominae insects, although the classical transmission route through the vector still occurs in endemic areas [2,3]. Other routes such as blood transfusion, organ transplantation and congenital transmission can also occur [4]. This illness presents two phases (acute and chronic) that have distinct characteristics. During the acute phase, pathogenesis is associated with high parasitemia [5,6]; however, the chronic phase is divided into indeterminate and symptomatic forms, which present digestive symptoms and/or cardiomyopathy, the primary clinical manifestations [7].
The T. cruzi biological cycle involves vertebrate and invertebrate hosts and different parasite forms [8]. The infection of the mammalian host is triggered by the entry of metacyclic trypomastigotes, which invade cells and differentiate into replicative amastigotes. After the intracellular proliferation of amastigotes, they differentiate into trypomastigote, and these parasites then reach the bloodstream to infect new cells and tissues. The infection of triatomine bugs occurs during insect foraging, through the ingestion of trypomastigotes. In the insect midgut, trypomastigotes are differentiated into proliferative epimastigotes, which colonize the vector. In the triatomine´s posterior rectum, a novel differentiation occurs to form metacyclic trypomastigotes, which will then be eliminated with the insect feces, completing the life cycle when the parasite reaches the vertebrate bloodstream again [9].
At present (2016), the nitroheterocyclic agents benznidazole and nifurtimox are the only commercial drugs available for Chagas disease chemotherapy. These compounds are very efficient against acute cases, but their severe side effects and limited efficacy make their use controversial for the chronic phase. Research on the discovery of novel molecular drug targets in the parasite is justified by the high number of chronic patients without an effective treatment [10]. The preclinical active azoles posaconazole and a ravuconazole derivative named E1224 are now in clinical trials, although they have presented a high percentage of treatment failures in chronic patients, indicating that the search for alternative compounds must be continued [11].
In searching for alternative Chagas disease chemotherapies, our group has been working on the trypanocidal effect of naturally occurring quinones, especially naphthoquinones and their derivatives, for the last 15 years. Among all the screened compounds, three naphthoimidazoles derived from β-lapachone N1, N2 and N3 (Fig 1) were the most promising [12–16]. The mechanisms of action of N1, N2 and N3 were previously assessed by cell biology techniques, and they exhibited cell cycle blockage, the inhibition of succinate cytochrome c reductase activity in epimastigotes as well as ultrastructural evidence of mitochondrial swelling, the abnormal condensation of nuclear chromatin, kinetoplast disruption and plasma membrane blebbing in bloodstream trypomastigotes. DNA fragmentation was also detected by flow cytometry and electrophoresis for the latter form of the parasite [13,14]. A cell death analysis strongly indicated autophagy as part of the naphthoimidazole mode of action, given the increase in monodansyl cadaverine labeling, the inhibition of the death process by the autophagic inhibitors wortmannin or 3-methyladenine, the overexpression of ATG genes and ultrastructural evidence [17].
Proteomics could play a crucial role in identifying potential drug targets because of the detection of metabolic shifts related to the pathogenesis of a great variety of diseases [18,19]. In trypanosomatids including T. cruzi, the open reading frames are organized into large polycistronic clusters that lead to post-transcriptional gene expression regulation. This molecular peculiarity supports the use of proteomic techniques as valuable technical alternatives, especially in these protozoa [20,21]. After the first description of the T. cruzi proteomic profile in 2004 [22], different high-throughput proteomic studies were performed with all stages of the parasite, and distinct strains were used to identify a maximum number of proteins from this organism [19,21,23–26].
In the chemotherapy field, only two proteomic analyses of the drug mechanism of action were performed, both of which were performed in epimastigotes. In 2008, T. cruzi's resistance to benznidazole was assessed by using a two-dimensional gel electrophoresis (2DE) approach, which showed that the modulation of thirty-six proteins was involved in distinct metabolic pathways in resistant epimastigotes [23]. The second proteomic evaluation of trypanocidal drugs was performed by our group, and we analyzed the mechanisms of action of naphthoimidazoles N1, N2 and N3 in the T. cruzi insect stage. The most remarkable modulation was detected in mitochondrial proteins, reinforcing the electron microscopy studies [13,14]. The modulated proteins are involved in several pathways such as reactive oxygen species detoxification, protein metabolism, and structural proteins such as tubulin, among others [19]. Recently, our group assessed the proteomic map of bloodstream trypomastigotes by using a shotgun approach. T. cruzi protein entries (5,901 of them) were described in almost all the cellular compartments and metabolic pathways of the parasite, with 2,202 protein entries exclusively detected in the bloodstream forms in comparison with the culture-derived and metacyclic trypomastigote proteomic profiles reported in the literature [27], which justifies the use of bloodstream trypomastigotes as a model for pathogenesis and chemotherapy studies. In the present work, we further evaluated the mechanisms of action of the naphthoimidazoles in bloodstream trypomastigotes, which are the clinically relevant forms of the parasite, by differential gel electrophoresis (DIGE).
The naphthoimidazoles were obtained from a reaction of β-lapachone with aromatic aldehydes in the presence of ammonium acetate and acetic acid, leading to 4,5-dihydro-6,6-dimethyl-6H-2-(phenyl)-pyran[b-4,3]naphth[1,2-d]imidazole) (N1), 4,5-dihydro-6,6-dimethyl-6H-2-(3´-indolyl)-pyran[b-4,3]naphth[1,2-d]imidazole (N2) and 4,5-dihydro-6,6-dimethyl-6H-2-(4´-methylphenyl)-pyran[b-4,3]naphth[1,2-d]imidazole) (N3), as previously described [12,15,16].
T. cruzi bloodstream trypomastigotes (the Y strain) were obtained by performing a heart puncture on infected albino Swiss mice (Mus musculus) at peak parasitemia (7th day) as previously described [27]. Six independent infections and heart punctures were performed in different days to generate six distinct pool of parasites which were considered distinct biological replicates. The parasites (5 x 106 cells/mL) were treated with the three naphthoimidazoles in RPMI for 24 h at 37°C at less than a half of a dose of inhibitory concentration that led to the lysis of 50% of the trypomastigotes (IC50/24 h) as previously determined (10 μM of N1 and 5 μM of N2 and N3) [13,14].
In this study, the infected mice euthanized for the trypomastigotes purification in strict accordance with the recommendations of the Guide for the Care and Use of Laboratory Animals of the Brazilian National Council of Animal Experimentation (COBEA). The protocol was approved by the Committee on the Ethics of Animal Experiments of the Fundação Oswaldo Cruz (CEUA-FIOCRUZ, License Number: LW16/13).
After the treatment, the parasites were washed three times with phosphate-buffered saline (PBS, pH 7.4) and then incubated in sample lysis solution (7 M urea, 2 M thiourea, 4% CHAPS, 40 mM Tris, and 60 mM dithiothreitol) containing Complete Mini protease inhibitor cocktail (Roche Applied Science, Indianapolis, USA). Subsequently, 10 freezing-thawing cycles were performed and the parasite homogenate was centrifuged to separate only the soluble protein fraction as previously described [19]. The protein concentration was determined using 2D Quant kit (GE Healthcare, Buckinghamshire, England). For 2D-DIGE analysis, six independent extractions (which were considered distinct biological replicates) of the four sample types [1 control and treatments with the three naphthoimidazoles (N1, N2 and N3)] were performed.
In each 2D-DIGE gel, 150 μg of protein was applied, being 50 μg of control sample, 50 μg of one of the treatments from the same extraction and 50 μg of an internal standard [a pool of trypomastigotes that was made by mixing equal amounts of protein from each sample from all the extractions (45 μg of the 24 samples)]. This experimental design (S1 Table) was used to prevent the possible impairment of gel image overlays among all the samples in the DeCyder software. The sample quantity for each treatment was limited and the control sample and internal standard were present in all gels, and then the experiments were performed once as described above.
The first two sample types were alternately labeled with 400 pmol of Cy3 and Cy5, and the internal standard for all of the gels was labeled with Cy2, according to the manufacturer's protocol (GE Healthcare, Piscataway, NJ, USA). Labeled samples were applied to Immobiline DryStrips (IPG 18 cm pH 4–7) (GE Healthcare, Piscataway, NJ, USA) using an in-gel rehydration method [28]. In summary, the rehydration step was 30 V for 12h followed by voltage increment to 200 V for 1h, 500 V for 1h, 1,000 V for 1h and from 1,000 to 8,000 V in 30 min. Finally, the isoelectric focusing step was set to reach a total of 64,000 Vh at 8,000 V [29].The electric conditions for isoelectric focusing (IEF) were set to a total of 64,000 Vh at 8,000 volts. After the IEF, each strip was incubated for 15 min in 50 mM Tris-HCl pH 8.8, 6 M urea, 30% (v/v) glycerol, 2% SDS and 0.002% bromophenol blue (BPB) containing 100 mg of DTT, followed by a second 15 min incubation step with the same buffer that contained 400 mg of iodoacetamide instead of DTT. The strips were positioned on the top of a 12% T SDS-PAGE gel and overlaid with 0.5% agarose in Tris-glycine electrode buffer and 0.002% BPB. The gels were run at 2.5 W/gel for 30 min and then for a total of 100 W until the dye front reached the bottom of the gel. The first and second dimensions of gel electrophoresis were performed using IPGPhor and Ettan DaltSix systems (GE Healthcare, Piscataway, NJ, USA), respectively. The gels were then scanned with Typhoon Trio (GE Healthcare, Piscataway, NJ, USA) at a resolution of 100 μm. The photomultiplier values were adjusted to optimize sensitivity and avoid oversaturation. The resulting digitalized images were analyzed using DeCyder 5.0 software (GE HealthCare, Piscataway, NJ, USA). A differential expression analysis was performed by comparing matched spots between the treated and control groups. A paired Student´s t-test was used for a statistical analysis of the differences (p ≤ 0.01). To identify the protein spots, electrophoresis was performed on a preparative 2D-PAGE containing 400 μg of protein from the pool of all the trypomastigote samples from this study (which made up the internal standard), and the gel was stained using colloidal Coomassie blue G-250. An image overlay was performed with Image Master 2D Elite 4.01 software (GE Healthcare, Piscataway, NJ, USA).
The Coomassie-stained gel spots were excised, destained with 25 mM ammonium bicarbonate pH 8.0 and 50% (v/v) acetonitrile (ACN), dehydrated with ACN and rehydrated in 15 μL of trypsin solution (20 ng/μL) for 45 min on ice (Promega, San Luis Obispo, CA, USA). A digestion was performed for 16–24 h, at 37°C. Tryptic peptides were transferred to clean tubes, and the remaining gel pieces were subjected to peptide extraction through 2 cycles involving the addition of 30 μL of 5% (v/v) formic acid/50% (v/v) ACN solution with vigorous vortexing for 20 s, resting for 15 min at room temperature, ultrasound (Ultra Cleaner 1400, Unique, Indaiatuba, SP, Brazil) for 2 min and further vortexing for 20 s. The final 80 μL was concentrated by vacuum centrifugation to approximately 10 μL and stored at -20°C until use [30]. Prior to mass spectrometry analysis, the tryptic peptides were desalted and concentrated with Zip-Tip C18 (Millipore, Billerica, MA, USA) according to the manufacturer's protocol. Peptides were eluted in 1.5 μL of 0.1% trifluoroacetic acid (TFA)/ 50% acetonitrile.
The eluted peptides were mixed with an equal volume (0.3 μL) of matrix solution [10 mg/mL α-cyano-4-hydroxycinnamic acid (Sigma-Aldrich, St. Louis, MO, USA) in 50% acetonitrile, and 0.3% TFA] for analysis with a MALDI TOF/TOF 5800 Proteomics Analyzer (Applied Biosystems, Foster City, CA, USA). The instrument was operated in positive-ion delayed-extraction-reflector mode. The peptides were ionized/desorbed with 2,000 total shots per spectrum, and the spectra were acquired at a 1.77 keV accelerating potential. An external calibration was performed with the following mixture of standard peptides: Arg-bradykinin (m/z 904.468), angiotensin I (m/z 1296.685), Glu1-fibrinopeptide B (m/z 1570.677), ACTH (1–17) (m/z 2093.087) and ACTH (18–39) (m/z 2465.199) (Applied Biosystems, Foster City, CA, USA). The calibration provided a mass accuracy of 50 ppm across the mass range from 800 to 3,500 Da. Up to 10 of the most intense ion signals with signal-to-noise ratios above 30 were subjected to fragmentation, excluding three of the most common trypsin autolysis peaks (m/z 1045.560, m/z 2211.100, and m/z 842.500). PSD spectra were acquired using 2000 laser shots and 2.01 keV of collision energy. MS/MS spectra were calibrated using the fragment ion mass spectra of Glu1-fibrinopeptide B. All MS/MS data were analyzed using Mascot (Matrix Science, London, UK; version 2.4.1) with the following parameters: 0.40 Da of fragment ion mass tolerance and 50 ppm of parent ion tolerance; the deamidated asparagine and glutamine, oxidated methionine, carbamidomethylated cysteine and propionamide-cysteine were set as variable modifications. A search was performed against a database containing 354,024 entries [NCBInr Kinetoplastida (downloaded at May 23rd 2014), Mus musculus (downloaded at June 16th 2014) and a common contaminants database] assuming the use of trypsin as the digestion enzyme. After the search, the data were statistically validated using Scaffold 4.4.1.1 software (Proteome Software, Portland, OR, USA) [31–33]. Protein and peptides were considered identified when their Peptide Prophet-calculated probabilities were greater than 95%. Proteins without proteotypic peptide identification were grouped to satisfy the principle of maximum parsimony [34].
The identified proteins were individually categorized by subcellular localization according to the information collected from the Uniprot database (http://www.uniprot.org/) and Basic Local Alignment Search Tool of NCBI (http://blast.ncbi.nlm.nih.gov).
The treatment of bloodstream trypomastigotes with naphthoimidazoles yielded a total of 61 differentially abundant protein spots by using 2D-DIGE (Figs 2 and 3). Comparisons between the control group and N1, N2 or N3 treatments yielded 44, 16 and 9 differentially abundant protein spots, respectively (Fig 3A and 3B). The differentially abundant protein spots were not exclusive of one comparison, e.g., there are 3 common spots that are simultaneously differentially abundant in N1- and N3-treated parasites (spots 614, 939 and 1453) and 5 spots simultaneously found in N1 and N2 treatments (spots 567, 1084, 1294, 1480 and 1488) as presented in Fig 2. A total of 36, 11 and 6 spots were only found to be differentially abundant in parasites that were treated with N1, N2 and N3, respectively.
After the MALDI TOF/TOF analysis, 34 protein spots were identified by 27 distinct protein entries (Table 1 and S2 Table). Of the total, only 36 spots were found to be differentially abundant following N1 treatment, but only 21 were identified. Of these identified spots, 15 protein spots were found to be more abundant in treated samples, and 6 spots were found to be less abundant. Among the more abundant spots, the following were identified: stress response proteins [a heat shock protein at 60 kDa (spots 665, 666, and 674), a heat shock protein at 70 kDa (525) and glucose-related protein 78 (spot 461)], energy metabolism regulation proteins [vacuolar ATP synthase subunit B (spot 787), asparagine synthetase A (spot 1070) and enolase (spot 949)], nucleic acid metabolism [guanine deaminase (spot 942)], protein transport [RAB GDP dissociation inhibitor alpha (spot 941)], cell motility and structural proteins [paraflagellar rod protein (spot 473) and major paraflagellar rod protein (spot 477)], proteins involved in protein folding [chaperonin (spot 607) and hypothetical protein (spot 781). The structural protein gelsolin (spot 943) from Mus musculus was found to be more abundant with N1 treatment. Among the less abundant protein spots that were only detected after N1 treatment, a hypothetical protein entry (spot 1612), alpha tubulin (spot 1624), succinyl-CoA ligase (spots 1278, 1393), heat shock protein at 70 kDa (spot 1075) and peptide methionine sulfoxide reductase (spot 1674) were found. In our results, the heat shock protein at 70 kDa was identified in spots showing higher (spot 525) and lower (spot 1075) abundance after N1 treatment.
For N2 treatment, of the 11 spots that were found to be differentially abundant with this treatment, 7 were identified. Actin entries from Mus musculus were to be found more (spot 970) and less (spot 1218) abundant, and arginine kinase (spot 1440), beta tubulin (spot 1235), heat shock protein at 70 kDa (spots 1409 and 1414) and activated protein kinase C receptor (spot 1264) were identified as less abundant. Distinct spots, namely spot number 1624 from the N1 treatment analysis and spot number 1235 from the N2 treatment were identified as tubulin, and both were found to be less abundant in comparison with the control sample.
Evaluating the spots simultaneously uncovered those that were differentially abundant in response to N1 and N2 treatments; a hypothetical protein (spot 567) was identified as being more abundant in both treatments, and the 85 kDa protein (spots 1084 and 1294) and the heat shock protein of 70 kDa (spots 1480 and 1488) were identified as less abundant. Of the 6 spots that were found to be differentially abundant with N3 treatment, 1 was identified as elongation factor 2 (spot 264), and it was more abundant with this treatment.
The subcellular localization of the 24 T. cruzi protein entries from trypanosomes is shown in Fig 4. A great number of mitochondrial proteins was identified (29%), followed by cytosolic entries (21%). Cytoskeletal, flagellar and plasma membrane proteins represent 8% each.
Many natural quinones that have been isolated from plant resources can participate in multiple biological oxidative processes because of their structural properties, given that their biological function is associated with their redox potential [35]. Quinoidal compounds are also sources of heterocycles, and there are few studies about the synthesis of quinone derivatives based on the reactivity of 1,2-quinoidal carbonyls towards nucleophilic reagents. Several classes of compounds were synthesized from β-lapachone, and among them, the naphthoimidazoles N1, N2 and N3 presented the highest activity against T. cruzi [12,15,16] in terms of efficacy against the three forms of the parasite [13,14].
As mentioned before, proteomics has only been employed twice in Chagas disease chemotherapy studies; for resistance/susceptibility analyses of benznidazole [23], and in a study about the mode of action of naphthoimidazoles N1, N2 and N3 [19]. In both cases, epimastigotes were used as a model, especially because of the axenic proliferation of the insect form of the parasite. The present analysis is the first proteomic study of trypanocidal drugs on T. cruzi trypomastigotes that were purified from the mammalian bloodstream, which is the most clinically relevant form. Two previous proteomic analyses of compounds against T. cruzi epimastigotes employed a two-dimensional gel approach (Coomassie blue staining) followed by MALDI-TOF/TOF identification [19,23], making it possible to obtain large amounts of samples crucial to completing all the performed analyses. However, the sample quantity represents the most important limitation during the proteomic evaluation of non-proliferative bloodstream trypomastigotes, and the need for a great number of animals makes this study very difficult to do.
Mass spectrometry-based proteomics is widely used for quantitative studies. The two main peptide-centered approaches are the label free techniques and mass-difference and isobaric tagging. The first one measure the abundance of proteins in their native state based on peptide ion intensities or spectral counts, but lack the throughput of labelling methods and the traceability of variations through analysis [36,37]. The isotopic (binary or tertiary comparisons, e.g.: ICAT, SILAC) and isobaric (multiplex analysis, e.g.: iTRAQ, TMT) labelling methods introduce respectively increased complexity in MS acquisitions and isobaric interference creating chimeric MS/MS spectra, both leading to decrease in accuracy of quantitation and limiting the sampling depth of proteome [38,39]. Withal, targeted proteomics methods (e.g., SRM and PRM) are analytically suitable experiments for quantification based on MS because of their exquisite selectivity and sensitivity [40]. However, continuing improvements in targeted approaches are currently under way, and an improved analytical workflow is still needed to ensure precise quantification [41]. The main advantage of gel-based approaches is the evaluation of the protein map (or proteoform map) of a sample regarding the hydrophobicity and molecular weight characteristics from intact polypeptides. In this context, DIGE is the state-of-the-art in two-dimensional gels because of its sensitivity, reproducibility and increased linear dynamic range for protein spot comparisons [42]. Here, the internal standard, a control sample and a treated sample of the same extraction were run in the same gel, as shown in Supplementary Table 1. As previously described [43], this experimental design was used to prevent the impairment of gel image overlays among all the samples in the DeCyder software because of the potential substantial differences among the gel images (protein patterns) that could be generated by each treatment. By having the control sample (of the corresponding extraction sample) in each gel, a comparison among the gels can also be performed as a traditional two-dimensional electrophoresis comparison, and the image overlay is guaranteed in each gel (control and treatment). No substantial differences among the gel images were observed, and the overall image overlay was successfully performed using DeCyder software.
The DIGE approach was clearly decisive for assessing the quantitative proteomic map of bloodstream trypomastigotes that were treated with naphthoimidazoles. This assessment is better evidenced by the reduction in the sample amount in comparison with our previous epimastigote work. In that study, 500 μg of each sample was added per gel [19], and it was reduced 10-fold (50 μg/sample) in our experimental design here. The fluorescence approach led to the detection of 1,724 protein spots (three-fold more than those found by Coomassie blue staining technique for the epimastigotes) in the narrower pH 4–7 gradient applied here. Interestingly, despite the increase in the number of detected protein spots, the number of differential protein spots that was identified was quite similar, at 30 and 34 for epimastigotes and trypomastigotes, respectively.
As with epimastigotes, the choice of the concentration for the three compounds was based on previously calculated IC50/24 h values [13,14], never exceeding half the dose that induces lysis in 50% of the parasites, to prevent undesirable and non-specific effects. Our previous proteomic study indicated that the most remarkable number of modulated proteins in epimastigotes were mitochondrial [19]. Here, we showed that a great number of mitochondrial proteins (7 of 27 differentially abundant proteins) were present at altered levels in treated trypomastigotes. These data were reinforced by ultrastructural and biochemical evidence showing that this organelle was the primary target of naphthoimidazoles [13,14]. Unlike what was described for treated epimastigotes, none of the differentially abundant proteins identified in trypomastigotes treated with these compounds was detected in benznidazole-resistant epimastigotes by Andrade and co-workers (2008), likely because a different parasite form was used.
Among the trypomastigote proteins that were differentially abundant after the treatment, chaperones were the most recurrent proteins that were modulated. Heat shock proteins 60, 70 and 85 (spots 461, 525, 607, 665, 666, 674, 1075, 1084, 1294, 1409, 1414, 1480, and 1488) were modulated by N1 and N2, as observed in epimastigotes that were treated with the same naphthoimidazoles [19]. Interestingly, an overexpression of heat shock proteins 60 and 70 was detected in both trypomastigotes and epimastigotes after N1 treatment, suggesting a parasite injury derived from proteolytic and/or oxidative stress that led to an increase in the chaperone content. However, a decrease in the chaperone levels was observed, especially with regards to the heat shock protein 85 content, in N1- and N2-treated bloodstream and insect forms; this finding deserves further functional analysis. This reduction could be attributed to induced oxidative dysfunction, as described for other cell models [44].
After treatment with N1, N2 and N3, tubulin was the most down-regulated protein in epimastigotes as previously observed. ELISA assays indicated a decrease in the tyrosinated tubulin content, but the levels of acetylated tubulin were not altered in treated epimastigotes [19]. The tyrosinated isoform is directly related to labile microtubules that participate in vesicular trafficking and also make up the intranuclear mitotic spindle, and detyrosination has been shown to act as an important checkpoint in the trypanosomatid cell cycle [45]. The acetylated isoform is present in stable microtubules that were localized in the flagella and in the subpellicular cage of the parasite [46]. Our present data showed that tubulin was also down-regulated in trypomastigotes that were treated with N1 and N2 (spots 1235 and 1624). As observed in treated epimastigotes, no ultrastructural injury was detected in the subpellicular and flagellar microtubules of trypomastigotes after naphthoimidazole treatment [13,14], reinforcing the hypothesis that a reduction in the tyrosinated tubulin levels could also compromise vesicular trafficking in the bloodstream forms.
Here, elongation factor 2 (spot 264) was up-regulated in N3-treated trypomastigotes. In epimastigotes, this elongation factor as well as elongation factors 1-alpha and 1-beta were down-regulated by N1 and N2 [19]. These GTP-dependent enzymes participate in protein synthesis in eukaryotes [47], suggesting an impairment of the T. cruzi protein synthesis machinery, but complementary experiments must be performed to confirm the hypothesis.
An activated protein kinase C receptor (spot 1264) was down-modulated by N2, as found before in N3-treated epimastigotes [19], indicating that signaling cascades could participate in the mode of action of these compounds. The regulation of several transporter systems was performed by protein kinase C, which was also implicated in the host cell infection [48]. A decrease in protein kinase C receptor expression could lead to a consequent reduction in T. cruzi transduction signaling, impacting the parasite's infectivity.
Enolase (spot 949) was up-modulated by N1 treatment, similar to that observed in N3-treated epimastigotes [19]. This enzyme is responsible for an essential step in energetic metabolism, which is the reversible conversion of 2-phosphoglycerate to phosphoenolpyruvate in glycolysis and gluconeogenesis [49]. The increase in this enzyme level can represent a mechanism to compensate for the energetic misbalance caused by these drugs.
Furthermore, some proteins were differentially abundant in bloodstream forms that were treated with naphthoimidazoles, but these differences were not detected in epimastigotes after the treatment. This result is expected because of the significant variation between the proteomic map of both parasite forms [24]. Some enzymes from distinct metabolic pathways were up-regulated in N1-treated parasites. Asparagine synthetase A (spot 1070) is responsible for the production of asparagine from aspartate, a non-essential amino acid that is present in several proteins. Recently, an in vitro study suggested one inhibitor of this enzyme as an alternative target in African trypanosomiasis [50]. One possible hypothesis is that N1 treatment may reduce the content of asparagine-containing proteins, and, as a result, the level of asparagine synthetase is increased, but assays with this labeled amino acid should be performed for confirmation.
ATP synthase subunit B (spot 787) is directly related to energetic metabolism by using the ions that flux to ATP synthesis in the mitochondrion [51]. The succinyl-CoA ligase [GDP-forming] beta-chain (spots 1278, 1393) catalyzes the reversible reaction of succinyl-CoA to succinate in the citric acid cycle [52]. The extensive mitochondrial swelling and the decrease in this organelle membrane potential was previously observed [13,14], together with the modulation of mitochondrial proteins described here, which strongly indicated that the essential pathways in T. cruzi mitochondrion are a primary target of N1, and the likely energetic failure that ensues leads to parasite death.
However, peptide methionine sulfoxide reductase (spot 1674) was down-regulated in trypomastigotes after N1 treatment. This enzyme reduces methionine sulfoxide to methionine, protecting the cells from oxidative damage. In T. cruzi, antioxidant studies focused especially on the trypanothione system, given that the mechanisms for repairing oxidized proteins are very poorly described [53]. A reduction in the methionine sulfoxide reductase in treated parasites leads to an accumulation of non-repaired oxidized macromolecules in the protozoa. Interestingly, trypanothione synthetase was up-regulated in epimastigotes after treatment with three naphthoimidazoles; however, this enzyme was not differentially abundant in treated bloodstream trypomastigotes. Despite the absence of redox properties in these compounds that could lead to reactive oxygen species generation, unpublished data from our group pointed to the reversion of the trypanocidal effect of N1, N2 and N3 by classical antioxidants such as tocopherol and urate in both epimastigotes and trypomastigotes. Pre-incubation with these antioxidant agents also led to a decrease in the reactive species produced by naphthoimidazole treatment in the insect forms (Menna-Barreto, personal communication). Moreover, the up-regulation of guanine deaminase (spot 942) after N1 treatment was observed. This enzyme converts guanine to xanthine in purine metabolism. Subsequently, the reaction of xanthine with water and molecular oxygen, as catalyzed by xanthine oxidase, produces urate and hydrogen peroxide [54]. This reaction could indirectly explain the reactive oxygen species produced by these naphthoimidazoles. However, a complementary analysis that employs different biochemical and molecular techniques will be performed to elucidate the mechanisms that are involved.
Some cytoskeleton-associated proteins were up-regulated in N1-treated parasites as follows: RAB GDP dissociation inhibitor alpha (spot 941) and two different components of the paraflagellar rod (spots 473 and 477). This inhibitor prevents the Rabs function in the docking step during vesicular trafficking through the inhibition of the GDP dissociation and the subsequent GDP/GTP exchange [55]. This finding suggests an impairment of cellular trafficking that is reinforced by the reduction in tubulin levels as discussed above. However, the paraflagellar rod is an extra-axonemal structure typical of trypanosomatids that are made of several proteins, and it plays a pivotal role in flagellum beating [56]. The overexpression of some elements of this structure could represent a compensatory mechanism for the misassembly of some flagellar accessory proteins that culminates in a reduction of parasite motility as previously detected [13,14].
Interestingly, one protein was down-regulated only in N2-treated trypomastigotes. Arginine kinase (spot 1440) is the transferase that is responsible for the production of phospho-L-arginine from L-arginine during an ATP-dependent reaction. Phospho-L-arginine was found to represent the energetic reservoir, and it is crucial for epimastigote proliferation [57]. Arginine kinase activity has been considered as a regulator of this process, and the decrease in this enzyme level could lead to the energetic collapse of the parasite.
In 2005, El-Sayed and colleagues sequenced the complete genome of the T. cruzi CL-Brener strain, but a huge quantity of the genes were annotated as hypothetical genes because of their absence of any putative biological functions [58]. Three different hypothetical proteins were identified in trypomastigotes that were treated with N1 (spots 567, 781 and 1612) and N2 (spot 567). A BLAST analysis showed 38% shared identity and 76% coverage (E-value: 5e-110) of the hypothetical protein entry of spot 567 with a putative mitochondrial nucleolar protein (gi|928109699), and 99% shared identity and 86% coverage (E-value: 0.0) for the entry of spot 781 with hydroxymethylglutaryl-CoA synthase (gi|686631215). The predicted sequence of spot 1612 did not show similarities with any protein family.
Ultimately, mammalian gelsolin and actin (spots 943, 970 and 1218) were also identified as modulated proteins after the treatment. In our recent description of the proteomic map of bloodstream trypomastigotes, we discussed the presence of blood components, especially the plasma, erythrocytes and platelets of Mus musculus in these samples [27]. Even the high stringency of the parasite purification process is not enough to avoid the identification of host proteins. The adsorption or the specific binding of the mammalian proteins on the parasite surface could not be discarded.
Our data together with a previous analysis indicate that the mechanisms of action of the three naphthoimidazoles are complex, and they involve distinct metabolic pathways such as cellular trafficking, protein synthesis, transduction signaling and energetic metabolism, among others, open interesting perspectives for trypanocidal strategies. Further studies on these metabolic interactions are necessary to answer some outstanding questions. Even so, new strategies for drug design have been improved by recent outcomes in T. cruzi biochemistry, allowing for better comprehension of the effects of trypanocidal agents.
The accession numbers for proteins mentioned in the text are listed as follows: elongation factor 2 [T. cruzi] gi|407835084; glucose-regulated protein 78, putative [T. cruzi] gi|407842744; paraflagellar rod component [T. cruzi] gi|2209137; major paraflagellar rod protein [T. cruzi] gi|162179; heat shock protein 70 [T. cruzi] gi|205278868; hypothetical protein, conserved [T. cruzi] gi|70871170; chaperonin, putative [T. cruzi] gi|70885659; heat shock protein 60 kDa [T. cruzi] gi|1495230; heat shock protein 60 kDa [T. cruzi] gi|1495230; heat shock protein 60 kDa [T. cruzi] gi|1495230; hypothetical protein, conserved [T. cruzi] gi|70883145; vacuolar ATP synthase subunit B, putative [T. cruzi] gi|70870795; RAB GDP dissociation inhibitor alpha, putative [T. cruzi marinkellei] gi|407410583; guanine deaminase, putative [T. cruzi] gi|70874663; gelsolin, isoform CRA_a [Mus musculus] gi|148676699; enolase [T. cruzi] gi|407849788; cytoplasmic beta-actin, partial [Mus musculus] gi|49868; asparagine synthetase A, partial [T. cruzi] gi|348658746; heat shock protein 70 [T. cruzi] gi|205278868; 85 kDa protein [T. cruzi] gi|162111; A-X actin [Mus musculus] gi|309090; beta tubulin 1.9 [T. cruzi] gi|18568139; activated protein kinase C receptor, putative [T. cruzi] gi|70882943; succinyl-CoA ligase [GDP-forming] beta-chain, putative [T. cruzi] gi|407849036; 85 kDa protein [T. cruzi] gi|162111; succinyl-CoA ligase [GDP-forming] beta-chain, putative [T. cruzi] gi|407849036; heat shock protein 70 [T. cruzi] gi|205278868; 70 kDa heat shock protein [Trypanosoma rangeli] gi|119394469; arginine kinase, putative [T. cruzi] gi|407844351; heat shock protein 70 [T. cruzi] gi|205278868; heat shock 70 kDa protein, putative [T. cruzi] gi|70876223; hypothetical protein TCDM_14147 [T. cruzi Dm28c] gi|557861434; alpha tubulin [T. cruzi] gi|1220545; peptide methionine sulfoxide reductase [T. cruzi strain CL Brener] gi|71405176.
|
10.1371/journal.pcbi.1006684 | Dynamic filopodial forces induce accumulation, damage, and plastic remodeling of 3D extracellular matrices | The mechanical properties of the extracellular matrix (ECM)–a complex, 3D, fibrillar scaffold of cells in physiological environments–modulate cell behavior and can drive tissue morphogenesis, regeneration, and disease progression. For simplicity, it is often convenient to assume these properties to be time-invariant. In living systems, however, cells dynamically remodel the ECM and create time-dependent local microenvironments. Here, we show how cell-generated contractile forces produce substantial irreversible changes to the density and architecture of physiologically relevant ECMs–collagen I and fibrin–in a matter of minutes. We measure the 3D deformation profiles of the ECM surrounding cancer and endothelial cells during stages when force generation is active or inactive. We further correlate these ECM measurements to both discrete fiber simulations that incorporate fiber crosslink unbinding kinetics and continuum-scale simulations that account for viscoplastic and damage features. Our findings further confirm that plasticity, as a mechanical law to capture remodeling in these networks, is fundamentally tied to material damage via force-driven unbinding of fiber crosslinks. These results characterize in a multiscale manner the dynamic nature of the mechanical environment of physiologically mimicking cell-in-gel systems.
| Many cells in the body are surrounded by a 3D extracellular matrix of interconnected protein fibers. The density and architecture of this protein fiber network can play important roles in controlling cell behavior. Deregulated biophysical properties of the extracellular environment are observed in diseases such as cancer. We demonstrate, through an integrated computational and experimental study, that cell-generated dynamic local forces rapidly and mechanically remodel the matrix, creating a non-homogeneous, densified region around the cell. This substantially increases extracellular matrix protein concentration in the vicinity of cells and alters matrix mechanical properties over time, creating a new microenvironment. Cells are known to respond to both biochemical and biomechanical properties of their surroundings. Our findings show that for mechanically active cells that exert dynamic forces onto the extracellular matrix, the physical properties of the surrounding environment that they sense are dynamic, and these dynamic properties should be taken into consideration in studies involving cell-matrix interactions, such as 3D traction force microscopy experiments in physiologically relevant environments.
| The Extracellular Matrix (ECM) is a scaffolding medium that helps transmit mechanical signals among cells in cancer [1,2], fibrosis [3,4], vascular networks [5,6], and more generally, morphogenesis [7,8]. The mechanical and biochemical properties of the ECM impact cell behavior. The stiffness of the local environment and the tensional response from cells can induce invasive phenotypes in tumors [9–12], guide differentiation in stem cells [13], and alter vascular function [14]. The fibrillar nature and local architecture of the ECM can lead to directed cell migration [15], and increased density and alignment in the tumor stroma are correlated with more aggressive disease and worse prognosis in preclinical and clinical samples [16,17]. ECM remodeling through cell contractility is also potentially a fundamental factor in tissue folding and shaping during development [18]. It is not clear, however, how ECM spatio-temporal evolution in living systems is controlled by cells to promote physiological and pathological states.
Many studies have quantified the mechanical signals transmitted by ECMs, mostly assuming ideal ECM material properties. Studies usually derive the magnitude of forces exerted by cells through imaging of fluorescent markers tethered to the ECM [19,20]. Because it is difficult to back-calculate forces in heterogeneous, dynamic environments, these approaches rely on 3D biopolymers or 2D substrates with time-invariant mechanical responses. The spatiotemporal evolution of the ECM is however relevant in many mechanobiological processes [3,18,21,22]. For instance, in angiogenesis and vasculogenesis, together with chemical signaling driving formation or inhibition patterns [23], endothelial cells mechanically sense each other [24], and cooperate to form tubular shapes by remodeling the fibrous 3D ECM [5,6]. Furthermore, mechanical signals are amplified, resulting in long-range force transmission, when ECMs are fibrillar, via local alignment and force-driven anisotropy [4,25,26]. More complex descriptions of fibrous materials are taken into account in recent 3D studies of forces in biological processes [27].
ECM remodeling remains very challenging to decipher, despite its biological ubiquity. Remodeling entails dynamic molecular processes such as cell-fiber interactions, proteolytic degradation, and crosslinking sites binding and unbinding that ultimately lead to global changes in the ECM network. Additionally, cell-scale forces are sufficient to drive ECM remodeling, and remodeled ECMs can in turn modulate mechanosensing in cells, resulting in dynamic feedback. The dynamic mechanical states of cells and the ECM, especially in physiologically relevant conditions, are not well understood.
Here we investigate cell-induced remodeling of physiologically relevant 3D fibrillar ECMs, specifically fibrin and collagen. We focus on cell force-induced non-reversible remodeling of the ECM, which interestingly occurs on the time scale of minutes and drastically changes local architectures. By toggling the tensional state of the cell, we capture and distinguish both plastic and elastic strains in the ECM. We find that cell-generated mechanical forces are sufficient to accumulate ECM at the cell periphery in a dynamic process that depends on actin nucleation factors. We then perform computational simulations using a network model with discrete ECM fibers to examine quantitatively the impact of dynamic, filopodial-like cellular forces and reaction kinetics at the ECM component level on tension and concentration profiles within the ECM network. We lastly propose that constitutive damage and plastic softening at the continuum level are capable of recapitulating both experimental and fiber-level simulation findings. Collectively, these findings might have profound implications in mechanobiology, especially in the context of cell traction force studies.
We explore remodeling by quantifying ECM dynamics in two physiologically relevant cell-ECM combinations cultured in 3D in vitro conditions. As a first combination, we use endothelial cells in fibrin gels given the well-known ability of these cells to form physiologically mimicking microvascular network structures [28]. As a second combination, we use breast cancer cells in collagen I gels, a highly abundant component in the tumor stromal microenvironment. Using both combinations, in one set of experiments, we inhibit cell-generated forces at the time of gelation and seeding by pre-treating cell-ECMs with Cytochalasin D. This treatment allows us to start from a force-free configuration (Fig 1A). After cells are seeded, we remove the Cytochalasin D and let the cells recover their ability to generate forces over a period of several hours. Finally, after recovery, we lyse the cells (decellularization) with a detergent to fully relax all forces in the cell-ECM construct. We use confocal microscopy to quantify deformations through 3D Digital Volume Correlation (DVC) algorithms using fluorescent signal intensity from pre-labeled fibrin and collagen gels [29–31].
In both cell-ECM combinations, we find that remodeling involves non-elastic, i.e. non-recoverable, deformations of the ECM. This plastic remodeling of recruited fibers is mainly force-driven; it is prevented by Cytochalasin D pre-treatment that inhibits cell-generated forces (Fig 1A). Also, increasing the crosslinking of fibrin results in a decrease in the displacement length, a measure of the average radial displacement of the ECM toward the cell (see Methods for details) (Fig 1B). Crosslinking has an important effect on plastic remodeling, with poor crosslinking resulting in the enhancement of fiber concentration at the cell-matrix boundary (Fig 1A–1C). For each cell, this effect is quantified by the densification factor (DF), the ratio of the average ECM fluorescence intensity near the cell to that far from the cell (Fig 1C). We further quantify elastic recoverability through the metric recoverability index (RI), the ratio between the reduction in displacement length caused by decellularization and the displacement length prior to decellularization, expressed as a percent. Details of these metrics can be found in the Methods section.
For all cases, after washing Cytochalasin D from the cells, we observe a significant degree of matrix remodeling in less than 1 hour (Fig 1B and 1C). To discriminate the time scale for eliminating Cytochalasin D from the cell from that for intrinsic cell contraction, we also quantify deformations immediately after seeding without Cytochalasin D pre-treatment (Fig 1D). We demonstrate that (i) the intrinsic cell contraction and ECM remodeling dynamics occur over the course of minutes and (ii) the remodeling rate diminishes within an hour (Fig 1D, S1 Fig, and S1 Video (MDA-MB-231 cell in 3mg/mL collagen), S2 Video (MDA-MB-231 cells in 1.5mg/mL collagen), and S3 Video (MDA-MB-231 cells in 1.5mg/mL collagen, overlay of fluorescent F-actin and collagen)). All tested cell-ECM combinations exhibit a plateau in displacement length indicating that remodeling stabilizes in the course of hours with substantial irreversible components (Fig 1B and S2 Fig). For endothelial cells in fibrin, lower crosslinking increases the degree of plastic deformation corresponding to a lower RI (Fig 1E).
We further consider how remodeled ECMs absorb and transmit forces in space. In general, both endothelial and cancer cells apply centripetal tractions, as demonstrated by the directions of the local ECM displacements (Fig 2A–2C). Plastic recruitment leads to a substantially higher magnitude of cumulative matrix displacement magnitudes ‖uoverall‖ than the elastic displacement ‖udecell‖ alone (Fig 2C). To assess how cell-generated ECM deformations propagate spatially, we measure the radial profile (from the cell) of ECM displacement magnitudes (or lengths), normalized by the displacement magnitude at the cell boundary (Fig 2B). We find that force transmission depends strongly on the specific cell-ECM pair, with the displacements decaying steeper in collagen than fibrin. In all cases, displacements decay with a lower gradient than in the ideal case of an isotropic linear elastic material, which confirms the long-range mechanical reach of cell forces in fibrous ECMs (Fig 2B). Fully and partially crosslinked fibrin matrices behave similarly in their ability to propagate displacements (Fig 2B, solid lines). We also observe similarities when comparing the decay of the overall displacement during remodeling to the decay of the purely elastic component of the displacement (based on measurements right before and after decellularization).
We next investigate possible mechanisms of cell force-driven remodeling at the ECM fiber scale by targeting actin nucleating factors that are important in filopodial dynamics; CK666 inhibits Arp2/3 and SMIFH2 inhibits formins. Both of these are observed to significantly reduce ECM remodeling (Fig 3A and 3B). We further show that proteolytic activity, when inhibited with GM6001 during the early remodeling process, does not appear to have a substantial effect on matrix recruitment on the timescale of hours (Fig 3A and 3B). These findings implicate dynamic cell force generation transmitted to the ECM fibers via filopodial projections in ECM accumulation, with little or no reliance on matrix degradation.
Our experimental results demonstrate that mechanical forces, mediated by dynamic actin nucleation-driven processes, and ECM crosslinking can modulate the plastic recruitment of ECM to the vicinity of cells. Here, through discrete fiber network computational simulations, we examine how the interplay between applied dynamic mechanical forces, mimicking filopodia-driven events, and kinetic fiber-fiber connections (crosslinks) lead to varying degrees of ECM remodeling and stress profiles. We extract quantitative details of how local molecular features can influence global reorganization dynamics of the ECM under cell-generated forces. A 3D fiber network, mimicking the ECM, is generated by polymerizing monomeric units into elastic fibers that can stretch and bend. Each fiber is a chain of cylindrical segments, and each segment follows the following potentials:
Us=12κeΔr2
(1)
Ub=12κbΔθ2
(2)
where Us is the potential energy from stretching, Ub is the potential energy from bending, κe is the extensional stiffness, κb is the bending stiffness, Δr is the deviation from the equilibrium length, and Δθ is the deviation from the equilibrium angle. Each monomeric unit adds a cylindrical segment to the fiber during polymerization, and fibers nucleate in random directions during initial network formation. Neighboring fibers are connected with crosslinks, if present, that can unbind in a force-sensitive manner in accordance with Bell’s model [32]:
ku=ku0eλFkBT
(3)
where ku is the crosslink unbinding rate, ku0 is the zero-force unbinding rate, λ is the mechanosensitivity (i.e. mechanical compliance) of the crosslink, F is the magnitude of the extensional force acting on the crosslink (only positive stretching forces contribute), kB is the Boltzmann constant, and T is temperature. Model parameters are listed in S1 Table.
The network is athermal and the components (fiber segments, crosslinks) follow the equation of motion:
Fc,i+Fi−ζidridt=0
(4)
where i is the index of the component under consideration, Fc,i is the cell generated loading force near the–z boundary, Fi is the mechanical force from the fiber network, which includes extension, bending, and repulsion (volume exclusion) [33] of the fibers and crosslinks, ζi is the drag coefficient, and ri is the position. Eq 4 is solved over time through Euler integration at discrete time steps to determine the position of each element in the network. Crosslink unbinding is modeled stochastically. Each bound crosslink has an unbinding probability at each time step Δt equal to:
Punbind=1−exp(−kuΔt)
(5)
Additional details of the discrete fiber network model, which has been applied previously to simulate other filamentous networks, can be found in [33,34]. Sample simulation are shown in Fig 3C and 3D, S4 Video (high loading forces, high ECM recruitment), and S5 Video (moderate loading forces, moderate fiber recruitment). Force loading in these simulations mimics filopodia and is described in more detail later.
Parameter values for the simulated fibers and network are chosen based on plausible values for ECM fibers (collagen I and fibrin) [35–39], and experimental network features (S3 Fig). We simulated moderately thick ECM fibers, which are ~ 100nm in diameter [40,41]. The Young’s modulus of an ECM fiber can be on the orders of tens of MPa’s for fibrin [35] and hundreds of MPa’s for collagen [37–39]. For simplicity, we picked an arbitrary value in this range (125MPa) and focused on the kinetic features of the model, driven by the force-sensitive unbinding of the crosslinks. Crosslinks have two arms, each 20nm, mimicking ECM molecular subunits that can connect fibers [42]. We explored a range of crosslink behaviors that spans relatively extreme cases (near permanent to highly transient), relative to expected fibrin bonds [36] to capture limiting network-level behaviors. For simplicity and computational feasibility, we only consider the thick fiber structures and one type of crosslink (fiber-fiber connections), which enable us to capture the dynamic connectivity of the ECM network, of focus here (S3 Fig, which shows that fiber-fiber contacts are prominent).
Once the crosslinked network is generated, it is allowed to relax to a stable state, in which the prestress built up during network formation has relaxed to a plateau close to zero. Thereafter, loading forces simulating filopodia are applied on one side of the network for a fixed duration of time and then reduced to zero to allow the network to relax to a new, potentially remodeled state. In our time series analyses, t = 0 corresponds to the initiation of force loading and t = 1 corresponds to the cessation of active forces, where t is the normalized time. In our simulations, the fiber ends at the +z boundary are fixed to mimic the resistance from fibers far away. The x and y boundaries are periodic, and the domain size is 20x20x20μm3. Filopodial force loading is applied such that any fiber segment that reaches within a certain distance (2μm) of the–z boundary experiences a local point force pulling it toward that boundary. We explore a range of force magnitudes from 1pN to 1nN to capture the impact of physiologically plausible cell generated forces. This loading condition mimics a pulling process where new filopodia are continuously generated that adhere to and pull new fiber segments near the cell. This type of loading is needed in order for fibers to be continuously recruited toward the cell, and many dynamic actin protrusions are indeed observed on the periphery of cells embedded inside a 3D ECM as shown in Fig 3E and S3 Video (overlay of fluorescent F-actin and collagen fibers during dynamic ECM remodeling). A schematic illustrating this loading condition is shown in Fig 3F.
We then examine the network remodeling dynamics under different conditions, modulating experimentally tunable or physiologically relevant parameters. Specifically, we consider different loading forces, crosslink densities, zero-force unbinding rates of crosslinks, and crosslink bond mechanosensitivities. These parameters aim to capture the impact of cell traction, the degree of ECM crosslinking, and the kinetic nature of crosslinks. ECM fibers are initially recruited to the loading boundary once applied forces are activated (Fig 3C and 3D). Temporal profiles of ECM accumulation near the cell boundary (region within 3μm of the force loading boundary) and the peak accumulated ECM concentration (over time) under varied loading forces are shown in Fig 4A and 4B, respectively. Similar plots for varied crosslink concentrations are shown in Fig 4C and 4D, respectively. The temporal profiles show that in some conditions (relatively low applied force magnitudes, high crosslink concentrations), after the loading forces are deactivated (at the normalized time of 1), the network recovers primarily elastically and the normalized ECM concentration in the accumulation region returns close to 1, the uniform network state prior to loading. Note that for high crosslinking cases, the accumulated concentration does not fully reverse after relaxation due to crosslink unbinding still having occurred, resulting in some plastic remodeling (Fig 4C). For very low loading forces, near full recovery is observed after relaxation (Fig 4A). Conversely, higher loading forces and lower crosslink concentrations lead to relatively high plastic remodeling, in which the recruited ECM fibers do not relax back to their original positions after force loading is stopped. S4 Video, S5 Video, and Fig 3D show the 3D ECM network evolution and fiber recruitment due to loading forces. In these cases, the network permanently remodels over time with recruited fibers remaining near the loading boundary after the cessation of applied forces. Note that after forces are relaxed, there are no adhesions between fibers and the force-loading boundary, mimicking decellularization in our experiments. The elastic restoring forces will then tend to pull the accumulated fibers away from the loading boundary, thus shifting the position of the maximum concentration (Fig 3D, S4 Video, S5 Video).
Furthermore, fiber recruitment vs. loading force (log scale) displays a sigmoidal trend, in which minimal ECM fiber recruitment occurs under low loading forces below a threshold, increasing ECM fiber recruitment occurs with increasing loading forces at an intermediate range, and a plateau is reached for forces above a second threshold (Fig 4B). Under the same loading forces, increases in crosslink concentration first lead to more ECM recruitment, followed by a trend reversal and decline in network remodeling (Fig 4D). This suggests that the network gains connectivity with higher crosslink concentration, enabling connected fibers farther away to be recruited. However, beyond a certain concentration, plastic recruitment is reduced, as loading forces are distributed between more crosslinks, leading to reduced crosslink unbinding rates. Note that while the crosslink reactions in our model are simulated in a stochastic manner, the overall network behavior is robust, as demonstrated by the results from repeated simulations of selected conditions (S4 Fig). We additionally explore ECM concentration profiles for different crosslink kinetics (S5 Fig). Notably, relatively high zero-force unbinding rates and mechanosensitivities, i.e. weak crosslinks, lead to plastic accumulation of the ECM, and there is a biphasic relationship between the amount of recruited ECM and the crosslink mechanosensitivity, similar to the effect of varying crosslink concentration.
We next consider the overall stress profiles in the ECM network under our loading condition. Stresses are calculated by summing the normal component of forces acting on fibers crossing a plane parallel to the cell surface divided by the area of the plane. Stress profiles during loading (normalized time 0 to 1) are highly dynamic, often exhibiting a sharp initial peak followed by relaxation, especially under high force, as fibers are being recruited plastically and fiber crosslinks unbind (Fig 5A–5D). At relatively low applied forces, the stress profile does not decay and instead reaches a plateau, as crosslink unbinding and network relaxation are minimal during the loading period. Larger loading forces lead to larger overall peak stresses in the network, but also more unstable stresses (Fig 5B and 5C). When crosslink concentration is varied under the same loading forces, network stress is low at low crosslink concentrations as unbinding prevents a high build-up of stress. Network stress can build up to a plateaued level with more crosslink support (Fig 5D–5F, S6 Fig). Furthermore, when the kinetics of the crosslinks is tuned, more stable crosslinks lead to larger stress build-up and sustained stress levels, while crosslinks that unbind more quickly lead to reduced and less sustainable network stresses (S7 Fig). Note that the simulations discussed so far do not consider the possibility of the rebinding of unbound crosslinks. We find that enabling rebinding partially diminishes ECM recruitment and network stress dissipation (S8 Fig). Overall, our simulation results demonstrate that filopodial or filopodial-like forces acting on a kinetically connected ECM can spontaneously lead to ECM densification near the cell surface and dynamic stress profiles in the surrounding microenvironment. The amount of ECM recruitment and the temporal stress profile depend on the interplay between the magnitude of the dynamic pulling forces and the concentration and kinetic properties of the ECM crosslinks.
A direct comparison of our simulation and experimental results (Fig 6) shows that by varying crosslink concentration alone in our simulations we can capture some of the differences observed in our experimental results between fibrin (low crosslinking), fibrin (high crosslinking), and collagen. Fibrin (high crosslinking) displays relatively low ECM accumulation, followed by recovery toward the initial state after relaxation, consistent with highly crosslinked simulated networks (crosslinking of 0.5–1). Both fibrin (low crosslinking) and collagen demonstrate high accumulation, much of which is non-recoverable, consistent with weakly crosslinked simulated networks (crosslinking of ~0.1).
Our results implicate possible consequences for traction force microscopy (TFM) in complex 3D ECMs. In TFM studies, typically a continuum material model is used for calculating forces from strains measured through imaging fluorescent markers embedded in a deformable substrate. Here, we develop a continuum model that essentially coarse-grains fiber-level mechanics and kinetics into continuum scale parameters, and we emphasize the impact of crosslink unbinding on material properties. To capture the impact of crosslink unbinding at the continuum material scale, we utilize viscoplastic and damage features that enable creep and stress relaxation responses, as guided by our fiber network simulation results. We start from a general viscoplastic model (Norton-Hoff) (Fig 7A and S1 Note). In this model, the viscous element simulates the creep response, and the plastic element simulates permanent, inelastic deformations. To further recapitulate the effects of crosslink unbinding on the elastoplastic properties, we add both (i) ‘elastic damage’, i.e. an exponential decay in elastic stiffness (E) starting from above a critical maximum tensile elastic strain (ε1),E=Ae−Bε1, and (ii) ‘plastic damage’ (softening), i.e. a linear decay of yield stress starting from above a critical plastic strain (Fig 7B). For elastic damage, we use a general form of the exponential decay, with positive parameters A and B obtained from fitting based on start and end points (ε1s, Es) and (ε1e, Ee), respectively, of damage principal strain and stiffness (see below for a sensitivity study on the damage start and end points). For softening, a linear decay is used in the yield stress–plastic strain space (Fig 7B and S1 Note). Conceptually, the elastic damage feature relates to the stiffness of the material becoming lower as crosslinks unbind, and the softening feature relates to the higher unbinding rate of crosslinks as fewer bound crosslinks remain due to the increased load per remaining crosslink.
We test this viscoplastic model with damage and softening in a finite element simulation with spherical symmetry of a cell contracting centripetally and displacing the surrounding ECM with similar magnitudes as in the experiments. We then relax the contractile force to simulate the experimental decellularization. This force is applied slightly outward from the edge, simulating the action of filopodia recruiting relatively close fibers (Fig 7C). The force is meant to produce mostly tensile stresses in the continuum but also some degree of local compression at the cell-ECM interface. The presence of elastic damage recapitulates (i) the decrease in equivalent stress at the edge in perfect viscoplasticity (Fig 7D, blue lines), which is further decreased by the presence of plastic softening (red lines), in agreement with the filament model when lowering the density of crosslinks (Fig 5D) and (ii) the increase in bulk displacement at the edge with long-term loading (Fig 7E) observed when lowering the density of crosslinks (Fig 1B). However, as the elastic damage lowers the elastic modulus, it still results in more elastic recoverability than the stiffer, non-damaged material, and thus contradicting the experimental RI (Fig 1E). Instead, plastic softening is needed to reproduce the loss in recoverability observed experimentally as the density of crosslinks decreases (Fig 7F), which occurs in conjunction to the increase in plastic strain (Fig 7G).
We next run parametric analyses to assess how sensitive the predictions of both the elastic recoverability and the accumulation of ECM damage are to the level of force and creep time (Fig 7H and 7I). We find that both an increase of cell traction forces and persistence in loading and recruitment simulated through longer creep periods can lead to a dramatic decrease in elastic recoverability, which is accompanied by an increase of damaged regions. Thus, larger forces and longer creep times will lead to more irreversible remodeling, as also suggested by the discrete model. We also study the model sensitivity to elastic damage parameters. Note that damage is programmed to initiate at an onset strain level, ε1s, and ends at a saturating strain level, ε1e (Fig 7B). Because both of these are elastic strain constants, we chose to link ε1e to the maximum elastic strain observable experimentally during relaxation (S9 Fig), and leaving the choice of ε1s as arbitrary. Nonetheless, we assess how these constants can influence the damaged region at the end of the loading process (S10 Fig). As expected, an earlier onset strain (e.g. 1%) for Young’s modulus decay will produce larger damaged regions. Also, increasing the end strain will lead to a lower damage radius, since more strain is required in order to reach larger damage levels (S10 Fig). We further show that the damage radius plateaus and can be from one to three cell diameters away from the cell edge, and it expands as the loading is held constant during creep (Fig 7H and 7I, S10B and S10C Fig, S12B Fig).
For the applicability of these continuum concepts toward quantifying cellular traction forces, we finally seek to estimate the errors introduced when elastic and plastic damage phenomena are ignored. For example, tracking the cumulative ECM deformation from an initial zero-force state via imaging, on its own, does not enable the separation of elastic and plastic deformations. An ‘apparent’ stress, back-calculated based on this total deformation and assuming linear elasticity of the ECM material, would be higher than the true stress because of plastic yield. The ratio between the true stress and the apparent stress is given by the recoverability index that has been defined and studied experimentally in Fig 1. Furthermore, a typical traction force study quantifies reference cellular stresses based on relaxing substrate deformations at the experimental end time, e.g. via cell relaxation by trypsinization, lysis, or actomyosin inhibition, and assuming nominal substrate stiffness values. Our model suggests that this quantification would be inaccurate due to plastic changes to the substrate material properties. Considering only the phenomena introduced here, i.e. elastic damage and plasticity, we estimate–with the set of parameters chosen, and experimental evidence available–that this apparent stress can again be much higher than the true stress (S11 Fig, S12C Fig, up to five time higher). We find that this is largely due to the damage process locally degrading the initial elastic modulus, but the creep duration can also have a notable effect (Fig 7I, S11 Fig, S12 Fig).
Many questions remain regarding the mechanics and dynamics of cell-ECM interactions in complex 3D matrices. Early studies demonstrated that cells, particularly fibroblasts, can contract collagen gels in a manner in which the degree of reversibility was dependent on the duration of contractile activity [43]. Furthermore, the mechanical properties of ECMs are irreversibly altered due to remodeling by cells [44]. These findings point toward non-elastic mechanical responses in ECMs due to cell force-mediated mechanisms. In more recent studies, confocal reflectance microscopy imaging has been widely used as a label-free technique to visualize and characterize heterogeneities in collagen matrices under cellular forces [27,45], e.g. assessing strain stiffening during cell migration [46]. Recently, using this imaging method, Mohammadi et al. found that fibroblasts seeded on top of a thin collagen gel induce inelastic matrix deformations, which can influence mechanosensing [47]. However, this flat “2D” cell-substrate geometry is non-physiological when considering cancer cells inside the tumor stroma or endothelial cells undergoing angiogenesis or vasculogenesis–scenarios better mimicked by cells embedded inside a 3D ECM. The geometry of the microenvironment has been shown to be critical in regulating cell phenotypes [48–50]. Moreover, confocal reflectance microscopy is a non-specific, label-free technique, which produces signals that are influenced by the presence of cells and fiber orientation, thus hindering accurate quantification of local ECM concentrations [51]. In another study, Nam et al. provided mechanistic, molecular-level insights into the plastic remodeling of biopolymer networks and implicated the unbinding of fiber-fiber bonds in stress relaxation, although the results were not based on direct, physiologically relevant cellular interactions with the ECM [52]. Recently, in a study using instead 3D DVC algorithms on fluorescently-labeled fibrin matrices, Notbohm et al. characterized the remodeling process and inferred that fibroblasts plastically push and pull fibrin to form tubular, protrusion-like permanent structures [53]. Other assessments using fluorescently labeled fibers in experiments have focused on continuum properties and characterizing the viscoplasticity of the ECM in reconstituted and living microtissues [54,55]. While useful and informative, these studies do not provide full assessment of the effects of crosslink concentration and dynamic actin-driven processes on ECM accumulation mechanics and dynamics by cells. Also, while molecular-level mechanistic insights were recognized, a multiscale computational approach linking physiological cellular forces and discrete fiber networks with local force-sensitive crosslink kinetics to a coarse-grained continuum representation of the ECM has not be fully investigated.
Our study takes an integrated computational and experimental approach toward understanding mechanical remodeling in heterogeneous, inelastic physiologically relevant biopolymer ECMs–collagen I and fibrin. We established an experimental strategy to capture sequentially the 1) initial zero-stress configuration, 2) rapid remodeling steps, and 3) final, plastically remodeled, cell stress-free state of cell-ECM systems. We used this strategy to assess the importance of the degree of crosslinking in fibrous ECMs in matrix remodeling. Increasing crosslink density has been shown to produce more than a two-fold increase in shear storage modulus, and a ten-fold increase in single filament stiffness in fibrin gels [35].
By disrupting key steps in intracellular actin dynamics, we found that ECM remodeling could be prevented, suggesting an important role for cell contraction, presumably in combination with filopodial protrusion and adhesion. This leads to a conceptual model in which the cell sends out filopodia via actin polymerization that adhere to individual matrix fibers, then retract, pulling the fibers closer to the cell. These forces then disrupt bonds between or within the matrix fibers, which may potentially form again in a new configuration, leading to matrix remodeling and plastic deformation.
Our findings confirm that mechanical force propagation is enhanced in fibrous ECMs as compared to ideally isotropic matrices. This has been previously attributed to non-linear phenomena such as strain stiffening [24], or local fiber alignment by cell-generated forces–a load-driven anisotropic effect in elastic matrices [25,26]. Here, we further show that such mechanical signals can change dynamically because the matrix is plastically remodeled–crosslink unbinding leads to force relaxation. Note that in our study, it is not possible to directly compare how far a cell can sense in collagen vs. fibrin, because different cell types were seeded in different gels at different concentrations, and we cannot precisely control cell-generated forces exerted on these biopolymers. Moreover, the same cells can have different affinities for different ECMs. However, from the normalization in Fig 2B, we can speculate that differential crosslinking or remodeling would not significantly alter displacement propagation profiles, pointing to the robustness of the connectivity of these networks for the purpose of long-range force transmission and mechanosensing [25]. These results point to the need for more investigations on the possible role of force-induced ECM remodeling in cell mechano-sensing dynamics in 3D physiological environments.
While ECM networks are rich with multiscale features that can influence their behavior, in this study we focus specifically on the roles of transient crosslinks and dynamic filopodial forces on ECM accumulation at the cell periphery. Our experimental and computational results show consistency (Fig 6) and indicate that the interplay of kinetic crosslinks and dynamic loading modulates the amount of accumulation and plastic remodeling of the ECM in the vicinity of the cell. Intuitively, molecular bonds can break under mechanical forces, and various simplified models have been developed to capture the unbinding kinetics of molecular bonds under load, including slip bonds and catch bonds, which have increased and decreased unbinding rates, respectively, when under tension [56]. For simplicity, we model crosslinks as slip bonds following Bell’s law [32]. We note that many types of bonds can exist that link ECM fibers together, such as molecular knob-hole connections that link fibrin fibrils [36], hydrogen bonds, and various intermolecular and polypeptide bonds that can slip and rupture [57]. In our simplified ECM model, we reduce the system to having only one type of bond that connects fibers, and we define that bond as a crosslink. In our simulations, we explore crosslinks with both high and low unbinding rates, thus examining the impact of both highly transient and relatively permanent bonds, providing insights toward the impact of diverse bond types (S5 Fig, S7 Fig). Future, higher resolution experimental studies and higher order, multiscale computational work can help reveal the impact of finer features on global ECM remodeling dynamics. ECM fibers are typically complex, multiscale, hierarchical structures. For example, during gelation, collagen and fibrin molecules polymerize into fibrils that bundle into thicker fibers. Thicker fibers can be linked to other fibers, forming a connected network [58,59]. The bundled fibrils in a fiber can also split, leading to branching [40]. It is possible that mechanical forces can disrupt both inter-fiber and intra-fiber bonds (fibril-fibril bonds within a fiber). It is possible that intra-fiber bonds can lead to additional effects. For example, the unbinding and rebinding of fibril-fibril bonds inside a fiber can lead to intra-fiber sliding and fiber elongation [60]. If unbinding occurs at a branch point, that can lead to the further splitting of a fiber and eventually the peeling off and breakdown of the fiber into thin fibrils. In these cases, we would expect an increase in the ratio of thinner to thicker fibers in the vicinity of the cell during force loading. Furthermore, as intra-fiber bonds unbind, it may be possible in the extreme limit for the fiber to break and fail, resulting in further relaxation mechanisms in the network.
At the coarser, material continuum level, unbinding and elastic failure of crosslinks would translate in (elastic) damage and (plastic) softening, the latter simulating the drop in average material yield stress onset, as fewer crosslinks are present. We show that both might act synergistically and be necessary to model inelastic remodeling of the ECM. Plasticity at the level of collagen fibrils has been tested experimentally and has shown high variability in plastic response, spanning from hardening to softening [61]. Further experimental assessment is needed to separate the effect of elastic damage and the effect of softening at the level of ECM fibrous hydrogels. Both effects could be present as the material is remodeled and fiber networks are re-arranged, with damage assumed to act on the elastic component and softening attributed to the plastic response to yield.
We can infer from our continuum model the effect of such elasto-plastic damaging features on the values of traction forces that one can back-calculate in a 3D experiment using these cell-ECM systems. ECM viscoplasticity has already been recently targeted for accurate descriptions of tissue dynamics [54,62]. We have additionally integrated novel experiments with discrete fiber and continuum simulations to elucidate mechanistic insights toward the dynamic physical state of the ECM during cell-matrix mechanical interactions. For instance, damage parameters can ultimately be linked to the degree of crosslinking in our experiments and discrete simulations. An early damage onset and rapid modulus decay mimic weak network crosslinking and increased force-driven crosslink unbinding, while a late damage onset and more gradual modulus decay simulate strong network crosslinking. We finally relate these experiments and simulations to a continuum reasoning that is very useful for traction force microscopy studies. Although 3D fibrous biopolymer networks have been recently characterized from a material point of view for traction microscopy [27], as well as viscoelastic inverse numerical algorithms been recently introduced for more accurate force computation [63], our study is explicitly targeting inelastic behavior of such networks. We have assessed that most traction forces would be highly overestimated in the presence of elastic damage—linked to filament-level crosslink unbinding—which is thus highlighted here as an important continuum feature that requires future experimental investigations.
Overall, our results suggest that during ECM recruitment, cells do not exhibit a stable tensional state, but rather a highly dynamic one due to relaxation from crosslink unbinding. Thus, for highly motile cells that recruit matrix, including endothelial cells and metastatic cancer cells, as they migrate to and recruit fibers from new locations inside a 3D ECM, their tensional profile is dynamic rather than static. This is a starkly different picture compared to cells on 2D artificial elastic substrates, as those cells tend to exhibit relatively stable stress profiles since the substrates do not undergo remodeling [64]. Many studies have shown that substrate stiffness affects cell tension and this, in turn, affects cell behavior. However, the dynamic state of cell tension, which appears to be characteristic of cells inside a more physiologically relevant environment of a 3D ECM, has not been fully investigated. Our study shows that certain key properties–cell loading forces, ECM crosslink density, and the kinetic nature of the crosslinks–are important in regulating this dynamic tensional state in cells, which can help guide future experiments in systematically tuning these parameters to assess cell behavior.
The physiological microenvironment is often composed of a complex, fibrillar ECM that exhibits non-linear, non-elastic properties. We have demonstrated that dynamic forces generated by the actomyosin machinery are capable of mechanically reassembling the local ECM, leading to substantially increased local ECM density in the course of minutes, which is not fully reversible when the cells are relaxed. Differences in ECM ligand density can alter cell signaling and overall phenotypes [65–67]. The results demonstrated here highlight the dynamics of cell-ECM interactions in a more physiological context. The local environment sensed by cells, both physically and biochemically, is highly distinct from acellular matrices and gels in their initial states, with nominal concentration values based on stock solutions. ECMs with active cells are rapidly remodeled by cells to generate heterogeneous local environments with significantly different ligand densities and architectures. This behavior is often not considered, as only nominal ECM concentrations are usually reported, and is not captured by widely used non-physiological, elastic substrates that cannot be plastically remodeled by cells. Physical properties of the microenvironment have already been shown to lead to diverse ramifications in cell behavior, from guiding stem cell differentiation to modulating tumor dissemination and tissue morphogenesis. Our results directly implicate cell mechanics–the actomyosin machinery and dynamic filopodial or filopodial-like forces–in driving active remodeling of the ECM and the creation of new microenvironments that can dynamically modulate cell behavior.
For fibrin experiments, we culture Human Umbilical Vein Endothelial Cells (HUVEC) (Lonza) on collagen I-coated flasks in EGM-2 (Lonza) growth medium and used in experiments between passages 6–8. For collagen experiments, we culture MDA-MB-231 cells expressing fluorescent actin filaments (via LifeAct, gift from the Lauffenburger Lab at MIT) were cultured at 37°C, 5% CO2 with DMEM supplemented with 10% fetal bovine serum and 1% penicillin-streptomycin.
Both rat tail collagen I, solubilized in 0.02N acetic acid, (Corning) and bovine fibrinogen proteins (Sigma) are fluorescently labeled in stock solutions. A fluorescent reactive dye binding to free amine groups (Alexa Fluor 647 NHS Succinimidyl Ester, ThermoFisher) is used to produce cell-compatible, purified gels that can be visualized in 3D confocal live imaging with no known alterations of functionality of monomers reported in previous mechanobiology studies of the ECM[31,53,60]. Stock solutions are purified from the unreacted dye by using dialysis cassettes (Thermo Fisher) with a 7 kDa molecular weight cut-off. Fluorescently labeled fibrin is then obtained by mixing over ice (i) bovine fibrinogen dissolved in PBS (Lonza) at twice the final concentration (6 mg/mL) and (ii) bovine Thrombin (Sigma), dissolved at 2U/mL in EGM-2 growth medium with HUVECs. Briefly, HUVEC’s are spun down at 1200 rpm for 5 min and the cell pellet is resuspended in EGM-2 growth medium containing the thrombin and mixed with the fibrinogen solution at a 1:1 ratio. The mixture is quickly pipetted into a microfluidic device using the gel filling ports. The device is placed in a humidified enclosure and allowed to polymerize at room temperature for 10 min before fresh growth medium is introduced before the experiment to hydrate the gel.
For the lowering crosslinking in fibrin gels, a synthetic inhibitor of transglutaminase (1,3-dimethyl-4,5-diphenyl-2[2(oxopropyl)thio] imidazolium trifluoromethyl-sulfonate) (DDITS, Zedira) is used. Fibrin gels were polymerized in the absence (high crosslinking) and presence (low crosslinking) of 0.2 mM DDITS. This method has been characterized previously in fibrin gels[35]. Furthermore, it has been shown at the molecular level that in the absence of this inhibitor, there is ligation of the γ chains and α chains of fibrin, which results in an increase in instantaneous bulk stiffness[68,69].
Collagen gels are prepared by mixing acid solubilized type I rat tail collagen with a neutralizing solution (100mM HEPES buffer in 2X phosphate buffered saline at pH 7.3) at a 1:1 ratio and then diluting with 1× PBS and suspended cells in media to a final collagen concentration of 1.5 mg/mL[70]. The final solution is then allowed to gel in a humidified chamber at 37°C and 5% CO2.
Microfluidic devices with gel and media chambers are used because of the convenient fluid flow access for on-stage media and reagent exchange necessary for the experiments. Device design and protocol are described previously[71]. Briefly, 130 μm thick devices were fabricated using PDMS soft lithography. The chambers are 1.3 mm wide and are injected with the gel encapsulating cells. Similarly shaped chambers for media flank these gel chambers and allow the quick washing and re-introduction of small volumes of reagents in all stages of the experimental procedure.
For assessing the effect of cytoskeletal drugs on densification and plasticity, the fluorescently labeled gels are polymerized together with cells treated with each of the tested drugs at the following concentrations: Latrunculin A (Calbiochem, 0.8 μM), GM6001 (Calbiochem, 10 μM), CK-666 (Sigma, 100 μM) and SMIFH2 (Sigma, 50 μM). These working concentrations, for substantial inhibition effects, were taken from literature data [72,73]. Control cases–both untreated and vehicle (1μL/mL dimethyl sulfoxide (DMSO), which is the maximum concentration for drug-treated conditions)–are included in the study. Cells are cultured for 4h with the same concentration of drug in the culture media, fixed with PFA4% and stained (DAPI, phalloidin). In the dynamic force recovery experiments, the gels are polymerized together with cells treated with Cytochalasin D (Santa Cruz Biotechnology, 5 μM), which is an inhibitor of actin polymerization and leads to highly diminished cellular force generation [74]. First, images are captured of cells encapsulated in the 3D ECMs under the action of Cytochalasin D to have a force-free initial configuration. Second, the chambers are washed through the microfluidic channels with fresh media on-stage three times to remove the Cytochalasin D, and to observe the onset of ECM remodeling. During this process, fluorescently labeled fibers are imaged at small time increments and these sequential images are cross-correlated through the Fast Iterative Digital Volume Correlation (FIDVC) algorithm [30] to determine the 3D displacement field while remodeling occurs. After plastic remodeling of the ECM begins to plateau (~4h), a non-ionic detergent (Triton X, 0.1%) that preserves the gel structure while permeabilizing the cell membrane is used to lyse cells. Thus, active cellular forces are fully relaxed, as cells are eliminated from the system, at the final fiber network configuration [19]. To obtain an estimation of the remodeling dynamics without the delay from Cytochalasin D recovery, FIDVC-based displacement estimations are also performed on time-lapse videos of cells right after seeding.
We apply three key metrics to quantify our experimental data: the displacement length, the densification factor (DF), and the recoverability index (RI). The displacement length, ‖u‖, is defined as the spatially averaged displacement magnitude, computed from FIDVC, inside a 60x60x60μm3 ROI containing one cell. DF is calculated as Idens/Ifar, where Idens is the integral of the radial intensity profile calculated within 5 μm from the cell membrane (averaged over 4 profiles per cell) and Ifar is the integral of the intensity profile of a 5 μm line far away from the cell (averaged over 4 profiles per cell). RI is defined as the percent from the ratio between the displacement length caused by decellularization and the displacement length right before decellularization (i.e. overall displacement length), RI = 100×‖udecell‖/‖uoverall‖. Note that RI aims to quantify elastic recoverability, as decellularization is expected to cause elastic relaxation, ‖udecell‖~‖uelast.‖, and the overall displacement contains both elastic and non-elastic deformations, i.e. ‖uoverall‖~‖uelast.+non−elast.‖.
All confocal images in Figs 1 and 2 and related quantifications were acquired with a 20× objective and an Olympus IX81 microscope (Olympus America, Inc.). Images in Fig 3 and related quantifications were acquired with a 60× or 63× oil-immersion objective using a spinning disk confocal microscope (Yokogawa) or a spectral confocal microscope (SPE, Leica microsystems).
|
10.1371/journal.pntd.0000924 | Subversion of Innate Defenses by the Interplay between DENV and Pre-Existing Enhancing Antibodies: TLRs Signaling Collapse | The phenomenon of antibody dependent enhancement as a major determinant that exacerbates disease severity in DENV infections is well accepted. While the detailed mechanism of antibody enhanced disease severity is unclear, evidence suggests that it is associated with both increased DENV infectivity and suppression of the type I IFN and pro-inflammatory cytokine responses. Therefore, it is imperative for us to understand the intracellular mechanisms altered during ADE infection to decipher the mechanism of severe pathogenesis.
In this present work, qRT-PCR, immunoblotting and gene array analysis were conducted to determine whether DENV-antibody complex infection exerts a suppressive effect on the expression and/or function of the pathogen recognition patterns, focusing on the TLR-signaling pathway. We show here that FcγRI and FcγRIIa synergistically facilitated entry of DENV-antibody complexes into monocytic THP-1 cells. Ligation between DENV-antibody complexes and FcR not only down regulated TLRs gene expression but also up regulated SARM, TANK, and negative regulators of the NF-κB pathway, resulting in suppression of innate responses but increased viral production. These results were confirmed by blocking with anti-FcγRI or anti-FcγRIIa antibodies which reduced viral production, up-regulated IFN-β synthesis, and increased gene expression in the TLR-dependent signaling pathway. The negative impact of DENV-ADE infection on the TLR-dependent pathway was strongly supported by gene array screening which revealed that both MyD88-dependent and –independent signaling molecules were down regulated during DENV-ADE infection. Importantly, the same phenomenon was seen in PBMC of secondary DHF/DSS patients but not in PBMC of DF patients.
Our present work demonstrates the mechanism by which DENV uses pre-existing immune mediators to defeat the principal activating pathway of innate defense resulting in suppression of an array of innate immune responses. Interestingly, this phenomenon specifically occurred during the severe form of DENV infection but not in the mild form of disease.
| Dengue is the most common vector-borne viral disease in humans, with 50–100 million infections per year. The severity of dengue ranges from an acute febrile illness, DF, to a life-threatening vascular leakage syndrome with or without shock, DHF/DSS. Determinants of these syndromes are mainly host factors including non protective but cross reactive antibodies which are known as preexisting enhancing antibodies. These antibodies enhance disease severity through increasing the virus infected cell mass and facilitating intracellular virus replication. Here we demonstrate that DENV exploits preexisting subneutralizing antibodies to defeat the pathogen recognition system and to down regulate the TLR signaling pathway resulting in suppression of an array of innate immune responses. Furthermore, we also show that this phenomenon specifically occurs in the severe form of dengue but not in the mild form of disease.
| Dengue is the most prevalent vector-borne disease occurring in tropical and subtropical regions with an estimated 50 to 100 million people infected each year. This includes 500,000 cases of life-threatening disease which are dengue hemorrhagic fever (DHF) and dengue shock syndrome (DSS) [1], [2]. Dengue viruses, members of family Flaviviridae, are a group of four genetically distinct serotypes known as DEN-1 to -4. The genome of these viruses is a single-stranded positive sense RNA which is approximately 11 kb in length and encodes for three structural (C, prM, E) and seven non-structural (NS1, NS2A, NS2B, NS3, NS4A, NS4B, NS5) proteins [3].
Infection with dengue virus (DENV) causes two clinically distinct syndromes which are dengue fever, a mild form of the disease and DHF/DSS, a life-threatening disease. The pathophysiology of DHF/DSS development is of interest among researchers since the incidence of DHF/DSS is 25–80 times higher in people previously exposed to DENV than in DENV-naïve individuals, indicating the significance of pre-existing immune mediators such as aberrant T cells, cytokine storms and the enhancing activity of the subneutralizing antibodies [4], [5], [6]. Among these mediators, the presence of enhancing antibodies stand out the most because it is the only risk factor that can explain DHF/DSS development in primary infected infants due to the finding that the peak incidence of DHF/DSS development in infected infants correlates with the decline of maternally derived protective antibodies to non-protective or subneutralizing levels. Moreover, these infants do not experience DHF/DSS accompanying a primary dengue virus infection after the maternally derived antibodies have completely disappeared [7], [8], [9], [10]. This epidemiological data is supported by in vitro enhancement infection experiments in which neat plasma from healthy infants born to dengue-immune mothers enhances dengue virus infection in a manner that correlates with the age-related DHF/DSS development in infants [11], [12]. To further support the role of subneutralizing antibodies, investigators have been able to mimic this phenomenon in a mouse model and in rhesus monkeys [13], [14]. The role of enhancing antibodies in exacerbating disease severity has been reported in other types of infection. For example, Leishmania is known to exploit host IgGs facilitating the entry of Leishmania amastigotes into macrophages. The entry of amastigotes-antibody complexes via Fcγ receptor ligation does not only allow the numerous parasites to penetrate into macrophages but also suppress the development of cell-mediated immunity resulting in progressive non-healing leishmaniasis in mice [15], [16].
The mechanism by which enhancing antibodies exacerbate dengue disease severity has not been fully established. However, severe dengue is associated with high levels of circulating DENV, and enhancing antibodies have been proposed to facilitate DENV production by at least two mechanisms. Firstly, enhancing antibodies function as a bridge between infectious DENV particles and FcR on cell surfaces resulting in an increased number of infected cells [17], [18]. Interestingly, Rodenhuis-Zybert et al. recently demonstrated that enhancing antibodies not only promote entry of the mature DENV but also assists entry of non-infectious virions or immature DENV particles into FcγR bearing cells [19]. Once inside the target cells, these immature viruses replicate effectively. This phenomenon, if occurring in natural dengue virus infections, could significantly contribute to disease severity.
The second mechanism proposed is one in which infection via Fc and FcR ligation switches the intracellular response from an antiviral mode into an immune suppressive mode [20]. This suppression mediates the severity of the secondary dengue virus infection. Thus, to gain more information on the intrinsic role of enhancing antibodies, we further investigated the mechanism of immune evasion induced by DENV-ADE infection.
Once attacked by viruses, host cells immediately recognize the invaders using several types of sensing systems [21]. One of these systems is the Toll-like receptors or TLRs pathway, and six TLRs have been reported to recognize viral invaders. For example, the extracellular TLR-2 and TLR-4 detect viral particles/viral proteins on the cell surface, while the endosomal TLRs recognize viral nucleic acid components such as dsRNA, ssRNA and unmethylated DNA with a CpG motif [22]. Upon ligation to the invader, TLRs trigger a signaling cascade through the recruitment of a set of TIR-domain-containing adaptors including MyD88, TIRAP (MAL), TRIF (TICAM) and TRAM (TICAM2). Based on the MyD88 molecule, the TLR signaling cascade can be divided into two principle pathways, the MyD88-dependent and MyD88-independent (or TRIF-dependent pathway) signaling pathways. While most TLRs trigger the MyD88-dependent signaling pathway via the TIR-containing cytosolic adaptor MyD88, TLR-3 and TLR-4 initiate their signals through TRIF activation [23]. Both MyD88-dependent and TRIF-dependent signaling pathways can activate type I IFN and inflammatory cytokines via NF-κB and the IRFs family [24], [25].
Activation of the TLRs signaling pathway in response to viral infection has been intensively studied [26], [27], [28], [29]. For example, the response against hepatitis C virus infection is mediated by the TLR2 and TLR3 signaling [30], [31], while West Nile Virus (WNV) can be recognized by TLR-3, eliciting an antiviral response shaping innate as well as adaptive immunity in in vivo experiments [32], [33]. TLR-3 and TLR-7 have been reported to play important roles in inhibiting dengue virus infection in U937 and HEK293 cells, respectively [34], [35].
The present study investigated the effect of DENV-antibody complex infection on TLR-dependent signaling in a monocytic cell line. The experiments were conducted in vitro and ex vivo, meaning that infected THP-1 cells and PBMCs from infected patients were used, respectively. This is the first study to show the negative effect of enhancing antibodies on the expression and function of the antigen recognition pathway in human monocytic cells. Results showed that preexisting subneutralizing antibodies were able to ligate infectious DENV particles to both FcγRI and FcγRIIa. Upon ligation, activation of TLR-negative regulators and down-regulation of membrane as well as cytoplasmic TLRs was pronounced, resulting in suppression of TLR-dependent immune activation. These results were also found in secondary DHF PMBC but not in secondary infection DF PBMC.
The protocol for patient enrollment and sample collection is approved by The Committee on Human Rights Related to Human Experimentation, Mahidol University, Bangkok, Thailand. Dengue-infected patients, which hospitalized at Queen Sirikit National Institute of Child Health, Bangkok, Thailand, were enrolled to the study after the parents/guardians have giving written informed consent. All clinical investigation must have been conducted according to the principles expressed in the Declaration of Helsinki.
The enrolled patients were 5–10 years of age. Blood samples were obtained twice, once on the day of admission (fever day) and the other on 30 days after admission (convalescence day). Plasma and PBMCs were separated immediately and were kept frozen at −80°C until required. The patient's disease severity was graded as DF or DHF according to WHO criteria. All enrolled cases were classified as secondary infection by haemagglutination inhibition (HI) titre and IgM ELISA assay [36].
Convalescent serum from a patient infected with DENV serotype 3 (DENV-3) at a 1∶ 10,000 dilution was used in all DENV-ADE infection experiments [38].
Antibody-dependent enhanced infection of DENV-2 16681 into THP-1 cells was conducted as described [38]. Briefly, THP-1 cells were infected with a complex between DENV-2 16681 and the enhancing antibodies at the MOI of 0.01 pfu/cell. After an hour of incubation at 37°C, cells were washed and were further cultured in growth medium. The infected cells and supernatants were harvested at 3, 6, 12, 18, 24 hours and every 24 h for 3 consecutive days. In this experiment, sets of control were performed which were THP-1 cultures infected with DENV at the MOI of 5.0 and 10.0 pfu/cell, THP-1 cells infected with UV-treated-DENV-Ab complexes and the mock-infected THP-1 cells.
THP-1 cells were pre-incubated at 37°C for 90 minutes with either an anti-human FcγRI MAb or an anti-human FcγRIIa MAb (R&D System, Inc., Minneaspolis, MN) or both antibodies, at a concentration of 5 µg/ml of each antibody. After incubation, cells were washed with IMDM before being infected with DENV or DENV-antibody complexes.
RNA was extracted from supernatants using TRIZOL (Invitrogen, CA, USA). The purified viral RNA was then monitored by real time RT-PCR using QuantiTect Probe RT-PCR (Qiagen, Germany) as described by Houng et al [39]. Real time RT-PCR amplification, data collection and analysis were performed using a Roter-Gene™ 3000 (Corbett Research, Sydney, Australia). The RNA copy number was calculated using dengue serotype-specific copy standards.
Total RNA was extracted from infected cells using the QIAgen RNAeasy Kit (QIAgen, Germany). Biotin-UTP labeled cRNA probes were synthesized using 3 µg of cDNA amplified from the total RNA template (Superarray Inc., Frederrick, MD, USA). The labeled cRNA (6 µg) from each sample was then incubated with Oligo GEArray Human Toll-Like Receptor Signaling pathway (OHS-018.2) membranes containing 113 TLR-related genes. The hybridized membranes were washed and hybridization signals were detected using a chemiluminescent system according to the manufacturer's instruction. The intensity of hybridization was determined by ImageMaster TotalLab version 2.00 (Amersham Pharmacia, England) and was acquired in TIFF format. The digital TIFF image files were then analyzed by ClonTech AtlasImage software, version 2.7 (CloneTech, CA, USA). The background was automatically subtracted and standardization of all the signals was performed by normalizing the raw data with β2-microglobulin (β2-m) and Glyceraldehyde 3-phosphate dehydrogenase (GADPH). The correlation between two data sets was tested using Pearson's correlation with P-value<0.05. Two-fold and 0.5 fold difference in expression between normalized gene intensities compared between control, DENV, and DENV-ADE samples were considered as significant up-regulation and down-regulation, respectively.
The qRT-PCR was used to investigate levels of gene expression. In brief, RNA was extracted from harvested cells using the QIAgen RNAeasy Kit (QIAgen, Germany) and then subjected to first-strand cDNA synthesis before amplification by qRT-PCR using specific primers. The primer sets for TLR-3, TLR-4, TLR-7, TRIF, TRAF-6, TRAM, IRAK-4, and ACTIN are : TLR-3: forward,5′-AGG AAC TCC TTT GCC TTG GT-3′ ; reverse, 5′ – TTT CCA GAG CCG TGC TAA GT- 3′; TLR-4: forward, 5′ –TGG ATA CGT TTC CTT ATA AG- 3′ ; reverse, 5′ –CAA GTA CAA GCA AAG TCA TTC- 3′ ; TLR-7: forward, 5′ –CCT GGA AAC TTT GGA CCT CA- 3′ ; reverse, 5′-CCA CCA GAC AAA CCA CAC AG- 3′ ; TRIF: forward, 5′ – CCC TGT GGA CAG TGG AAG AT- 3′ ; reverse, 5′ –CAA GAC CCT TCA CCC AGA AA- 3′; TRAF-6: forward, 5′-GTT GCT GAA ATC GAA GCA CA- 3′; reverse, 5′ –CGG GTT TGC CAG TGT AGA AT- 3′ ; TRAM: forward, 5′ – GGG TGA TGT TCG TGT CTG TG- 3′ ; reverse, 5′ –ACT GAG GCG CTG AGG TAA AA- 3′ ; IRAK-4: forward, 5′ –CCT TTG CCT TCC ATT GTG AT- 3′ ; reverse, 5′ –GTT TTG GCT TAC GGT TCT GC- 3′; SARM: forward, 5′ –TTG CCA AGC AGC AAT GTT AG- 3′ ; reverse, 5′ –TCT CCT CCC AAC CAG AAA TG- 3′; TANK: forward, 5′ –CAG GCA TGC ATG GAT AGA GA- 3′ ; reverse, 5′ –TTC AAG CAG AGG AAC ACA GC- 3′; Beta-actin: forward, 5′- CCT GGC ACC CAG CAC AAT-3′ ; reverse, 5′GGG CCG GAC TCG TCA TAC- 3′ The qRT-PCR was carried out using the SYBR system (Invitrogen, Oregon, USA), using actin as a control.
Levels of SARM, TRAF6, IRAK4, TLR7, IKK-α, and Rel-A protein production were semi-quantitated using immunoblotting. The intensity of each specific protein was detected using monoclonal antibodies as previously described [38].
Level of IFN-β production was quantitated using a PBL Medical Laboratories Kit (Piscataway, New Jersey, USA) according to the manufacturer's protocol.
Values were expressed as mean ± standard deviation (SD) of at least three independent observations. Statistical significance was tested by Student's t-test, One-way ANOVA, as indicated in the legend of figure. P-values<0.05 were considered significant.
Two types of FcR, FcγRI (CD64) and FcγRIIa (CD32), have been reported by several investigators to participate in the entry of DENV-antibody complexes in in vitro systems [17], [40], [41], [42]. In the present study, the synergistic role of FcγRI and FcγRIIa in DENV-ADE infection in THP-1 cells was investigated. THP-1 cells were pretreated with either anti-FcγRI or anti-FcγRIIa antibodies or both before being infected with either DENV or DENV- enhancing antibody complexes. The level of viral production was monitored by assaying RNA copy number and the number of infectious virions using real-time RT-PCR and plaque assay, respectively. As demonstrated in Fig. 1a–b, blocking of FcγRI or FcγRIIa significantly suppressed viral production in DENV-ADE infection of THP-1 cells. The largest reduction in viral production was found in cells pre-treated with both anti-FcγRI and anti-FcγRIIa antibodies, suggesting that FcγRI and FcγRIIa synergistically mediate the entry of DENV-enhancing antibody complexes into THP-1 cells.
Our previous report demonstrated that DENV-ADE infection significantly suppresses IFN-β production in THP-1 cells (20). Therefore, the level of IFN-β production was used as a marker to test the synergistic role of FcγRI and FcγRIIa on DENV-ADE infection. THP-1 cells were pretreated with anti-FcγRI and anti-FcγRIIa before being infected with DENV-enhancing antibody complexes, and the production of IFN-β was assayed at 24 hr of infection using ELISA. As shown in Fig. 1c, blocking of ADE-infection via FcγRI and FcγRIIa completely restored IFN-β production. Taken together, these results show that DENV-enhancing antibody complexes use both FcγRI and FcγRIIa for entry.
We previously reported that one of the intrinsic roles of ADE-infection is suppression of type I interferon via the RIG-I/MDA-5 signaling pathway [20]. Since type I interferon production is also activated via the TLR pathogen recognition pathway [43], we therefore investigated whether DENV-ADE infection has any effect on TLR expression and/or the TLR-dependent signaling pathway. To answer this question, THP-1 cells were infected with either DENV alone or infected with DENV- enhancing antibody complexes. The expression of a surface membrane TLR (TLR-4), endosomal TLRs (TLR-3, TLR-7) and TLR-signaling molecules (TRIF, TRAF-6 and IRAK4) were monitored using qRT-PCR and immunoblotting. As illustrated in Fig. 2, THP-1 cells infected with DENV-2 significantly stimulated TLR-3, TLR-4, TLR-7, TRIF and TRAF-6 expression. This data correlated with the level of IFN-β production as shown in Fig. 1c. In contrast, DENV- enhancing antibody complex infection significantly suppressed TLR and TLR-signaling molecules in comparison to infection by DENV alone. This data is supported by anti-FcγRI and anti-FcγRIIa treatment in which blocking of DENV-ADE infection via these two receptors restored the expression of TLR(s) and TLR-signaling molecules. This data is also supported by the increased IFN-β production as shown in Fig. 1c. Collectively, DENV-ADE infection could interfere with TLR-dependent signaling via FcγRI and FcγRIIa ligation which corresponded to the reduction of IFN-β production. In addition, pre-treatment with anti-FcγRIIa antibodies restored a higher level of TLRs and TLR-signaling molecules than pretreatment with FcγRI (Fig. 2).
To ensure that phenomenon found in this study is not due to the effect of higher level of viruses produced during ADE-infection mode, control experiments were preformed. THP-1 cultures were infected with DENV at the MOI of 5.0 and 10.0 pfu/cell, or with UV-treated-DENV-enhancing Ab complexes, or were mock infected. THP-1 cultures infected with the MOI of 10.0 replicated DENV to the same level as ADE-infected mode. In contrast, infection by higher MOI of DENV activates stronger TLR-3 and -4 expressions (Figure S1). THP-1 cells infected with UV-DENV-Ab complexes revealed no suppressive effect on IFN-β production (data not shown).
Taken together, this information indicated that suppression of TLRs and TLR signaling pathway demonstrated in our study is due to the infectious immune complexes infection.
Suppression of the TLR-dependent signaling pathway may due to down-regulation of TLR synthesis and/or blocking of TLR-signals. Unfortunately, negative regulators of TLR synthesis have not yet been identified. Thus we investigated whether or not down-regulation of the TLR-dependent signaling pathway is due to DENV-ADE infection activating negative regulators of TLRs signaling such as SARM and TANK and so the levels of SARM and TANK gene expression were investigated. As shown in Fig. 3, expression levels of SARM and TANK were significantly increased at 3 hr post DENV-ADE infection, but not in DENV infection.
Since THP-1 is a monocytic cell line it may not be an ideal physiological model of the natural response during DENV infection. Therefore, to confirm the phenomenon found in THP-1 cells, expression levels of TLR-3, TLR-4, TLR-7 and TRAF-6 in PBMCs obtained from secondary DF and secondary DHF patients, on fever and convalescent days were determined by qRT-PCR. Interestingly, expression of these genes was significantly down-regulated in secondary DHF patients but not in DF patients (Fig. 4), suggesting that the TLR-dependent signaling pathway is activated during the mild form of DENV-infection but not in the severe form of the infection.
To further elucidate the impact of ADE-infection on TLR-dependent signaling, a TLR-specific oligonucleotide array analysis was conducted to differentiate responses between DENV infection and infection by DENV-enhancing antibody complexes. As shown in Table 1, the expression of 27 out of 113 TLR-related genes was significantly altered during DENV-ADE infection. These genes were categorized on the basis of their functions as TLR, TIR containing adaptor molecules, effector molecules, NF-κB associated molecules, JNK/p38 pathway, IRF pathway, IFN-inducible genes and others. Among these 27 genes, 21 and 6 genes were down- and up-regulated during DENV-ADE infection, respectively. Most of the down-regulated genes are associated with both the MyD88-dependent and -independent signaling pathways such as TLR-4, TIRAP, IRAK-2, IRAK-4, TRIF (TICAM1) and TRAM (TICAM2). This data indicates that both MyD88-dependent and -independent pathways were suppressed. The expression of TLR-3 and -7 were not included in this table since their expression were suppression less than two folds, 1.7 and 1.6 folds, respectively. Expression of the other types of TLR was undetectable in our array analysis.
The suppressive effect was also strongly seen in the NF-κB signaling pathway, in which genes required for IκB degradation and NF-κB activation such as MAP3K7IP1 (TAB1), MAP3K7IP2 (TAB2), TRAF-6, UBE2N and MAP3K7 were down-regulated. This observation was confirmed by the reduction of NF-κB signaling molecules including the REL complex, NF-κB2, CHUK (IKK-α) and MAP4K4 while NF-κBIE, an inhibitor of NF-κB, was up-regulated. In addition, suppression of the IRF pathway and IFN-inducible genes was also pronounced during ADE-infection. In addition, the expression of SARM gene was 1.5 folds increased while the activation of TANK gene was undetectable during ADE-infection.
Taken together, these data imply that DENV-ADE infection may activate host negative regulators which in turn down regulate the MyD88-dependent, MyD88–independent and NF-κB signaling pathway, supporting the in vitro and ex vivo experiments described above.
To validate data obtained from the oligonucleotide array analysis, qRT-PCR was used to monitor the expression of 3 genes including of TICAM2, TIRAP and IRAK-4 at 3, 6, 12, 18, 24 hours post inoculation. The observed copy numbers of these representative genes are shown in Fig. 5a–c. The levels of expression of these genes were significantly down-regulated in THP-1 cells infected with DENV-ADE infection meaning that data from qRT-PCR confirmed the cDNA analysis. In addition, the protein levels of IKK-α and Rel-A were determined using specific monoclonal antibodies. As shown in fig 5 d–e, degradation of phosphorylated IKK-α and suppression of Rel-A production was significant in DENV-ADE infection mode suggesting that immune complexes infection suppesses NF-κB pathway.
Even though the presence of antibodies that enhance dengue viral infectivity has been known since 1977 [44], the mechanism(s) as to how these antibodies increases viral infectivity and exacerbate disease severity is only just being understood. Our reports and others show that enhancing antibodies is not only facilitating virus entry, but also alter the intracellular responses and that the synergism between the extrinsic and the intrinsic roles of enhancing antibodies significantly increases viral burst size and the total virus yield.
The first mechanism by which antibodies enhance DENV infectivity occurs at the plasma membrane. In this process, enhancing antibodies facilitate the interaction between virus particles and the FcR on target cells. This event gives rise to a higher chance of virus penetration resulting in a greater number of infected cells. The significance of Fc and FcR ligation on ADE-infection has been confirmed using genetically engineered antibody variants which can not bind to FcR [45]. These engineered antibodies abrogate ADE-infection and protect mice from ADE-induced lethal challenge. The types of FcR involved in DENV-antibody complex infection have been investigated intensively by several groups of investigators as well as by our group and all agree that both FcγRI and FcγRIIa facilitate ADE-infection in natural DENV target cells and in DENV susceptible cell lines [12], [16], [20], [41]. Moreover, we observed that FcγRIIa enhanced the infectivity in THP-1 cells more efficiently than FcγRI did (Fig 1a and Fig. 2). Our finding is supported by the previous report using FcγR transfected COS-7 cells in which FcγRIIa enhances dengue virus immune complex infectivity more efficient than FcγRI. This difference may due to mode of virus-immune complex internalization mediated by these two types of FcR [42]. In the other experimental system, engatement of immune-complexes to FcγRI signals through γ-chain to initiate proinflammatory cytokines production and to transport the complexes to MHC-II mediated antigen presentation while interaction between immune-complexes and FcγRIIa impaires proinflammatory cytokine production and antigen presentation [46]. Whether this phenomenon can be applied to DENV-immune complex infection remains unclear.
Investigation of the intrinsic role of enhancing antibodies has pointed toward suppression of the innate immune response in which type I interferon and proinflammatory cytokine production are revealed as the main targets [38] and the mechanism of suppression is partly due to ADE infection up regulating negative regulators of the RIG-I/MDA5 signaling pathway [20]. In the present work, we expanded our investigation horizontally to another type I interferon stimulating pathway, the TLR-signaling pathway. Toll-like receptors, some of the most important pattern recognition receptors, are abundant on monocytes/macrophages and dendritic cells, the main in vivo target cells for DENV, and TLRs are key players in priming innate responses upon viral infection. They detect invaders and trigger antiviral defenses, interferon and pro-inflammatory cytokines. Interferon then exerts an antiviral activity through activating the JAK/STAT signaling pathway resulting in the activation of interferon stimulated genes which subsequently inhibit viruses by a non-cytolytic mechanism. In turn, invaders can circumvent the interferon response to be able to propagate in the host cell. DNA viruses including hepatitis B virus (HBV) use their envelop and non-envelope proteins to suppress TLRs expression as well as to inhibit responses elicited by TLRs stimulation [47], [48], [49]. The vaccinia virus uses the A46R and A52R proteins to inhibit TLR-signaling molecules such as TRIF, TRAM and IRAK-2 resulting in ablation of type I IFN production [50], [51]. Respiratory syncytial virus (RSV) strain A2 and Measles virus (MeV), a member of the Paramyxoviridae family, can antagonize TLR-7 and TLR-9 induced type I IFN and proinflammatory cytokine production in epithelial cells, hematopoietic cells (T lymphocytes, B lymphocytes, monocytes) and pDC [52]. Similar to A46R and A52R of the vaccinia virus, the NS3-4A heterodimer of Hepatitis C virus inhibits the TLR-3 mediated antiviral response by degrading TRIF while NS5A has been reported to bind directly to MyD88 leading to inhibition of the MyD88-dependent signaling pathway [53], [54]. Moreover, the entire genome of hepatitis C virus has also been found to suppress TLR-3, -4 and -7 in HepG-2 cells [55]. Similar events are also reported during DENV infection. DENV use nonstructural proteins to block phosphorylation and to down-regulate expression of major components of the JAK/STAT pathway causing reduced activation of IFNα/β stimulating genes [56], [57]. All of the antagonists mentioned above are viruses or viral products. However, high-jacking of pre-existing host immune factors by viruses to interfere with the TLR-dependent signaling pathway has not been reported. We are the first group that has been able to show that DENV exploits pre-existing subneutralizing antibodies to defeat the TLRs system. Upon engagement between FcR and DENV-antibody complexes or entry of DENV into monocytic cells via FcR, expression of TLR-3,-4, -7 and TLR signaling molecules were dramatically decreased in parallel to the decreased production of IFN-β. This observation was further confirmed in experiments that showed that production of IFN-β and expression of TLRs were restored when ADE-infected cells were pretreated with anti-FcR antibodies. This data indicates that entry of DENV via FcR preferentially switches off the TLR-dependent IFN stimulating pathway. The switch off mechanism was mediated at the TLRs gene expression level and through activation of the negative signaling regulators, TANK and SARM (Fig. 6). Unfortunately, the events occurring upstream of TLRs expression and of SARM and TANK activation are unknown, and therefore require further investigation. However, Kurane and colleagues have demonstrated that functional ITAM is essential for ADE infection [58].
The events shown in Fig. 6 are well supported by the array analysis in which ADE-infection suppressed TLR gene expression and down-regulated the TLR-signaling cascade while several negative regulators of TLR-cascade were up-regulated. Importantly, this phenomenon was also found in natural DENV infection in which TLRs (TLR-3, -4, and -7) and TRAF6 were strongly suppressed in PBMC from secondary DHF patients but not in PBMC of mild disease, secondary DF patients. Taken together, the data obtained from in vitro as well as ex vivo studies indicate a significant collapse of the TLR-dependent signaling pathway during DENV-enhancing antibody complex infection.
In conclusion, the present study and our previous report on the suppression of TLR-signaling during DENV-ADE infection of THP-1 human monocytic cells clearly show that initiation of infection by DENV-enhancing antibody complexes defeats the major pathogen recognition pattern pathway resulting in suppression of innate antiviral responses. How dengue immune complexes can have such broad effects on cells is not clear. FcRs are well known in their roles in regulating a multitude of innate and adaptive immune responses. After crosslinking by immune complexes, ITAM initiates either negative or positive signals through several types of adaptor molecules such as Syk/ZAP family PTKs, Src family kinase and SHIP-1, SHP-1 etc. [59], [60].The inhibitory activities of SHIP-1, SHP-1 and Lyn/P13k can be can be seen on multiple signaling pathways including TLRs [61], [62]. Even though direct role of these adaptors on RIGI/MDA5 remain unclear but TLRa and RIGI/MDA5 pathways crosstalk at several steps, thus, the negative effect against TLRs possibly block RIGI/MDA5 pathway. Finally, DENV immune complexes formed with neutralizing or partially neutralizing antibodies fail to suppress innate immunity but permit limited infection of monocyte/macrophage resulting in mild disease is crucial problem requiring further study.
|
10.1371/journal.pcbi.1000384 | Minimum Criteria for DNA Damage-Induced Phase Advances in Circadian Rhythms | Robust oscillatory behaviors are common features of circadian and cell cycle rhythms. These cyclic processes, however, behave distinctively in terms of their periods and phases in response to external influences such as light, temperature, nutrients, etc. Nevertheless, several links have been found between these two oscillators. Cell division cycles gated by the circadian clock have been observed since the late 1950s. On the other hand, ionizing radiation (IR) treatments cause cells to undergo a DNA damage response, which leads to phase shifts (mostly advances) in circadian rhythms. Circadian gating of the cell cycle can be attributed to the cell cycle inhibitor kinase Wee1 (which is regulated by the heterodimeric circadian clock transcription factor, BMAL1/CLK), and possibly in conjunction with other cell cycle components that are known to be regulated by the circadian clock (i.e., c-Myc and cyclin D1). It has also been shown that DNA damage-induced activation of the cell cycle regulator, Chk2, leads to phosphorylation and destruction of a circadian clock component (i.e., PER1 in Mus or FRQ in Neurospora crassa). However, the molecular mechanism underlying how DNA damage causes predominantly phase advances in the circadian clock remains unknown. In order to address this question, we employ mathematical modeling to simulate different phase response curves (PRCs) from either dexamethasone (Dex) or IR treatment experiments. Dex is known to synchronize circadian rhythms in cell culture and may generate both phase advances and delays. We observe unique phase responses with minimum delays of the circadian clock upon DNA damage when two criteria are met: (1) existence of an autocatalytic positive feedback mechanism in addition to the time-delayed negative feedback loop in the clock system and (2) Chk2-dependent phosphorylation and degradation of PERs that are not bound to BMAL1/CLK.
| Molecular components and mechanisms that connect cell cycle and circadian rhythms are important for the well-being of an organism. Cell cycle machinery regulates the progress of cell growth and division while the circadian rhythm network generates an ∼24 h time-keeping mechanism that regulates the daily processes of an organism (i.e. metabolism, bowel movements, body temperature, etc.). It is observed that cell divisions usually occur during a certain time window of a day, which indicated that there are circadian-gated cell divisions. Moreover, it's been shown that mice are more prone to develop cancer when certain clock genes are mutated resulting in an arrhythmic clock. Recently, a cell cycle checkpoint regulator, Chk2, was identified as a component that influences a core clock component and creates mostly phase advances (i.e., jet lags due to traveling east) in circadian rhythms upon DNA damage. This phase response with minimum delays is an unexpected result, and the molecular mechanism behind this phenomenon remains unknown. Our computational analyses of a mathematical model reveal two molecular criteria that account for the experimentally observed phase responses of the circadian clock upon DNA damage. These results demonstrate how circadian clock regulation by cell cycle checkpoint controllers provides another layer of complexity for efficient DNA damage responses.
| Circadian rhythms are periodic physiological events that recur about every 24 hours. The importance of circadian rhythms is well recognized in many different organisms' survival as well as in human physiology. Misregulations in circadian rhythms may lead to different conditions such as depression, familial advanced sleep phase syndrome (FASPS), delayed sleep phase syndrome (DSPS), or insomnia, which largely impact our society [1],[2]. Recent studies indicate higher incidents of cancer in clock defective individuals [3],[4] and chronic jet-lag is associated with higher mortality rate in aged mice as well as faster growth of tumor [5],[6]
The molecular mechanism of circadian rhythms began to become clear beginning with the discovery of the period (per) gene in Drosophila melanogaster in 1971 [7], and the frequency (frq) gene in Neurospora crassa in 1973 [8]. Through analysis of the genetic variants of these genes, pieces of the clock's mechanism could be described. The consensus idea is that it involves interlocked feedback loops largely based on a transcription-translation related time-delayed negative feedback loop [9]. Most of the genes encoding proteins involved in the mechanism of circadian rhythms have been found simply by screens aimed at cataloging the components or by analysis of the regulation of the components. Several studies of mathematical modeling and systems approaches helped further understanding of circadian rhythms in various organisms [10]–[14].
One of the defining properties of circadian rhythms is the ability to phase shift upon a stimulus from external cues. This property allows organisms to adapt efficiently to the external environment. For example, a person traveling east to Europe from the U.S. will experience a jet-lag in the process to adapt advanced phase. Even a brief pulse of light may cause phase advances or delays depending on the timing and influence of the pulse [15]. It is intuitive to assume that a phase shifting agent will create both phase advances and delays depending on the timing and strength of the pulse by uniformly affecting molecular pathways in the circadian system [16]. It has been observed that 2 h treatments of Rat-1 fibroblasts with dexamethasone (Dex) result in large advances and delays (Type 0 resetting of the phase), possibly by inducing transcription of both rPer1 and rPer2 [17],[18]. This Dex-dependent PRC is also observed in the NIH3T3-Bmal1-Luc-1 cells [19]. If the Dex-dependent induction of Per transcripts causes both phase advances and delays, we would also predict that DNA damage-dependent phosphorylation and degradation of PERs by Chk2 [20],[21] would result in similar PRCs. Recent findings indicate that this prediction is wrong [18],[21]. Upon experiencing DNA damage, the cell cycle machinery influences the circadian clock in such a way that creates predominantly phase advances in Rat-1 fibroblasts and mice [18], as well as in Neurospora crassa [21]. These data strongly suggest that there is a conserved pathway across different species that affects the phase of the clock after DNA damage, and involves physical interactions of ATM and/or Chk2 with a core clock component (i.e. PER1 or FRQ) [18],[20],[21]. This interaction leads to phosphorylation of PER1 and FRQ [21],[22]. The molecular mechanism for this unique phenomenon, however, remains unexplained.
In this paper, we explore the minimum criteria in the molecular network of circadian rhythms that simulate the above PRCs with tools of computational modeling. Theoretically, a time-delayed negative feedback is sufficient to create robust oscillations. Both cell cycle and circadian rhythms, however, contain both negative and positive feedbacks in their wiring networks. Positive feedback mechanisms are essential for proper eukaryotic cell divisions [23] whereas their roles in circadian rhythms remain elusive. Recently, Tsai and colleagues indicated that a general function of positive feedbacks in different networks is to create tunable robustness in the system [24]. In our study, we address two questions 1) what is a molecular mechanism that accounts for Chk2-dependent PRC in circadian rhythms?, and 2) is the positive feedback mechanism necessary for the observed PRC? In the conditions that we have tested, we discovered that we can only simulate the Chk2-dependent PRC with predominantly phase advances when Chk2 only affects PERs that are not bound to BMAL1/CLK in the presence of an autocatalytic positive feedback mechanism. Both conditions are required for proper simulations. Our study is the only in silico experiment to indicate the necessity of an autocatalytic positive feedback mechanism in simulating specific phenotype in the circadian system.
We explored our simple mammalian circadian clock model (Fig. 1) from our previous work [25] to investigate whether we can simulate different PRCs from the Dex and IR treatment experiments [17],[18]. Note that an autocatalytic positive feedback mechanism is already embedded in our model [12],[26]. Based on the experimental data, we added the following in our previous model: 1) Dex increases the transcripts of Per but not Bmal1 [18], and 2) Chk2 phosphorylates PERs and facilitates their degradation upon DNA damage [20],[21]. Our simulations show that the Dex-dependent increase of Per messages creates both Type 0 (as shown in the experiment, strong resetting of the phase) and Type 1 PRCs (weak resetting of the phase) depending on the strength (concentration) of the Dex treatments (Fig. 2A). It is, however, not trivial to simulate a PRC with mostly phase advances reproducing the phenotype from the IR treatment experiments [18]. We observe a PRC with large advances and delays if we follow the simplest possible assumption that DNA damage induces Chk2-dependent phosphorylation and degradation of all forms of PER (monomer, dimer, and complex with BMAL1/CLK) (Fig 1 and Fig 2B). Through in silico experiments, however, we observe minimum phase delays as seen in experiments [18],[21] only when Chk2 does not affect the PER that is in a complex with BMAL1/CLK (i.e. due to conformational changes of PER upon complex formation) (Fig. 2B). In other words, Chk2 prematurely degrades PERs that are not bound to BMAL1/CLK to advance the clock, while allowing continued repression of BMAL1/CLK by not degrading the PERs that are in complex with BMAL1/CLK (Fig. 2C). This prolonged repression on BMAL1/CLK creates small delays when Chk2 affects PERs around their minima as observed in experiments [18],[21].
It is interesting to note that an inhibition of CKIε, another kinase that is known to phosphorylate PER, generates a PRC with only delays [27]. This PRC is qualitatively different than the PRC after DNA damage as there are no advances. We can simulate a mirror image of the PRC with mostly advances, which creates mostly delays, by reducing the rates for Chk2-dependent phosphorylations (not shown). Our data, however, is qualitatively different as we do see small advances whereas Badura and colleagues did not observe any advances [27]. This difference are possibly due to the following reasons: 1) Badura et al. administered a CKIε inhibitor not as a pulse (there was no removal of the drug after administration), and 2) it is possible that Chk2 and CKIε results in different types of phosphorylations which can lead to different consequences. We plan to further investigate this with an extended version of circadian clock module.
Our simple model is adapted from Tyson and colleagues' earlier paper where both negative and positive feedbacks play essential roles in creating a robust oscillator [12],[26]. The autocatalytic positive feedback mechanism in the model arises from different stabilities between PER monomers vs. PER complexes. Based on molecular data from Drosophila system [28]–[31], we assume that PER monomers are more susceptible to degradation than PER in complexes (i.e. PER/PER, PER/CRY, etc.). This creates autocatalytic PER dynamics as PER stabilizes itself by forming complexes. To date, this is the only circadian rhythm model that employs an essential positive feedback mechanism that is necessary to maintain a robust oscillator [32]. Hence, we wondered whether the incorporated essential positive feedback is required (or disposable) in simulating the unique PRCs upon DNA damage.
In order to test our hypothesis, we removed the autocatalysis in the model by assuming no stability differences between PER monomers and complexes. Then, we re-parameterized the system to rescue oscillations (see materials and methods). Note that we had to use a Hill-coefficient = 4 for highly cooperative negative feedback in order to rescue oscillations in our four-variable model in the absence of the autocatalytic positive feedback mechanism. To our surprise, we were not able to generate the unique PRC with predominantly phase advances upon DNA damage even by assuming differential phosphorylation and degradation of PER monomers vs. PER complexes with BMAL1/CLK (lane 2, Table 1).
We wondered whether above conclusions from our simple model can be generalized to a more comprehensive model with distinct wiring network. Hence, we tested Leloup and Goldbeter's mammalian model [33],[34]. They used four sets of parameters in order to investigate possible functions of multiple feedback loops in the circadian system. For our purposes, we concentrated in parameter sets 1 and 3. In the parameter set 1, robust oscillations of their model can arise from two different time-delayed negative feedback loops: PER-driven and PER/CRY-independent BMAL1/CLK-driven negative feedback loops. For this parameter set, they can generate an oscillator based on BMAL1/CLK-driven negative feedback loop in the absence of the PER-driven negative feedback loop. In the parameter set 3, they disabled the BMAL1/CLK-driven negative feedback loop making the system a PER/CRY-dependent single negative feedback oscillator. We did not explore parameter sets 2 and 4 because PER is not required for oscillations in parameter sets 2 and 4. The wiring network of Leloup and Goldbeter's model is significantly different from our model which consists of an intertwined dynamics between an essential autocatalytic positive feedback and time-delayed negative feedback [12],[32].
We incorporated Chk2-induced degradation of PER molecules that are not bound to BMAL1/CLK in the Leloup and Goldbeter's model. Then, we tested Chk-2-dependent differential degradation of PER as in our simple model. Our simulations indicate that we see both TYPE 1 and TYPE 0 PRC depending on the strength of Chk2, but we do not observe asymmetric PRCs with mostly advances (lane 3 and 4, Table 1). These results show that the differential effect of Chk2-dependent degradation of PER complexes is not enough to create the observed DNA-damage induced PRCs with the innate wiring of the Leloup and Goldbeter's model.
Our next step was to introduce an autocatalytic positive feedback mechanism in the Leloup and Goldbeter's model and investigate its role in reproducing the asymmetric PRC upon DNA-damage. First, we added an autocatalytic positive feedback in the parameter set 1 of Leloup and Goldbeter's model in a similar way as in our simple model. PER complexes are assumed to be more stable than PER monomers. To our surprise, we were not able to generate the PRCs with predominantly phase advances with differential degradations of PER complexes by Chk2 even with an added autocatalytic positive feedback mechanism (lane 5, Table 1). We wondered whether this was due to the PER-independent BMAL1/CLK-driven negative feedback loop which is built in the parameter set 1. Hence, we tested the parameter set 3 which consists of the PER-driven single negative feedback. Interestingly, we were able to simulate the observed asymmetric PRC with predominantly phase advances as we have observed in our simple model only when both the autocatalytic positive feedback and the differential effect of Chk2 on PERs were implemented in the absence of BMAL1/CLK-driven negative feedback loop (lane 6, Table 1). This suggests that there exists an important dynamical relationship between negative feedback loops and an autocatalytic positive feedback mechanism.
What are the implications of DNA damage-induced phase responses of the circadian clock to the cell cycle? We hypothesize that cells utilize various pathways for different timing events in response to DNA damage. The Chk2 kinase directly inhibits the progress of the cell cycle by phosphorylating and removing Cdc25C (a phosphatase that is antagonistic to Wee1 which activates cell proliferation) from the nucleus [35]. Moreover, the cell cycle machinery also employs Chk2 in order to provide an additional mechanism that helps to delay the cell cycle progress for extended time by indirectly increasing the level of Wee1 via the circadian network. We believe that the above sequential roles of Chk2 maximize the efficiency of DNA damage-induced delay. With our model, we show that premature degradation of PER, resulting in phase advances, causes early activation of BMAL1 (Fig 2C). This creates an early transcriptional activation of the Wee1 (G2 inhibitor of the cell cycle) during the upcoming circadian cycle, which delays the cell cycle in the G2 phase. If the DNA damage-response induces large phase delays, it will generate a short-lived, transient increase of BMAL1, but a long delay in the activation of Wee1 by BMAL1/CLK for the upcoming circadian cycle. This late activation of Wee1 is probably not a desired result for an efficient DNA damage response.
Our model is simple and intuitive, and yet predicts a molecular mechanism that is responsible for the observed PRC. Our in silico experiments elucidate a molecular mechanism that accounts for Chk2-dependent phase advances and minimum delays of the circadian clock upon DNA damage. It seems counterintuitive to assume that Chk2 does not affect the PER that is in a complex with BMAL1/CLK. This may appear to prolong the repression on BMAL1, which will delay the activation of Wee1. However, due to the cyclic nature of the circadian clock, our simulations suggest that these unique Chk2-dependent phase responses are the best strategy for inducing large and prolonged induction of Wee1 by BMAL1/CLK, allowing extended time for the cell cycle to repair problems upon DNA damage. We propose that the cell cycle network is ingeniously wired with the circadian clock for an optimal response upon DNA damage. Previously, experimentalists showed that the functional circadian clock is important for optimum response to the chemotherapeutic agent cyclophosphamide or γ radiation [4],[36]. For example, reduced apoptosis is observed in mPer2 deficient mice compared to wild-type mice upon γ radiation, which resulted in tumorigenesis [4]. Based on these works, it can be assumed that DNA damage response is more efficient when the circadian clock is intact. We do not know, however, how the efficiency of DNA damage response is affected by the circadian clock. Hence, we suggest testing the efficiency of DNA damage response in the presence and absence of the circadian clock in both in cell culture (i.e. wild-type vs. cryko) as well as in vivo.
Another intriguing finding is the importance of the autocatalytic positive feedback mechanism in simulating the observed PRC upon DNA damage. Our simple model is adapted from Tyson and colleagues which implemented both negative and positive feedback mechanisms [12],[32]. DNA damage-induced PRCs with predominantly advances are lost upon removal of the positive feedback even with the differential degradation of PERs by Chk2. This observation is extended to the Leloup and Goldbeter's model [33],[34]. We tested four different combinations of positive and negative feedback loops with two different sets of parameters (Table 1). Our findings confirm that the autocatalytic positive feedback mechanism is required to simulate DNA damage-induced PRCs. Our results elucidate three important points: (1) the role of the autocatalytic positive mechanism in the circadian system, (2) the wiring of different negative feedback loops, and (3) the interplay between positive and negative feedbacks in response to DNA damage. We acknowledge that there are multiple feedback loops in the circadian system [9]. Therefore, it is essential to develop a more comprehensive model accounting detailed dynamics of different negative feedback loops in the clock network. Furthermore, it is important to experimentally verify autocatalytic positive feedback mechanisms in the context of circadian rhythms, the nonlinearity of negative feedback loops, and the possible interplay between the positive and negative feedback loops in the circadian clock.
Our objective is to create a simple mammalian circadian clock model that accounts for different phase response curves (PRCs) observed from various experiments [17],[18],[21]. For simplicity of the model, we only deal with PER protein and treat PER1, PER2, and PER3 as same proteins. CRY proteins (CRY1 and CRY2) are also part of core clock components that negatively regulate BMAL1/CLK. We do not consider, however, CRY proteins in this model for two reasons: (1) simplicity of the model, and (2) it is not yet known whether Chk2 phosphorylates and triggers degradation of CRY proteins as mPER1. We will include the function of CRY proteins in our future work. We assume that PERs exist in monomers (Clock Protein, CP), dimers (Clock Protein, CP2), and complex with the BMAL1/CLK (Transcription Factor, TF). We imagine that the BMAL1/CLK is inactive when bound to PER (Inactive Complex, IC) creating a negative feedback. We treat CLK as a parameter in the system since it does not cycle [37]. We also assume that the CP2 is more stable than the CP, which introduces a positive feedback in the system [12]. Dex induces the transcription of Per message (Message, M) [18], and DNA damage-activated Chk2 promotes phosphorylation and degradation of PERs [20],[21]. We use same equations and parameter values from our previous publication [25] other than the newly added effects of Dex or Chk2.
Messenger RNA of the clock proteins (Per mRNA):(1)Monomer clock proteins (PER):(2)Dimer form of clock proteins (PER/PER):(3)Transcription factor (BMAL1/CLK):(4)Inactive complex of clock dimers and transcription factor:(5)Total amount of clock proteins (PER on Fig. 2):(6)Rate constants (h−1):Dimensionless constants:
All protein concentrations in the model are expressed in arbitrary units (au) because, for the most part, we do not know the actual concentrations of most circadian proteins in the cell. All rate constants capture only the timescales of processes (rate constant units are in h−1).
Various parameters of the model of Zámborszky et al. [25] have been changed in order to remove the originally existing positive feedback from the system. The equations are the same as presented above. Many parameters were changed to create a robust circadian rhythm with approx 24 h period. Changed parameters: Rate constants (h-1): kms = 0.5, kmd = 0.045, kcps = 10, kcpd = 0.0001, ka = 100, kd = 0.001, kcp2d = 0.0001, kicd = 0.001, kica = 4, kp1 = 1.97, kp2 = 1.97. Dimensionless constants: TFtot = 1, Jp = 0.05, J = 0.4, n = 4.
The Chk2 induces degradation of PER monomers and PER-CRY dimers but not PER proteins that are in complex with BMAL1/CLK. To achieve this we replaced the original Vphos term by (Vphos+VChk2) in the original Leloup and Goldbeter models [33],[34]. In simulations we used VChk2 = 1 to simulate the effect of IR pulse treatment.
We increased the nonspecific degradation rate constant for destruction of nonphosphorylated PER monomers in the cytosol from 0.01 to 0.3, while keeping the background degradation rates of PER/PER dimers and PER/CRY complexes at the original 0.01 level. In this way PER has a positive influence on itself by forming complexes. This creates a similar autocatalytic positive feedback mechanism as the one we used in Zámborszky et al. [25].
We used XPP-AUT computer program [38] of G. Bard Ermentrout (freely available at http://www.math.pitt.edu/~bard/xpp/xpp.html) for simulations and analysis of our model. The ODE file of our model is available as online supplementary material of this article (see Text S1). The SBML version of the model is also downloadable from the BioModels Database (http://www.ebi.ac.uk/biomodels-main/) [39], as MODEL7984093336. For each simulation, we calculated the phase differences between unperturbed and perturbed systems after 10 days (10 circadian cycles). Treatments were induced at each circadian hour.
|
10.1371/journal.pntd.0005783 | Towards Chagas disease elimination: Neonatal screening for congenital transmission in rural communities | Chagas disease is a neglected tropical disease that continues to affect populations living in extreme poverty in Latin America. After successful vector control programs, congenital transmission remains as a challenge to disease elimination. We used the PRECEDE-PROCEED planning model to develop strategies for neonatal screening of congenital Chagas disease in rural communities of Guatemala. These communities have persistent high triatomine infestations and low access to healthcare. We used mixed methods with multiple stakeholders to identify and address maternal-infant health behaviors through semi-structured interviews, participatory group meetings, archival reviews and a cross-sectional survey in high risk communities. From December 2015 to April 2016, we jointly developed a strategy to illustratively advertise newborn screening at the Health Center. The strategy included socioculturally appropriate promotional and educational material, in collaboration with midwives, nurses and nongovernmental organizations. By March 2016, eight of 228 (3.9%) pregnant women had been diagnosed with T. cruzi at the Health Center. Up to this date, no neonatal screening had been performed. By August 2016, seven of eight newborns born to Chagas seropositive women had been parasitologically screened at the Health Center, according to international standards. Thus, we implemented a successful community-based neonatal screening strategy to promote congenital Chagas disease healthcare in a rural setting. The success of the health promotion strategies developed will depend on local access to maternal-infant services, integration with detection of other congenital diseases and reliance on community participation in problem and solution definition.
| Chagas disease is caused by a parasite transmitted by insects that infest households living in extreme poverty conditions. The parasite can also be transmitted from mother to child during pregnancy. If detected at birth, the infection can be treated effectively with available drugs. However, access to professional neonatal healthcare is limited in rural communities such as those affected by Chagas disease. We developed a strategy to promote access to a simple neonatal diagnostic test in a rural region of Guatemala considered at risk for Chagas disease. The strategy included collaboration between Health Center personnel, midwives and non-governmental organizations that play a local role in maternal-infant care. During the implementation of a health promotion campaign, screening revealed previous infection in almost one of every 25 pregnant women. Most babies born to positive women were tested at the Health Center for parasites in blood. The implementation of similar strategies to prevent congenital Chagas in other rural areas should consider local maternal-infant care practices. This strategy of collaboration between Ministry of Health, community health workers, non-government organizations, academia and external governmental support could be expanded to screen for other diseases, such as Zika, that require early detection to improve overall infant health.
| Chagas disease is a vector-borne illness that can also be transmitted congenitally, via blood transfusion, organ donation, lab accidents or ingestion [1]. With the implementation of vector control programs, insect transmission of Trypanosoma cruzi has become less common [2] and vertical transmission has increased in importance [3–6]. In Argentina, a prospective study showed that 67.3% of 107 patients enrolled were infected congenitally, while only 4.7% via vector transmission [7]. In 2005, the Guatemalan Ministry of Health (MoH) proposed to include congenital Chagas disease screening and treatment of children. [8]. However, program implementation has been limited by lack of evidence on congenital incidence rates. We are implementing a strategy to screen congenital transmission in populations at highest risk.
Congenital Chagas disease is an acute infection [9] with 27–57% asymptomatic cases in children [10, 11]. The consequences of infection in utero can be seen prior to birth, with spontaneous abortion and stillbirth and, upon birth, neonates have a higher mortality within the first two days [10]. Diagnosis of congenital Chagas can be achieved with varying degrees of sensitivity by screening the newborn´s blood within the first month after birth by microscopy, hemoculture or by polymerase chain reaction [5, 12, 13]. Infants may be screened serologically 10 months after birth, when maternal transplacental antibodies have waned [14]. Some potential risk factors for vertical transmission of Chagas disease include the degree of parasitemia [15–17], the presence of acute infection in the mother [18], and co-infection with HIV [17, 18]. Treatment should be implemented immediately after diagnosis to improve prognosis [19]. The oral treatment must be monitored by trained health personnel due to potential adverse effects [5, 15]. Thus, screening and treatment programs require access to maternal-infant care within an institutional platform.
Over the past five years, we have worked at the municipality of Comapa in the Department of Jutiapa. This is a region of eastern Guatemala that, prior to the launching of the vector control program in 2000, had some of the highest triatomine infestations (>40% infested households) [20] and seroprevalence in school-age children (13.75%) [10] in the country. We extended our previous multidisciplinary study of Chagas disease vector control [21], working in collaboration with the health personnel, communities and non-governmental organizations to establish a congenital Chagas disease healthcare program. This study aimed to improve congenital Chagas disease detection and treatment in this rural area of Guatemala through a multi-stakeholder driven strategy, based on the PRECEDE (Predisposing, Reinforcing, and Enabling Causes in Educational Diagnosis and Evaluation) PROCEED (Policy, Regulatory and Organizational Constructs in Educational and Environmental Development) model [22] for community interventions. After program implementation, newborns are being screened for Chagas disease at the Health Center (HC).
The study obtained ethical approval from both the Universidad del Valle de Guatemala (#108-10-2014, #100-04-2014) and the Ministry of Health of Guatemala (01–2014) Institutional Review Boards. Individual written consents were obtained from participants before interviews and health access surveys.
Comapa is a municipality located in the department of Jutiapa, in the southeastern region of Guatemala bordering El Salvador at -89°54′46.8″ and 14°6′38.6748″ (Fig 1). Comapa was selected to develop the congenital Chagas disease surveillance protocol due to the presence of a newly built maternity ward (2012), the relevance of the disease to local health authorities, an ongoing Chagas diagnosis and treatment program for children and adults, and an incipient Chagas disease prenatal screening program. Prenatal screening includes a rapid diagnostic test (if available) at the HC in Comapa, with the rapid test provided by the non-governmental organization (NGO), Médicos con Iberoamérica (IBERMED), followed by a single ELISA test performed at the Area Laboratory in Jutiapa.
Quantitative and qualitative research methods were used to understand the local socio-ecological system driving health behaviors. Ultimately, we aimed to develop and implement a sustainable community process for the surveillance of congenital Chagas disease. For this, we conducted the situational assessments of all phases of the PRECEDE component of the model: phase 1 (social), phase 2 (epidemiological), phase 3 (educational and ecological) and phase 4 (administrative and policy). We also conducted one phase of the PROCEED component of the model, phase 5 (design and implementation) of the health promotion strategy [22]. We partnered with the MoH and identified stakeholders (midwives, NGOs, Municipal offices, maternal HC and laboratory personnel) throughout the study to ensure the joint identification of the problems and solutions. Table 1 shows the project phases, timeline, activities and stakeholders in chronological order.
Maternal-infant care in Comapa, Jutiapa, involves public health services, midwives and NGOs. We implemented a multi-stakeholder strategy for neonatal screening to offer timely diagnostics and treatment of congenital Chagas disease. The strategy was generated at the local level through a process including participatory activities with midwives and HC personnel, followed by community-based health communication and educational programs regarding Chagas disease management.
To allow newborn screening and early treatment, the strategy requires (1) reaching the population at highest risk for infection through a community-based health communication program, (2) inclusion of midwives, clinic personnel and NGOs in the implementation of promotional materials for early diagnosis at the HC and (3) HC personnel trained to (a) take newborn blood samples, (b) perform a simple microscopic method to detect parasites in the blood sample, and (c) provide treatment and follow-up for infected neonates. The strategy takes into consideration current maternal-infant care policies and practices at the HC in Comapa, with inclusion of regional NGOs. It also takes advantage of the role played by midwives in informal maternal-infant care, as well as the current national policy requiring their training.
Before our study, infants born to positive women were not screened. To promote newborn detection and treatment, education of midwives and women 15–45 years of age must be developed in a culturally appropriate way. The participatory meetings allowed the development of a socioculturally appropriate strategy for the promotion of congenital Chagas disease screening and treatment in the region. The newborn screening procedure was designed to have a low cost, requiring only microscopic evaluation of the newborn´s blood. The PCR was proposed to confirm microscopic results during method implementation. Once optimized, the parasitological method alone could be implemented in other endemic areas with high seroprevalence. Despite the limitations in maintaining trained personnel in rural areas, we propose that the microscopic method has a potential for sustainability due to its low cost, and could become a standard of care for newborns in these regions. However, rapid test based on the detection of T. cruzi IgM antigen would be better and should be considered as an alternative once available. On the other hand, the training procedures with midwives can become part of the ongoing program to improve maternal-infant health in the country. Cost estimations have not been included given that all procedures can be implemented without additional expenditure to ongoing activities at the Health Center. Limitations of the proposed strategy will likely include the sustainability of the community-level education programs to promote maternal-infant follow-up visits, the inclusion of the program in current prenatal screening programs such as HIV and syphilis, and the ability to maintain HC competency in parasitological diagnosis and record keeping [29]. As observed in South America [30], social and technical constraints in Chagas disease management in Guatemala include lack of knowledge on the disease, loss to follow up, side effects that lead to treatment non-adherence, lack of communication between decentralized health system levels and lack of training on diagnostics and treatment. We propose that the inclusion of midwives as empowered stakeholders has resulted in referral of newborns to the health center. Future studies will evaluate the strengths and limitations of this strategy, and recommended improvement.
The scaling up of the strategy will require a train-the-trainer program targeting reproductive health and nurse coordinators at the department level for prioritized areas. In addition, evidence of local transmission and education campaigns are needed to empower stakeholders at all levels. Targeted communication campaigns should be developed based on in-depth knowledge of the sociological and cultural behavior of the communities regarding maternal and neonatal care, and how they interact with the health authorities. Forms for recording screening and treatment of mothers and neonates must be developed or modified, and methods of reporting to epidemiological, vector control and policy authorities strengthened. A supply of treatment medication must also be ensured.
The treatment of T. cruzi-infected women after delivery to reduce the risk of congenital transmission remains a challenge because there are no guidelines regarding treatment during the lactation period. Women with Chagas disease can breast feed, unless they are in the acute phase with high parasitemia, reactivated disease or have bleeding nipples [31]. In rural areas where women have multiple pregnancies, treating infected women during lactation would allow completion of the two-month course before another pregnancy [32]. Research in this area is needed to provide evidence. Finally, treatment before the first pregnancy reduces the risk of congenital transmission [33] and should be considered in future prevention strategies.
A 3.9% seroprevalence in pregnant women attending the HC indicates that early congenital detection and treatment should be a priority in areas with similar historically high triatomine infestations [34] and seroprevalence [14] in Guatemala. To achieve elimination, more studies are needed to understand the prevalence of congenital disease in such populations. In similar areas with persistent triatomine infestation, the MoHs must ensure that Chagas disease control and prevention programs integrate innovative vector control strategies and attention to treatable congenital disease. Future assessment of the strategy is needed to ensure its long term effectiveness and sustainability. The strategy could be expanded to other congenital diseases by strengthening the network of midwives and maternity ward personnel through training in symptom detection at the community level and case referral to health facilities in areas with low access to health services. Given the recent emergence of Zika as a new vector-borne congenital disease, we propose that this stakeholder driven strategy could be implemented in areas with limited access to maternal-infant health services.
|
10.1371/journal.pcbi.1003429 | Phylogenetic Gaussian Process Model for the Inference of Functionally Important Regions in Protein Tertiary Structures | A critical question in biology is the identification of functionally important amino acid sites in proteins. Because functionally important sites are under stronger purifying selection, site-specific substitution rates tend to be lower than usual at these sites. A large number of phylogenetic models have been developed to estimate site-specific substitution rates in proteins and the extraordinarily low substitution rates have been used as evidence of function. Most of the existing tools, e.g. Rate4Site, assume that site-specific substitution rates are independent across sites. However, site-specific substitution rates may be strongly correlated in the protein tertiary structure, since functionally important sites tend to be clustered together to form functional patches. We have developed a new model, GP4Rate, which incorporates the Gaussian process model with the standard phylogenetic model to identify slowly evolved regions in protein tertiary structures. GP4Rate uses the Gaussian process to define a nonparametric prior distribution of site-specific substitution rates, which naturally captures the spatial correlation of substitution rates. Simulations suggest that GP4Rate can potentially estimate site-specific substitution rates with a much higher accuracy than Rate4Site and tends to report slowly evolved regions rather than individual sites. In addition, GP4Rate can estimate the strength of the spatial correlation of substitution rates from the data. By applying GP4Rate to a set of mammalian B7-1 genes, we found a highly conserved region which coincides with experimental evidence. GP4Rate may be a useful tool for the in silico prediction of functionally important regions in the proteins with known structures.
| To understand how a protein functions, a critical step is to know which regions in its protein tertiary structure may be functionally important. Functionally important protein regions are typically more conserved than other regions because mutations in these regions are more likely to be deleterious. A number of phylogenetic models have been developed to identify conserved sites or regions in proteins by comparing protein sequences from multiple species. However, most of these methods treat amino acid sites independently and do not consider the spatial clustering of conserved sites in the protein tertiary structure. Therefore, their power of identifying functional protein regions is limited. We develop a new statistical model, GP4Rate, which combines the information from the protein sequences and the protein tertiary structure to infer conserved regions. We demonstrate that GP4Rate outperforms Rate4Site, the most widely used phylogenetic software for inferring functional amino acid sites, via simulations with a case study of B7-1 genes. GP4Rate is a potentially useful tool for guiding mutagenesis experiments or providing insights on the relationship between protein structures and functions.
| An important question in biology is the identification of functional residues in proteins. This information can help us understand the relationship between protein structures and functions as well as guide us to design new proteins by genetic engineering. However, experimental techniques for identifying functional sites, e.g. mutagenesis, are time consuming and expensive, which prohibits the brute force scanning of functional sites by experiments. Therefore, bioinformatics tools are useful, because they can narrow down the candidate sites for experimental investigation. Evolution operates similar to a high-throughput mutagenesis experiment: spontaneous mutations introduce protein variants in each generation and then the functional effects of the spontaneous mutations are “measured” by natural selection [1]. Therefore, protein sequences contain signatures of natural selection which reflect the functions of amino acid residues. For example, mutations at the functionally important sites tend to disrupt the proteins' normal functions, so these sites usually are more conserved than unimportant ones. If the sequences of a family of homologous proteins can be collected from multiple species, we may compare these sequences to infer which sites are more important than others.
A number of bioinformatics tools based on phylogenetics have been developed to infer functional sites by the simple idea that functionally important amino acid sites tend to be more conserved than unimportant ones [2]–[11]. Given the multiple sequence alignment and the phylogenetic tree of a protein family, these phylogenetic methods can infer the amino acid substitution rate at each site in the alignment and an unusually low substitution rate implies that the site is functionally important. It has been shown that the predicted conserved sites coincide with experimental evidence, which confirms that these bioinformatics tools are useful.
However, these existing methods are far from flawless. Most of the popular methods, e.g. Rate4Site [7] used in the ConSurf web server [11], assume that the substitution rates are independent across sites. In statistical terms, this means that the sites in the alignment are independent and identically distributed (i.i.d.). The i.i.d. assumption simplifies the statistical modeling, but it is unrealistic from the viewpoint of biology. The i.i.d. assumption implies that the slowly evolved functional sites are randomly distributed in the protein tertiary structure. In contrast, it is well known that functionally important sites tend to be close to each other in the protein tertiary structure and form functional regions, e.g. ligand binding sites or catalytic active sites. Clearly the i.i.d. assumption is inappropriate if a functional region consists of a number of sites.
Several methods have been developed to incorporate the spatial correlation of evolutionary patterns, e.g. substitution rates at the protein level or dN/dS ratios at the codon level, to overcome the drawbacks of the i.i.d. assumption [3], [5], [8], [12]–[16]. Most of these methods use a sliding window framework, in which the amino acid substitution rate or the dN/dS ratio at a focal site is approximated by the average substitution rate in a set of neighbor sites in the protein tertiary structure [3], [12], [13]. A site is considered to be a neighbor of the focal site if the Euclidean distance between the two sites is smaller than a predefined window size. Unfortunately, these sliding window methods also have intrinsic drawbacks. Firstly, in most, if not all, of sliding window methods the neighbor sites, including the focal site itself, are weighted equally in the inference of the substitution rate. However, clearly the focal site itself contains more information on its substitution rate than the sites near the boundary of the sliding window. Secondly, it is unclear how to determine the optimal window size [17], [18]. If the window size is too large, there will be too many distant sites in the window, which could bias the estimation at the focal site. In contrast, if the window size is too small, the sliding window methods will not be able to capture the spatial correlation of substitution rates and may lead to overfitting. Furthermore, there is evidence that the optimal window sizes may vary among different protein families [12].
Very recently, a Bayesian model which combines the Potts model in statistical physics and the phylogenetic model has been proposed by Watabe and Kishino to infer protein patches under positive selection in protein tertiary structures [16]. In Watabe and Kishino's model, the Potts model is used to define a prior distribution of dN/dS ratios over a protein tertiary structure. This model solved many problems of the sliding window framework. However, the prior distribution in Watabe and Kishino's model is unnormalized [16], which makes it difficult to design efficient algorithms to estimate hyperparameters. An advanced algorithm, thermodynamic integration [19], was used in Watabe and Kishino's model to infer hyperparameters. However, the algorithm may be very inefficient, especially if there are many hyperparameters in the Potts model.
Here we propose to incorporate a Gaussian process with the phylogenetic model to overcome the drawbacks of the existing methods. The Gaussian process has been widely applied in geostatistics and machine learning to capture the spatial correlation of interesting features [20], [21]. Here we will briefly introduce the basic idea of the Gaussian process. More details of the Gaussian process and its applications can be found in the geostatistics and machine learning literature, e.g. [20]. A Gaussian process defines a probability distribution over functions, namely that a single sample point of the Gaussian process is a function over a space, e.g. a 3D space. Because the sample points of the Gaussian process are “smooth” functions, the Gaussian process encodes an intrinsic spatial correlation. Thus physically closely located points in the space are more likely to have similar function values. Therefore, the Gaussian process is very useful for defining prior distributions over spatially correlated patterns. For example, in this paper we are interested in modeling the spatial correlation of site-specific substitution rates in protein tertiary structures. If we image each residue in a protein tertiary structure as a single point in the 3D space, the Gaussian process can be used to define a prior distribution of site-specific log substitution rates over these points (residues). The “smoothness” property of Gaussian process prior suggests that two physically closely located sites are more likely to have similar site-specific log substitution rates than two distantly located sites. Then, the Gaussian process prior can be combined with standard phylogenetic likelihood functions [22] to infer site-specific substitution rates from real data.
We name this kind of hybrid model of Gaussian processes and phylogenetics as a phylogenetic Gaussian process model (Phylo-GPM). In the Phylo-GPM framework, the spatial correlation of substitution rates can be naturally described and the strength of spatial correlation can be learned from the data. Therefore, it overcomes the common drawback of the sliding window methods that the window size must be manually specified. Unlike Watabe and Kishino's model [16], the phylogenetic Gaussian process model uses a normalized prior, so simple algorithms, i.e. the widely used Metropolis algorithm [23], [24], can be used to efficiently infer hyperparameters. We have developed software, GP4Rate, based on the Phylo-GPM framework. In both simulated and real datasets, GP4Rate outperforms Rate4Site, a widely used tool based on the i.i.d. assumption. Therefore, GP4Rate may be a useful tool for the identification of functionally important sites.
Simulations were implemented to evaluate the performance of GP4Rate and to compare it with the widely used software, Rate4Site [7]. In the comparisons, Rate4Site is used as a representative of the classic phylogenetic models which use the discrete Gamma distribution to describe the variation of substitution rates across sites [25] but do not consider the spatial correlation of site-specific substitution rates in the protein tertiary structure. Because the true site-specific substitution rates are known in the simulated alignments, the estimated site-specific substitution rates can be compared with the true rates to evaluate the performance of the two methods. We generated two sets of simulated alignments based on different assumptions. In this and the next section, we will describe the first set of simulations which were based on a 2D toy protein structure. Thereafter we will describe the second set of simulations which were based on more realistic assumptions.
To generate simulated alignments, we need a phylogenetic tree to describe the evolutionary relationship between simulated sequences, a protein structure to calculate the pairwise Euclidean distances between sites, a substitution model, and a vector of substitution rates. Note that the following discussions will be mainly based on the substitution rates rather than their log values. A simple phylogenetic tree was used in all simulations (Figure 1A). The tree consisted of four sequences and all the branch lengths were equal to 0.2 substitution per site. Because the total branch length was equal to 1 substitution per site, on average an amino acid site only contained a single substitution. Therefore, the accurate estimation of substitution rate at a single site is challenging. The JTT substitution model [26], [27] was used in all simulations. Note that the protein tertiary structure and the vectors of substitution rates used in the two sets of simulated alignments were different and will be described in detail below.
In the 2D toy protein model, the protein tertiary structure was described by a 20 by 20 regular 2D grid, in which each dot corresponds to an amino acid in the toy protein structure (Figure 1B). In addition, we assumed that the distance between adjacent sites in the 2D grid is equal to 5 Å. This distance is comparable to the average distance between carbon atoms of the physically interacting residues in real proteins. Even though the 2D toy protein model is artificial and no real protein has a similar structure, it is useful because the estimated site-specific substitution rates can be easily visualized by a heatmap (Figure 2). Therefore, we used the 2D toy protein model to check the correctness of the program and to get insights on the performance of GP4Rate.
Two different spatial configurations of site-specific substitution rates were used in the 2D toy protein simulations. In the first configuration, the 20 by 20 grid was divided into 4 non-overlapping blocks, each of which was a 10 by 10 grid (Figure 2A). Sites within a block had the same substitution rates but different blocks could have different substitution rates. Two substitution rates, 0.2 and 1.8, were used for simulations and the substitution rates of blocks were alternatively arranged in the 2D protein structure (Figure 2A). Therefore, the toy proteins consisted of two conserved blocks with low substitution rates (0.2) and two variable blocks with high substitution rates (1.8). The second configuration was similar to the first one, but the sizes of non-overlapping blocks were 5 by 5 instead of 10 by 10 (Figure 2B). Twenty simulated alignments were generated for each configuration of site-specific substitution rates. It is easy to notice that the average site-specific substitution rate is equal to 1 in both configurations.
A program based on Bio++ [28], [29] was developed to implement the simulations. For each simulated alignment, we ran two separate MCMC chains using GP4Rate to estimate site-specific substitution rates. For each MCMC chain, iterations were implemented and the trace plots of the MCMC outputs were monitored to ensure the convergence of the MCMC chains. The first of the iterations were discarded as burn-in. Then, the two chains were combined to calculate the average substitution rate at each site. To compare the performance of GP4Rate with that of Rate4Site, we also used Rate4Site to estimate the substitution rates. To make the results of GP4Rate and Rate4Site more comparable, the phylogenetic tree and branch lengths were fixed to the true values in both GP4Rate and Rate4Site.
We firstly randomly sampled two simulated alignments, one for each configuration, as examples to get insights on the performances of GP4Rate and Rate4Site. As shown in Figure 2C and 2D, the site-specific substitution rates estimated by GP4Rate are smoothly distributed within the 2D protein structures. In addition, GP4Rate segments the 2D protein structures into blocks which correspond to the true patches with different substitution rates. In contrast, the spatial distributions of substitution rates estimated by Rate4Site are far from smooth. The sites with similar substitution rates are not clustered together and do not form clearly bounded patches (Figure 2E and 2F). Thus, GP4Rate can capture the spatial correlation of substitution rates but Rate4Site cannot.
To quantitatively evaluate the performance of GP4Rate and Rate4Site, we used receiver operating characteristic (ROC) curves to measure the power of the two methods. ROC curves are widely used to evaluate the accuracy of binary classifiers. The area under a ROC curve is usually used as a measure of the power of the corresponding method. To apply ROC curves to the simulated datasets, we must divide the amino acid sites into two categories, functional sites and nonfunctional sites, before generating simulated alignments. The functional sites are used as true positives while the nonfunctional sites are used as true negatives. In the 2D toy protein simulations, functional sites evolved at the lower rate (0.2) while nonfunctional sites evolved at the higher rate (1.8). Then, the ROC curves were created by plotting the average true positive rates versus the average false positive rates using the ROCR library in R [30]. As shown in Figure 3A and 3B, the areas under the ROC curves generated by GP4Rate are larger than those generated by Rate4Site, which suggests that GP4Rate outperforms Rate4site.
ROC curves measure whether a model can distinguish slowly evolved functional sites from the other sites. If a model can assign relatively low substitution rates to slowly evolved sites and relatively high rates to the other sites, it will perform well in the evaluations based on ROC curves. However, ROC curves cannot capture potential systematic biases of the model. For example, if the model adds a constant bias to the site-specific substitution rates, its ROC curves will be exactly the same regardless of the magnitude of the constant bias. Therefore, we used a simple loss function complementary with the ROC curves to capture any potential systematic biases of the estimated site-specific substitution rates. The loss function is defined by the following formula(1)in which is the total number of sites in the alignment, while and are the true and estimated log substitution rates at site i, respectively. The log values of site-specific substitution rates are used in the right-hand side of Equation 1, since we want to emphasize the differences between low substitution rates. It is desirable because both GP4Rate and Rate4Site were designed to detect conserved regions with low substitution rates. Unlike ROC curves, a model which introduces a larger systematic bias will have a higher average loss than a model which introduces a smaller bias.
We plotted the losses of both GP4Rate and Rate4Site in the 2D toy protein simulations. As shown in Figure 3C and 3D, GP4Rate outperforms Rate4Site, as evident by the lower losses produced by GP4Rate (paired Wilcoxon test, for both of the two configurations). The improved accuracy originates from GP4Rate's ability to model the spatial correlation of site-specific substitution rates, since the performance gap between GP4Rate and Rate4Site becomes smaller in the second configuration which consists of smaller conserved and variable patches.
GP4Rate has two hyperparameters, i.e. the characteristic length scale and the signal standard deviation , which model the strength of spatial correlation of substitution rates and the marginal variation of substitution rate at a single site, respectively. An advantage of GP4Rate over the sliding window methods is that the hyperparameters can be learned from the data. In contrast, the window size of the sliding window methods must be predefined before analyses. To show that GP4Rate can learn the hyperparameters from the data, we plotted the estimated median hyperparameters of the simulated alignments. As shown in Figure 4A, the characteristic length scales estimated in the first configuration are about 3 fold larger than those estimated in the second configuration. Because the patches are much larger in the first configuration, the result suggests that GP4Rate can learn the magnitude of the spatial correlation of substitution rates from the data. The estimated signal standard deviations in the two configurations are similar, which matches the intuition that the two configurations are similar except in the strength of spatial correlations of substitution rates.
In summary, when spatial correlation of substitution rates exists in proteins, GP4Rate always outperforms Rate4Site. However, the spatial correlation of site-specific substitution rates may be insignificant in some proteins. Therefore, we also evaluated both GP4Rate and Rate4Site in simulated alignments in which the spatial correlation of site-specific substitution rates is absent. These simulated alignments were generated by randomly shuffling the columns in each alignment in the first spatial configuration of substitution rates (Figure 2A). The permutations of alignments destroyed the spatial patten of site-specific substitution rates. Here we only summarize the performance of GP4Rate and Rate4Site in the permuted alignments and more details can be found in the online Supplementary Material. The absence of spatial correlation results in close-to-zero characteristic length scales in GP4Rate, which confirms that GP4Rate can detect the absence of spatial correlation when there is none. Plots of ROC curves show that GP4Rate and Rate4Site have effectively the same power to distinguish slowly evolved sites from the other sites. In contrast, when we use the loss function (Equation 1) to measure the accuracy of estimated substitution rates, GP4Rate is less accurate than Rate4Site. Nevertheless, GP4Rate and Rate4Site have similar power to find slowly evolved functional sites, since in practice it is the relative rankings of sites instead of their absolute substitution rates tell us which sites may be more likely to be functional.
We generated a second set of simulated alignments based on more realistic assumptions. The basic idea is that if we have a large number of highly diverged sequences, a simple method which does not consider the spatial correlation of substitution rates may accurately estimate the site-specific substitution rates because of the rich information in a very large dataset. We may generate simulated alignments based on the real protein tertiary structure and the presumably accurately estimated site-specific substitution rates. These simulated alignments may have similar features as real proteins.
In this set of simulations, we used the same phylogenetic tree (Figure 1A) and the JTT substitution model [26], [27] used in the 2D toy protein simulations. The protein tertiary structure and the site-specific substitution rates were based on a real protein, B-cell lymphoma extra large (Bcl-xL). This protein has been studied using Rate4Site and the two predicted conserved patches coincide with the regions with known functions [31]. We downloaded the protein tertiary structure of Bcl-xL from Protein Data Bank (PDB ID: 1MAZ [32]). The site-specific substitution rates estimated by Rate4Site were obtained from the ConSurf-DB database [10]. In ConSurf-DB, 131 unique homologs of Bcl-xL were automatically collected and then Rate4Site was applied to estimate the site-specific substitution rates. Because of the very large number of sequences in the dataset, the estimation of site-specific substitution rates may be relatively accurate. We generated 20 simulated alignments based on the above assumptions and both GP4Rate and Rate4Site were applied to the simulated alignments using the same setting described in the 2D toy protein simulations.
To evaluate the performance of GP4Rate and Rate4Site by ROC curves, we divided the sites into two categories before generating simulated alignments: slowly evolved functional sites and others. Based on the site-specific substitution rates reported by ConSurf-DB, the 10 percent most slowly evolved sites were considered to be functional while the others were not. As shown in Figure 5A, GP4Rate is more powerful to distinguish slowly evolved sites from the other sites, since the area under the ROC curve of GP4Rate is larger than that of Rate4Site. In addition, based on the loss function defined by Equation 1, GP4Rate produces lower losses in 18 out of the 20 simulated alignments (Figure 5B) and the median loss of GP4Rate is significantly smaller than that of Rate4Site (paired Wilcoxon test, value<10−4). Therefore, GP4Rate still outperforms Rate4Site in the realistic simulations.
The B7-1 (CD80) family is a member of the immunoglobulin superfamily (IgSF) and is critical for the regulation of immune responses [33]. The protein tertiary structure of the human B7-1 protein has been determined [34], [35]. The human B7-1 protein consists of two IgSF domains (IgV and IgC), each of which shows an anti-parallel sandwich structure [34]. We applied GP4Rate and Rate4Site to 7 mammalian B7-1 sequences downloaded from the NCBI HomoloGene database [36] and compared their performances. The N-terminal and C-terminal sequences were trimmed in the alignment, because the corresponding atoms are absent in the X-ray crystal structure. The resulting alignment consists of 199 amino acid sites. Then the phylogenetic tree was inferred by PhyML with the model [37]. The protein sequences in the alignment are very similar to each other as evident by the lack of gaps in the alignment (data not shown). Therefore, the information in each site in the alignment is very limited and it is hard to infer site-specific substitution rates accurately.
We used the human B7-1 protein structure (PDB ID: 1I8L [35]) to calculate the pairwise Euclidean distances between the carbon atoms of amino acids. Then, we applied GP4Rate to the B7-1 alignment to infer site-specific substitution rates. We ran two independent MCMC chains for iterations, and the first of the iterations were discarded as burn-in. We first estimated the posterior marginal distributions of hyperparameters based on the MCMC samples. As shown in Figure 6, the estimated characteristic length scale is significantly higher than 0, which confirms that the substitution rates are correlated in real proteins. The presence of spatial correlation of substitution rates may facilitate the discovery of slowly evolved functional regions. To test this hypothesis, the mean site-specific substitution rates of the MCMC samples were calculated and the 20 most slowly evolved sites were considered to be functional. Then, the 20 most slowly evolved sites were superimposed onto the protein tertiary structure (PDB ID: 1I8L [35]). As shown in Figure 7A, the slowly evolved sites predicted by GP4Rate are not randomly distributed and instead form a single large region in the IgC domain. A systematic mutagenesis study has suggested that the IgC domains are important for binding CTLA-4 and CD28 [38], even though the effects of the IgC domain may be indirect [35]. To test whether the predicted slowly evolved sites overlap with the experimentally verified functional sites [38], the 7 experimentally verified functional sites in the IgC domain were mapped onto the human B7-1 structure (Figure 7A). Clearly 4 experimentally verified functional sites in the IgC domain, i.e. Q157, D158, E162, and L163, are within the slowly evolved patch predicted by GP4Rate, which highlights the potential usefulness of GP4Rate.
To compare GP4Rate with Rate4Site, we also applied Rate4Site to the same dataset. The superimposition of the 20 most slowly evolved sites predicted by Rate4Site is shown in Figure 7B. The sites predicted by Rate4Site are present in both the IgV and IgC domains and do not form clearly bounded regions. Even though 2 experimentally verified functional sites in the IgC domain, i.e. F106 and I113, overlap with the sites predicted by Rate4Site, the 4 experimentally verified functional sites detected by GP4Rate do not overlap with the sites predicted by Rate4Site. Therefore, GP4Rate and Rate4Site can provide complementary insights to real data.
To investigate which model, GP4Rate or Rate4Site, fits the B7-1 dataset better, we performed a Bayesian model comparison. The direct comparison between GP4Rate and Rate4Site is impractical, because Rate4Site is based on the maximum likelihood principle instead of the Bayesian principle. However, it is not very difficult to develop a Bayesian version of Rate4Site by specifying a prior distribution over parameters. Therefore, we developed a Bayesian version of Rate4Site and compared it with GP4Rate. Details of the Bayesian model comparison can be found in the online Supplementary Material and we only summarize the results here. We compared the site-specific substitution rates estimated by the original Rate4Site and its Bayesian version and found that the two programs produced essentially the same result. Therefore, the marginal likelihood estimated by the Bayesian version of Rate4Site may be used to evaluate how good the original Rate4Site fits the B7-1 dataset. The log marginal likelihood of GP4Rate is equal to while the log marginal likelihood of the Bayesian Rate4Site is equal to , which suggests a very large Bayes factor of GP4Rate compared with the Bayesian Rate4Site (). Therefore, GP4Rate fits the B7-1 dataset much better than the Bayesian Rate4Site.
Many phylogenetic methods have been developed to identify slowly evolved amino acid sites which may be functional. However, the most widely used methods, e.g. Rate4Site, ignore the spatial correlation of site-specific substitution rates. Some other methods use the sliding-window framework to capture the spatial correlation of substitution rates, but the statistical method for choosing the optimal window size is largely unknown. Since the strength of the spatial correlation of substitution rates is unknown in most of proteins, the sliding window methods are problematic in real data analyses. In GP4Rate, both of the two issues are solved under a Bayesian statistical framework. By using the Gaussian process to define the prior distribution of the site-specific log substitution rates, GP4Rate can naturally model the spatial clustering of functionally important sites and the hyperparameters which measure the strength of spatial correlation can be inferred from the data instead of being manually specified before the analyses.
In simulated datasets, GP4Rate significantly outperforms Rate4Site. The power of GP4Rate is mainly derived from the fact that GP4Rate has the added ability to model the spatial correlation of substitution rates. By borrowing statistical information from neighbor sites with similar substitution rates, GP4Rate can estimate the site-specific substitution rates with a much higher accuracy than Rate4Site. In the case study of B7-1 genes, GP4Rate predicted a slowly evolved functional patch in the protein tertiary structure and 4 sites within the region are well supported by experimental evidence. In contrast, the slowly evolved sites predicted by Rate4Site are scattered and do not form clearly bounded regions. In addition, we have shown that GP4Rate fits the B7-1 dataset much better than Rate4Site based on Bayesian model comparison.
The performance gap between GP4Rate and Rate4Site will be maximized when the protein sequences are very similar to each other and the spatial correlation is strong. Therefore, GP4Rate is most suitable to analyze small gene families, e.g. new genes or small gene families derived from recent gene duplication events. When the spatial correlation of substitution rates is weak, GP4Rate and Rate4Site may generate similar results. For example, we applied GP4Rate to 38 RH1 genes [39] and found that the spatial correlation of substitution rates is much weaker in the RH1 dataset than that in the B7-1 dataset (data not shown). In this case, the difference between GP4Rate and Rate4Site is subtle. Therefore, a rigorous model comparison as shown in the case study of B7-1 genes may be important in data analyses.
Because GP4Rate is based on MCMC simulations, it is slower than Rate4Site. For example, it took about 1 CPU day for GP4Rate to analyze the B7-1 dataset. However, GP4Rate is still very useful for small scale problems, e.g. guiding mutagenesis experiments, since the experimental time is much longer than the execution time of GP4Rate. The time cost of GP4Rate can be reduced in the future using advanced algorithms, e.g. more efficient MCMC sampling algorithms or sparse approximations of the Gaussian process [40]. The most time consuming step of GP4Rate is the Cholesky decomposition whose time complexity is a cubic function of the number of sites in the alignment. In practice, a simple method to reduce the computational time is to perform the analyses based on a selected subset of amino acid sites. For example, it is well known that surface residues are more likely to be involved in interactions with other proteins or ligands. If these interactions are most interesting to users, a fast analysis based only on the surface residues may be appropriate.
In addition to modeling the spatial correlation of site-specific substitution rates, protein tertiary structures have been used to improve phylogenetic models and the estimation of site-specific substitution rates in a few other studies [41]–[46]. These methods can be roughly divided into two categories. The first category of models assumes that the fixation probability of new mutations is determined by how the mutations influence the stability of the protein [41]–[43]. Typically it is assumed that mutations which stabilize the protein structure are more likely to be fixed than mutations which destabilize the protein structure. Unlike this category of models, the Phylo-GPM framework does not provide a mechanistic interpretation for the estimated substitution rates. However, GP4Rate may be more powerful to identify functional regions which are not directly relevant to the stability of proteins. The second category of models assumes that the site-specific substitution rates or dN/dS ratios are influenced by the local environment of the focal site in the protein tertiary structure [44]–[46]. For example, it has been shown that the dN/dS ratio of a site is influenced by its relative solvent accessibility (RSA) [44]–[46]. It is relatively straightforward to combine the Phylo-GPM framework with local features of amino acid sites. For example, in this study we assume that the site-specific log substitution rates follow a zero-mean Gaussian distribution. We may replace the zero-mean rate vector by a new one in which the mean of log substitution rate at a site is a linear function of its local features, e.g. RSA. It is very interesting to investigate whether adding local features to the Phylo-GPM framework improves model fitting in the future.
The Phylo-GPM framework proposed in this paper may be used as a general tool to model the spatial correlation of patterns in the protein tertiary structure. The phylogenetic hidden Markov model (Phylo-HMM) is a popular method which combines the hidden Markov model and statistical phylogenetics [47]. It has been used to model the spatial correlation of evolutionary patterns along primary sequences [17], [48]–[53]. The Phylo-GPM framework may be viewed as an extension of the Phylo-HMM to the protein tertiary structures. In the future, new methods based on the Phylo-GPM framework may be developed to identify functional divergence or positive selection in proteins.
GP4Rate is an open-source software application written in C++ and its source code is freely available from http://info.mcmaster.ca/yifei/software.html. GP4Rate combines the protein alignment and the protein tertiary structure to infer groups of close-located functional sites evolved at low rate. We assume that the protein alignment, the phylogenetic tree, and the tertiary structure of one protein in the alignment are provided by users. In GP4Rate, both the topology and the branch lengths of the phylogenetic tree are fixed to improve the speed of the program. In addition, we assume that the protein sequences in the alignment belong to the same gene family and have very similar functions, which implies that the functionally important sites do not vary among sequences and the site-specific substitution rates do not change over time. However, we do assume that the substitution rates can vary across different sites. The site-specific rates are used as proxies of functionality: very low substitution rates suggest the corresponding sites are functionally important.
In most molecular phylogenetic programs, e.g. Rate4Site [7], PAML [54], and PhyML [37], the site-specific substitution rates are assumed to be i.i.d. and follow a simple discrete distribution, usually the discrete Gamma distribution [25]. Recently, Dirichlet process pirors have been used to model the variable substitution rates over sites to overcome the inflexibility of the simple discrete distributions [55], but it is still assumed that the site-specific substitution rates are i.i.d. The i.i.d. assumption implies that slowly evolved functional sites are scattered in the protein tertiary structure. The major contribution of this paper is to relax the i.i.d. assumption using the Gaussian process [21] which can naturally capture the spatial correlation of site-specific substitution rates in the protein tertiary structure.
In GP4Rate, the parameters are estimated using the Bayesian principle. In Bayesian statistics, the parameters are random variables and the conditional distribution of parameters given data, i.e. the posterior distribution, gives us an estimation of parameters. For simplicity of presentation, first we focus on the vector of site-specific log substitution rates, which is the collection of log values of substitution rates at all amino acid sites, and defer the discussions on the other parameters. The posterior distribution of the vector of log site-specific substitution rates can be defined by the following equation,(2)In the equation, is the vector of site-specific log substitution rates, is the protein alignment while is its i-th column, and is the phylogenetic tree with the associated branch lengths. is the site-specific likelihood at site i, which is a function of the site-specific log substitution rate at site i. is the fundamentally important prior distribution of site-specific log substitution rates.
A realistic should be able to describe the spatial correlation of site-specific substitution rates. In GP4Rate, is specified by a zero-mean Gaussian process. A Gaussian process is a probability measure defined over a function space. In the statistical modeling of site-specific substitution rates, we are only interested in the marginal distribution of the Gaussian process over a finite set of spatial locations which correspond to the locations of residues in the protein tertiary structure. By the definition of Gaussian processes, the marginal distribution of a zero-mean Gaussian process is a zero-mean multivariate Gaussian distribution [21]. Therefore, may be rewritten in the following format,(3)The correlation of site-specific substitution rates is determined by the covariance matrix , in which is the pairwise distance matrix which measures the Euclidean distance between the carbon atoms of amino acids in the protein tertiary structure. Furthermore, the covariance function is parameterized by two hyperparameters, and , which measure the strength of spatial correlation and the variation of substitution rates across sites, respectively. By plugging and , the prior distribution of the hyperparameters, into Equation 2, it can be expanded to the following format,(4)In the following sections, we will provide more details on the specifications of the right-hand side terms of Equation 4 and the MCMC algorithm for the sampling of parameters, i.e. , , and .
As mentioned above, follows a zero-mean multivariate Gaussian distribution (Equation 3). In the multivariate Gaussian distribution, the covariance matrix is specified by a covariance function. By default, GP4Rate uses the Matérn 1.5 covariance function,(5)In the equation, is an element in the covariance matrix while is an element in the distance matrix which measures the Euclidean distance between site i and site j in the protein tertiary structure. is an indicator function which is equal to 1 if site i and site j are the same site and equal to 0 otherwise. The covariance function contains two free parameters, and . is the characteristic length which determines the strength of the spatial correlation of substitution rates. If it is small, the spatial correlation is weak and only nearby sites have similar log substitution rates. Instead, if it is large, the spatial correlation is strong and distant sites can have similar log substitution rates. is the signal standard deviation which measures the marginal variation of log substitution rates at a single site. Small implies that the variation of log substitution rates is small. is a fixed “jitter” term which introduces a small amount of noise to the diagonal elements in . The “jitter” term ensures that the Cholesky decomposition, a critical numerical algorithm in the MCMC simulations, is numerically stable and improves the mixing of the MCMC simulations [56]. The “jitter” term is usually a small positive number (e.g. ), so it does not significantly change the behavior of the covariance function [56]. Clearly Equation 5 implies that the covariance of log substitution rates are decreasing with increasing Euclidean distance between two amino acid sites, which is compatible with our intuition that nearby sites tend to have similar substitution rates due to similar functions.
In addition to the Matérn 1.5 covariance function, GP4Rate has two alternative covariance functions for users to choose. One is the Matérn 2.5 covariance function,(6)The other is the widely used squared-exponential covariance function,(7)
The three covariance functions are all special cases of the general Matérn covariance function [21]. The major difference between them is that the three covariance functions describe different levels of smoothness in the spatial distribution of site-specific log substitution rates [21]. In the squared-exponential covariance function, the site-specific log substitution rates are smoothly distributed in the protein tertiary structure. Therefore, it is most suitable to model proteins with relatively large functional regions. In contrast, the Matérn 1.5 covariance function is the least smooth one and is suitable to model proteins with small functional patches. In this paper, we used the Matérn 1.5 covariance function in all analyses to allow for proteins that may have relatively small functional patches and could have nearby sites with very different substitution rates.
The hyperparameters in the covariance functions, i.e. and , follow a prior distribution . We assume that the characteristic length, , and the signal standard deviation, , are independent and follow exponential distributions. Therefore, the prior distribution is defined by the following probability density function,(8)We choose and to be large so that the prior distribution has relatively weak information.
To fully define the unnormalized posterior distribution (Equation 4), the likelihood must be specified. We follow the standard phylogenetic model first described by Felsenstein [22]. We assume that the substitution model in the phylogenetic likelihood function is fixed to the JTT model [26], [27] while the phylogenetic tree is fixed to the one provided by the users. The likelihood can be calculated by the pruning algorithm and the gaps in the alignment may be treated as missing data [22]. However, the calculation of the likelihood function can easily become the most time consuming step in the MCMC sampling, because we need to evaluate the likelihood millions of times. We have applied a simple linear interpolation method to reduce the computational time of the likelihood evaluation [57]. GP4Rate calculates the site-specific log likelihoods at a set of evenly spaced substitution rates and then approximates the site-specific log likelihoods at other rates by interpolation. Note that the linear interpolation is performed based on the site-specific substitution rates while is the vector of their log values, so an exponential transformation, i.e. , must be performed for each site i before the interpolation. By default, GP4Rate calculates and caches the site-specific log likelihoods at 4000 evenly spaced substitution rates, ranging from to 20. In each step of the likelihood calculation, if is between and 20, the corresponding site-specific log likelihood is approximated by the following formula,(9)On the right hand side, and are the two cached substitution rates which are closest to , while and are the site-specific log likelihoods of and , respectively. In practice, is rarely bigger than 20 or smaller than . If it is indeed outside this, the log likelihood at the closest boundary is used as the approximate log likelihood.
GP4Rate uses MCMC simulations to sample parameters from their posterior distribution. The algorithm follows previous studies by Neal [56], [58]. As described in the previous sections, the parameters in GP4Rate have two components. The first one is and the second one consists of and . In each iteration, the two components are sequentially updated by the Metropolis algorithm with symmetric proposals [23], [24].
To update , GP4Rate uses a proposal distribution suggested by Neal [56],(10)In the equation, is the current vector of site-specific log substitution rates while is the new proposal. is the Cholesky decomposition of the covariance matrix and is a vector of independent standard Gaussian variables. The proposal distribution is tuned by the constant, . A large leads to large changes of while small leads to small changes. is chosen to make the acceptance rate of new proposals close to 0.25.
Instead of updating and in the original scale, we transform them to the log scale. The use of a log scale removes the boundaries of the two parameters and makes the MCMC sampling of and independent from the scale of the data [56]. The two parameters are updated by a sliding window method with a bivariate Gaussian proposal [58]. The Gaussian proposal is tuned so that the acceptance rate of new proposals is close to 0.25.
In practice, the update of is much faster than the update of and , because the update of and requires a Cholesky decomposition whose time complexity is , in which is the total number of sites in the alignment. Therefore, it is reasonable to update more often than and [56]. In each iteration is updated 50 times while the pair of and is updated once. After every 10 iterations, the values of , , and are recorded.
|
10.1371/journal.pcbi.1003170 | Characteristic Effects of Stochastic Oscillatory Forcing on Neural Firing: Analytical Theory and Comparison to Paddlefish Electroreceptor Data | Stochastic signals with pronounced oscillatory components are frequently encountered in neural systems. Input currents to a neuron in the form of stochastic oscillations could be of exogenous origin, e.g. sensory input or synaptic input from a network rhythm. They shape spike firing statistics in a characteristic way, which we explore theoretically in this report. We consider a perfect integrate-and-fire neuron that is stimulated by a constant base current (to drive regular spontaneous firing), along with Gaussian narrow-band noise (a simple example of stochastic oscillations), and a broadband noise. We derive expressions for the nth-order interval distribution, its variance, and the serial correlation coefficients of the interspike intervals (ISIs) and confirm these analytical results by computer simulations. The theory is then applied to experimental data from electroreceptors of paddlefish, which have two distinct types of internal noisy oscillators, one forcing the other. The theory provides an analytical description of their afferent spiking statistics during spontaneous firing, and replicates a pronounced dependence of ISI serial correlation coefficients on the relative frequency of the driving oscillations, and furthermore allows extraction of certain parameters of the intrinsic oscillators embedded in these electroreceptors.
| We explore how a neuron responds to a special type of input signal which is oscillatory but noisy (narrow-band noise). These fluctuations could be due to sensory input, due to oscillatory activity of a surrounding network, or due to a natural stimulus. We study theoretically the effects of noisy oscillations on an idealized model neuron, which would otherwise produce as output a series of action potentials at regular intervals. Because our model is comparably simple, we can describe the effects on ISI statistics analytically with formulas that we test against computer simulations of the model. Moreover, we can compare our theoretical predictions to experimental data from electroreceptors of paddlefish - a biological example for spiking neurons that are naturally stimulated by stochastic oscillatory input. In particular, our theory provides a simple explanation of the seemingly complicated patterns of correlations between interspike intervals, that are observed for the electro-afferents in paddlefish; the theory shows also good agreement with respect to other independent spike train statistics. Our findings further the understanding of how nervous activity is shaped by oscillatory noisy signals, which can emerge in the neural networks of the brain, in the sensory periphery, and in the environment.
| Oscillatory activity is common in neural systems. Mechanical oscillations form an important class of sensory stimuli, for instance, in hearing, but may also be generated autonomously by mechanosensory hair cells [1]. In single neurons, periodicities may occur in the form of subthreshold membrane potential oscillations [2]. Oscillations at the level of brainstem and spinal cord neural networks generate the coordinated motor patterns for breathing and locomotion. Cortical networks may cause periodicities of local field potentials [3] or electroencephalogram (EEG) or magnetoencephalogram (MEG) activity [4].
With few exceptions, e.g. motor rhythms and the precise rhythm of the electric organ discharge in weakly electric fish [5], the oscillations generated by neural systems are not coherent over long time scales, but instead show fluctuations in both phase and amplitude (see Fig. 1, middle panel, for an example). Such periodic signals with limited coherence are termed stochastic oscillations, and are characterized by a preferred frequency band of spectral power. An individual neuron's activity may be affected by stochastic oscillations via synaptic input to it, or from its own endogenous fluctuations. Although stochastic oscillations are frequently found in neural systems, there is generally poor understanding of how an input current of this kind affects the firing pattern of a neuron, its ability to transmit information about time-dependent stimuli, and its interaction with other cells in a neural network. This is in marked contrast to the often studied (non-stationary) problem of how a deterministic periodic driving affects neural activity (see e.g. [6]–[13]). The simplest yet non-trivial problem that comes up with stochastic oscillations is how they shape the spontaneous activity of a spiking neuron, our topic here.
For the strictly periodic (i.e. a deterministic) driving, different analytical approaches and results exist (see e.g. [8], [11], [12]). Explicit expressions for the spike statistics of neurons driven by stochastic oscillations, however, are still lacking even for simple integrate-and-fire type models (for a notable exception, see [14] for an approach to the count statistics of such models). Formulas, e.g. for the ISI statistics, are desirable for several reasons. First, the analytical approach gives us a more thorough understanding of the spike time statistics, along with opportunities to formulate falsifiable predictions from the model. Secondly, in many neurons, a stochastic oscillatory drive may arise from noisy background processes rather than from specific sensory input. Analytical results may help to understand this more complicated situation of oscillatory noise and sensory stimuli being present at the same time. Put differently, before we can characterize the signal transmission of such a cell, it is in many cases beneficial to first thoroughly understand its spontaneous (i.e. autonomous signal-independent) activity caused by intrinsic noise or massive synaptic background. Thirdly, the temporal structure of single neuronal spike trains is conserved even if many independent spike trains are superposed [15] (weak correlations between neurons will additionally shape the power spectrum of the sum). Hence, on the network level, characteristics like the ISI density and ISI correlations of presynaptic cells driven by stochastic oscillations still affect postsynaptic target cells and thus shape network dynamics. Last, by comparing the ISI statistics of real neurons to analytical expectations, it may in certain cases be possible to draw conclusions about intrinsic parameters of the neural dynamics, which may otherwise be inaccessible, as has been carried out recently for sensory neurons with spike-frequency adaptation [16], [17].
Extensive experimental results pertinent to this problem of how stochastic oscillations shape the spontaneous spiking of a sensory neuron exist for the peripheral electroreceptors in paddlefish, which embed two distinct types of stochastic oscillators, one running at approx. 25 Hz, residing in a population of epithelial cells, which drives another in the peripheral terminals of afferents, running at approximately twice higher frequency. It was shown that the forcing from stochastic epithelial oscillations leads to rather complicated firing statistics of afferent firing, with multiple peaks in spike train power spectra, and extended-range correlations in the ISI sequence, continuing for tens of ISIs [18], [19]. We made use of a database of digitized recordings of spontaneous firing of electroreceptor afferents, obtained from in vivo paddlefish preparations in which external environmental noise due to water motion was minimized.
In this paper, we present novel analytical results for the firing statistics of a perfect integrate-and-fire (PIF) neuron model, which is driven by noisy oscillations [14]. The PIF model is the canonical model for a supra-threshold, regularly firing neuron, in which the effective mean input current is so strong that it overshadows any voltage-dependence of the subthreshold membrane dynamics. The membrane potential obeys the dynamical equation(1)where denotes the temporal derivative of . The model generates spikes whenever hits the threshold at and is subsequently reset to . The driving consists of a so-called harmonic noise, representing the stochastic oscillation, given by the following Langevin equations [20](2)(3)together with a short-correlated Ornstein-Uhlenbeck process [21](4)which mimics broadband intrinsic fluctuations. The values of the noise are not reset after spiking. Important parameters of the model are: (i) the frequency ratio of the damped frequency of the harmonic noise to the mean firing rate , (ii) the quality factor which quantifies the bandwidth and coherence of the harmonic noise, (iii) the non-dimensionalized variance of the harmonic noise , (iv) the non-dimensionalized variance of an Ornstein-Uhlenbeck (OU) broadband noise process , and (v) its non-dimensional correlation time . Our model with stochastic oscillations is illustrated in Fig. 1. Note that it can be regarded as a generalization of previous models, in which a PIF model was driven by uncorrelated white noise [22], exponentially correlated noise [16], [23], or a white noise and periodic driving [8], [12], [24], [25].
For this simple model, we calculate approximations for the ISI density and the ISI serial correlations and compare them to numerical simulations of the model. When discussing our explicit results, we focus on changes of the ISI statistics upon varying the ratio of the frequency of stochastic oscillations to the neuron's firing rate, a parameter that also shows a remarkable effect for the electroreceptor afferents of paddlefish. In particular, we show that upon variation of the skewness of the ISI density and also the structure of the ISI correlation coefficient as a function of the lag both change drastically, changes that are well-described by our theory.
We then compare our formulas for the ISI statistics to experimental data from electroreceptor afferents of paddlefish, obtained previously [18]. The analytical results from our simple perfect integrate-and-fire model work reasonably well in predicting (matching) these experimental data, indicating that the limitations of this model are not severe for representing sensory neurons with a high ongoing firing rate. This accords with other reports of remarkably good performance of stochastic perfect integrate-and-fire models for mimicking the ISI statistics of spontaneously active sensory neurons [17], [22] (for the performance of more general IF models in reproducing spike statistics, see e.g. [26], [27]). We conclude with a short discussion of the implications of our results for oscillatory physiological systems in general.
In this work, we aim at (i) the statistics of individual interspike intervals (ISI) by means of their probability density function (pdf), its coefficient of variation (CV), and its skewness, and (ii) the correlations between ISIs as quantified by the serial correlation coefficient (SCC). We study these statistics for the perfect integrate-and-fire (PIF) model and compare the theoretical results to experimental data.
Despite the apparent simplicity of the PIF model, the fire-and-reset condition severely complicates the analysis. For the calculation of the ISI density and ISI correlations, one has to solve a first-passage-time problem in the form of a Fokker-Planck equation in a four-dimensional state-space spanned by the voltage and all the noise variables. The fire-and-reset condition imposes a complicated boundary condition on a half-space [28], which however can be ignored in the case of a weak colored noise where the standard deviation of the total noise is much smaller than the base current , or, in terms of a small parameter , if(5)In this case, based on the methods presented in [16], [23], the solution with natural boundary conditions can be used to calculate the ISI density. Furthermore, to obtain explicit expressions for the ISI moments and the SCC, a perturbation calculation of the characteristic function, in which enters as the small parameter, turns out to be advantageous. These approximations are outlined in Methods and lead, for the considered problem, to formulas of reasonable length for the statistics of interest. In the next section, we compare our formulas to results from numerical simulations for small fixed values of . In Methods, we also show some of the statistics as functions of our small parameter in order to give the reader some intuition about the range of validity of our formulas.
Our results are valid for arbitrary time scales of harmonic noise and OU noise; the general formulas are provided in the Methods section. However, because the effects of an exponentially correlated noise on ISI statistics are well-known [16], [23], [29], we focus on variations in the time scales of the harmonic noise, and set the correlation time of the OU noise to a small value if not stated otherwise. In most of the cases discussed, the latter noise thus acts essentially as a white-noise source. Direct inclusion of a white noise is not possible in our perturbation approach.
The theory developed in the previous sections was applied to experimental data obtained from in vivo electroreceptors of paddlefish. A single peripheral electroreceptor (ER) in paddlefish embeds two distinct oscillators. One resides in a population of epithelial cells (epithelial oscillator, EO), and can be recorded near an epithelium. This EO is coupled synaptically to another oscillator associated with the terminal of a given primary sensory afferent neuron (afferent oscillator, AO) [18]. Unidirectional coupling of these self-sustained oscillators, EO→AO, results in spontaneous biperiodic firing patterns of afferents having two fundamental frequencies, including the EO's at about , and another corresponding to the mean firing rate of an afferent, , ranging from 30 up to 78 Hz, depending on the particular electroreceptor [18]. These two fundamental frequencies are seen as separate peaks centered at and in the power spectral density of an afferent's firing. These peaks were used to determine the frequency ratio of the two oscillators as . Only the AO is affected by external electric field stimuli. The EO acts as a stimulus-independent source of narrow-band noise input to the AO [34], [35]. Thus, the paddlefish electroreceptor system is an appropriate source of experimental data for validating the theory developed here.
In the in vivo preparation of paddlefish, an extracellular single unit recording offers information about the firing of an ER afferent. However, parameters of the epithelial oscillator, such as its effective quality factor and its variance, are hidden (Discussion). Previous computational studies have shown that a model of two unidirectionally coupled oscillators reproduces well the spontaneous and response dynamics of paddlefish ERs [14], [19], [36]. Here, we use our theory for the PIF model with harmonic noise, and in particular analytical expressions for the SCCs, to extract statistical and dynamical properties of the oscillators embedded in these ERs, and to verify the theoretical predictions of how the statistics of ISIs depend on the parameters of the coupled oscillators.
We analyzed spontaneous spiking activity from a sample of ER afferents (Methods). External noise was minimized, and a criterion for stationarity of long data segments was imposed. The data were in the form of sequences of spike times , , where was of the order of 15000–50000 spikes, corresponding to recording times of 300–900 s. The mean firing rates of these units were in the range of 37.9–77.7 Hz, with mean and SD of 54.3±8.71 Hz. The ratio of the EO to AO frequencies, , was 0.48 ± 0.06 for the sample (range 0.40–0.61). The CV of the corresponding ISIs of these units was 0.19 ± 0.05 (range 0.11–0.31). Histograms in Fig. 6 summarizing these statistics illustrate the diversity of firing rates and variability among this sample of ER afferents.
From a given spike sequence, we estimated the SCCs, the probability density of ISIs, and the power spectral density of the spike train. Fig. 7 shows these measures for three representative afferents with distinct values of the EO-to-AO frequency ratio, , which were below, near, or above . For all afferents in the sample, the distributions of ISIs were unimodal, and peaked close to the mean ISI (Fig. 7A). They all showed extended decaying series of significantly non-zero serial correlations, arising from the interaction of the EO and AO, with a structure determined by the frequency ratio [19]. To assess the variation of SCC values due to unavoidable minor non-stationarity, we split the spike train into 20 segments, each 2000 ISIs long, and estimated the SCCs for each segment, which yielded error bars for the SCC values shown in Fig. 7B. The PSD (Fig. 7C) showed a characteristic structure of peaks, with a peak at the fundamental frequencies of the EO and AO ( and , respectively), sideband peaks at combination frequencies (), and their higher harmonics [18], [19].
To apply our theory to experimental data, we extracted from the serial correlations of an afferent spike train the four parameters needed for the PIF model: the quality factor of the EO (a metric of the bandwidth and coherence of harmonic noise), the frequency ratio , the SD of the EO (the magnitude of harmonic noise), and the intensity of broadband OU noise, . These parameters were extracted using a fitting procedure described in the Methods (last section).
Fig. 7 illustrates the outcome for three representative ER afferents. Our theoretical expression for the serial correlations of ISIs, Eq.(12), shown by the red lines in Fig. 7B, provided excellent fits for experimental data, as fitted values were within the error bars of most experimental SCCs. The extracted parameters of the PIF model for these three afferents are listed in Table 1.
To calculate the probability densities of ISIs and the PSDs, we needed to accept a value for the correlation time of OU broadband noise, (which was in units of the mean ISI interval, i.e. ). This was the only free parameter in our procedure of comparison of experimental data and theory. Probability densities of ISIs calculated according to the theory Eq.(6) (solid lines in Fig. 7A), with the parameters from Table 1, showed good correspondence with experimental data, and weak dependence on the correlation time of OU noise. Instead of tuning up , the correlation time of OU broadband noise was assumed to be fixed at for all afferents, which provided good correspondence of experimental and theoretical ISIs distributions for all units, such as those shown in Fig. 7A.
Finally, Fig. 7C,D compares power spectra of spike trains obtained from numerical simulations of the PIF model Eqs.(1–4), using parameters from Table 1, to the PSDs of ER spike trains. Although the PSD's from simulations reproduced well the overall shape of experimental PSDs, the agreement between them is incomplete, especially at low frequencies, Hz, suggesting that the PIF model is an oversimplification of the stochastic dynamics of these electroreceptors. In particular, ER afferents in another fish species are known to exhibit spike-frequency adaptation resulting in short-term negative correlations [37]–[39]. These anticorrelations result in reduced power at low frequencies and a sharper peak at the mean firing rate. A previous study [19] showed that introduction of spike-frequency adaptation in a spiking model of paddlefish ERs results in an additional subtraction of low-frequency power similar to that observed in the experimental PSDs shown in double log scale in Fig. 7D. Nevertheless, the overall agreement of our simple and analytically tractable model is clear.
The quality of fit is further illustrated in Fig. 8A showing measured and calculated correlation lag of the ISI sequence. Furthermore, correspondence of theory and experiment is demonstrated in Fig. 8B for the skewness of the ISI distribution, an independent variability measure derived from ISI distribution. As seen from the figure, theory estimate was biased towards somewhat smaller values of the skewness. The Spearman rank correlation coefficient was for the correlation lags and for the skewness ( for both).
The application of the fitting procedure to our sample of ERs provided the following sample-averaged values for the parameters of the PIF model:
(range 8.570–29.46), (range 0.129–0.443), and (range 0.084–0.303). It is noteworthy that the SD values of the broadband OU noise, , were close to or even larger than SDs of the harmonic noise, . Nevertheless, the first-order approximation used in our theory was adequate to provide close correspondence with the experimental data, as seen in Figs. 7 and 8.
Next, we analyzed how the statistical properties of afferent ISIs depend on the parameters of epithelial oscillations. In contrast to analytical or numerical analyses, which allow studying the dependence of a given statistical measure versus a single control parameter, other parameters being fixed, here instead each experimental data point on the scatter plots of Fig. 9 carries a set of 4 measurable parameters (, , , ) with fixed values. Variation of these parameter values between different ERs allows qualitative tendencies to be clearly seen in the sample of experimental data points, and these trends can be compared to theoretical predictions. We start with Fig. 9A, showing a scatter plot of the electroreceptor ISI correlation lag (i.e. how slowly the serial correlations of ISIs decayed; Eq.(14)) versus the values of epithelial oscillations (i.e. their bandwidth and coherence). According to the theory, SCCs will decay more slowly for more coherent epithelial oscillations (i.e. for larger values), such that increases with . This prediction was supported by a positive correlation between them in the experimental data (Fig. 9A), having a significant Spearman rank correlation coefficient of (), despite considerable scatter.
To quantitatively compare experimental data to expectations from theory, we calculated versus from the PIF model Eq.(15) for each ER in the sample, with the other three parameters (, , ) extracted using the SCC fitting procedure (Methods, final section), yielding a curve for that electroreceptor over the range of values along the abscissa of Fig. 9A. The family of n = 56 PIF model curves were then averaged to calculate the sample-averaged tendency (solid blue line) and its standard deviation (dashed blue lines):(16)where , , and are the parameters for the k-th afferent. The mean trend from theory formed a straight line with positive slope, correctly predicting that the correlation lag increases for more coherent epithelial oscillations. More than expected of the experimental data points (47/56 = 84%) fell within the predicted ±1 SD (68%) bands.
The ISI variability metric also depended strongly on the frequency ratio (Fig. 9B), with the largest correlation lag attained for a value of close to 0.5, i.e. when there were two afferent spikes per cycle of epithelial oscillations. This is consistent with the PIF theory (solid blue line) showing a local maximum of the correlation lag for . Furthermore, most experimental points were within theoretical ±1 SD bounds (blue dashed lines, calculated in the same way as for Fig. 9A but by varying instead of ). The large scatter of data points presumably was due to diversity of afferent variabilities.
Finally, Fig. 9C shows that the skewness of ISI distributions was negatively correlated with the frequency ratio , such that the tails of ISI distributions were significantly reduced at higher values of (Spearman correlation coefficient , ). This negative correlation was borne out by analytical calculations from the PIF model (blue lines).
This report analyzed a scenario in which the membrane potential and spiking of a neuron is forced by weak noisy oscillatory input, in a narrow but non-vanishing frequency band. Our goal was to study the effects of narrow-band noise input on the output spiking statistics of a neuron. Our analysis centered around a perfect integrate-and-fire model of a single neuron, stimulated by a mixture of stochastic oscillations and broadband noise. We obtained novel explicit expressions for the probability density and serial correlation coefficients of the model's interspike intervals (ISIs). By a perturbation calculation of the Fokker-Planck equation, we derived a structurally simple form for the serial correlation coefficient. This novel derivation helps to solve the inverse problem: using the spike statistics of a neuron to estimate parameters of the underlying stochastic processes that drive its firing. No other body of theory has rigorously addressed the implications of narrow-band stochastic input for neural firing statistics, despite much acclaim of the widespread roles of oscillators in nervous systems.
Our new analytical formulas compare extremely well with results from our numerical simulations of the spiking neuron model, provided that the variance of the total input noise is weak, such that the coefficient of variation of spiking output remains low, less than approximately 0.3.
We compared the PIF theory to spike time data from a well-defined experimental system, the electroreceptor afferents of paddlefish, which receive stochastic synaptic driving in a narrow frequency band from ongoing oscillations arising in their sensory epithelia. For a given afferent's sequence of ISIs, a fitting procedure was used to extract four parameters needed for the PIF model, and the model's output was computed. The only appreciable discrepancy between model and experiment was observed in the skewness of ISI distributions in which the model showed consistently smaller values and in the low frequency regime of spike train power spectra, in which the model showed excess noise power. This low frequency regime is presumably shaped by spike-frequency adaptation, which we did not incorporate in our model to keep it analytically tractable.
The fitting parameters varied considerably for different units, reflecting natural variability of the electroreceptors. These natural ranges of values permitted us to check whether different functional relationships were correctly predicted by the theory. For example, in both theory and experiment, the temporal extent of the SCCs (i.e., their correlation lag, ) increases monotonically with the quality factor of epithelial oscillations (Fig. 9 A), whereas depends in a non-monotonic fashion on the frequency ratio , attaining a maximum at (Fig. 9 B), for both theory and experiment. We note that previous computational work showed that this frequency ratio corresponds to a maximum mutual information rate for electroreceptor afferents stimulated by a time-varying stimulus [36]. Thus, our study provides further arguments in favor of the idea that oscillators embedded in the electroreceptor system are tuned to maximize stimulus encoding [34].
We applied our formulas to the inverse problem of whether the spike statistics of a neuron can be used to estimate parameters of the underlying stochastic processes that drive its firing. Using only our formulas for the firing rate, the CV, and the serial correlation coefficient, we were able to predict the parameters of the epithelial oscillator () and the variance of the broadband noise, . Using these predicted parameters, our analytical formulas provided excellent fits to the experimental serial correlation coefficients, and close correspondence between model and experiment in their ISI distributions and power spectra (except at low frequency).
It could be argued that the suggested solution of the inverse problem is too cumbersome in the case of epithelial oscillations of paddlefish ERs. Power spectra of afferent spike trains show a second fundamental peak due to synaptic input at the frequency of epithelial oscillations, , so this spectral peak provides direct information. However, the peak is of limited usefulness for measuring parameters of the epithelial oscillation such as their quality factor, , because afferent spike train spectra incorporate the effects of nonlinear transformations during synaptic transmission and spike generation, and also because the peak may overlap with other spectral peaks, e.g. a sideband.
In general, the good agreement of the simple PIF model and the experimental data indicates that the detailed voltage dependence of the neural dynamics, more faithfully modeled in a conductance-based Hodgkin-Huxley model, is less important for the spiking statistics than the stochastic oscillatory driving, provided that the mean input to the neuron is appropriate for tonic firing with low ISI variability. The PIF model is able to reproduce several complex features observed experimentally in the afferent spike timing [18], [19], including skewing of ISI probability densities in different ways, oscillations, beating, or seemingly chaotic patterns in the serial correlations of interspike intervals as a function of the lag. The ability of our theory to reproduce complex non-renewal spike timing may encourage experimentalists to look for and analyze seemingly complex looking patterns in ISI correlations.
Examples of narrowband noisy neural oscillations include the gamma band (25–90 Hz) extracellular field potentials prevalent in mammalian cortex [40], which have been suggested, along with transient synchronization between brain areas, to mediate or reflect higher cognitive functions [4], [41]. Such fast gamma oscillations interact with slower rhythms, including the theta rhythm in hippocampus, and slower oscillations in thalamic nuclei. From the point of view of a single cell in a specific brain region engaged in a specific rhythm, input from other brain regions could be regarded as a stochastic oscillation. What matters the most for the ISI statistics of this cell may be not so much the synchrony of the activity but the frequency ratio between the stochastic oscillatory driving and the mean firing rate of the driven cell. Put differently, instead of coherence and synchronization, an important signal for cognition might be the frequency ratio of narrowband stochastic oscillations in related brain areas. Our work provides a rigorous demonstration and model of how the operation and spiking statistics of neurons can change sharply when the frequencies of different stochastic oscillatory components approach or assume an integer ratio (i.e. a rational number). Perhaps integer ratioing could function as a trigger or gate for cognitive, memory, or other information processes, acting like an event detector.
Specifically, our results show how the structure of a neuron's serial ISI correlations depends characteristically on the frequency ratio of weak stochastic oscillatory input, and the intrinsic periodicity of a neuron receiving the input, with extreme SCC behavior occurring at integer multiples. We have delineated other parameters which strongly affect SCCs including the quality factor of stochastic oscillatory drive (i.e. its bandwidth and coherence), the neuron's mean firing rate, and the overall level of spike timing noise (its CV). Our results bear general importance for the effects of weak stochastic oscillations on the spiking statistics of neurons in other systems, and are relevant to the study of neuronal firing in many brain regions. We have defined a basis in theory for using serial correlations to detect and characterize weak interactions of physiological oscillators, which may apply to other organ systems as well [42]–[44]. For example, the breathing and heartbeat rhythms can assume integer frequency ratios, and are known to be coupled [45].
We used conventional metrics, summarized here for clarity, to characterize the statistics of a stationary spike train given by the set of spike times . The spiking statistics can be derived from its sequence of interspike intervals (ISIs) , where denotes the i-th ISI. Calculations are simplified without loss of generality by restricting the stationary ensemble of spike trains to those realizations having a spike at time , called the zero-th spike. Under this choice of the origin, the n-th order interval, defined as the sum of consecutive ISIs, is equal to the n-th spike time:(17)The stationary spiking statistics can be formulated in terms of the statistics of the nth-order intervals, for all . Knowing the probability density of the n-th-order interval(18)for arbitrary , yields complete information about the spiking statistics.
The ISI probability density is given by the first-order interval density: . Let the mean ISI be denoted by , which is independent of the index due to stationarity (here and in the following, the notation refers to the ensemble average). Then, the mean of the nth-order interval is , and the variance is . The coefficient of variation (CV), defined as the ratio between ISI standard deviation and mean is given by(19)The CV is a measure of irregularity of the spike train; it is equal to one for a Poisson process. The statistics of individual ISIs are further characterized by the skewness defined by(20)Correlations among the ISIs are characterized by the serial correlation coefficient (SCC)(21)which depends on the order of ISIs. The SCC measures the correlations between two ISIs that are lagged by an integer . This measure can be related to the nth-order variances by the formula [23](22)
In this paper we gave parameters of the model simulation in terms of and . Here we provide the inverse relationship, how to obtain the simulation parameters and given and :(23)Using these values, one can easily determine the noise intensities:(24)
The mean ISI in the PIF model is independent of the properties of a noise with zero mean [23] and is given by(25)In fact, for large times , the spike count is determined by the free running solution of Eq. (1) (i.e. without resetting, cf. [29]): . Averaging this expression, the integral term vanishes and we obtain the firing rate , from which follows Eq. (25). Furthermore, using Eq. (17), we find that the mean n-th-order interval is given by .
To obtain higher moments as well as the probability density of , it is important to recognize that the nth-order intervals can be interpreted as a first-passage time (FPT). In fact, in the PIF model the statistics of the sum of subsequent ISIs for a firing threshold is equal to the statistics of a single ISI with respect to a firing threshold at . The statistics of a single ISI is, however, nothing else than the statistics of the FPT with respect to the boundary for a “particle” that starts at and is not reset at . The equivalence between n-th spike time and the FPT with respect to the boundary is due to the fact, that the “velocity” of the particle is independent of according to Eq.(1). Consequently, the time of the nth spike depends only on the total distance that a particle has to cover.
The FPT problem can be solved by using the Fokker-Planck equation for the probability density , which is associated to our stochastic model (see e.g. [46]). This equation reads(26)The probability density has to satisfy certain boundary and initial conditions. Specifically, we demand that particles that have crossed the boundary are not allowed to re-enter the domain (see [28] for a discussion on a related problem). This precludes repeated threshold crossings. As a consequence, there is no probability flux through the boundary with negative velocity. Mathematically, this entails the boundary condition(27)because no particles are found just below the boundary if . Furthermore, we require that the probability density and the probability current vanish at infinitely distant boundaries (natural boundary conditions). In the following, we assume that the total noise is weak. In particular, we require that the standard deviation of is much smaller than , or(28)where(29)are the normalized variances of and . Under this assumption, it is highly unlikely that becomes negative and hence, the boundary condition Eq.(27) can be safely neglected.
The initial condition is determined by the fact that at time the neuron has just fired a spike and the membrane potential has just been reset to . This implies, that the initial probability must satisfy(30)where is the probability density of the variables , and , upon firing. How can one obtain this probability density? To this end, let us for the moment reconsider the original setup, where the trajectories are reset if . Then the dynamics are restricted to the domain and the probability density will in this case converge to some stationary probability density, which will be denoted by . The density upon firing must be proportional to the fraction of particles that exit the domain through the surface element per unit time. This fraction is equal to , where is the stationary probability current in the direction. Thus,(31)Under the weak noise assumption Eq.(28), the stationary distribution does not depend on , because for all values have equal probability due to the voltage-independence of the membrane dynamics and the loss of the memory about the initial condition. We hence find(32)where . Upon normalization, the initial condition can now be written as(33)
The time-dependent solution of the Fokker-Planck equation (26) with the initial condition (33) can be related to the nth-order interval density as follows: The probability per unit time to cross the boundary at time is equal to the total probability current across the boundary at time , hence(34)For the sake of notational convenience, we will henceforth use the dimensionless time and membrane potential . Furthermore, we introduce the non-dimensionalized variables(35)and the non-dimensional parameters(36)In these rescaled variables the Fokker-Planck equation takes the form(37)
Here we briefly discuss the range of validity for our approximations. In general, we expect our theory to be valid whenever . Let us recall that(62)and, hence, we can increase by increasing only , only , or both simultaneously. In Fig. 10, we chose the first option, i.e. we vary only the harmonic noise strength. In the three panels of Fig. 10, we show the CV, the skewness, and the serial correlation coefficient at lag one as functions of and for three selected values of the frequency ratio . Varying both and for the ratios or 1 yield very similar results (not shown).
The plots illustrate that the theory works well for , confirming its general validity. For the statistics of the single ISI (CV, skewness), only minor deviations are found even for . This is not so for , which shows strong deviations for and can even reverse its sign for a strong harmonic driving at frequency ratios and . However, deviations of between theory and simulations can be neglected for and , which covers the experimentally relevant ranges of and for paddlefish electroreceptor afferents (data for and look very similar but are not shown).
Data from afferents of 19 animals were from experiments at University of Missouri-St. Louis in 2000–2002, under an IACUC-approved animal use protocol (W01-13) there. The spontaneous discharges of electroreceptor afferents of paddlefish (Polyodon spathula) were recorded in in vivo preparations with procedures detailed in [18]. A fish was held at rest in a plastic chamber, maintained by a stream of oxygenated water. The water temperature was maintained at 22°C. No external electric field or any other relevant kinds of stimulation were applied while recording spontaneous afferent firing. Disturbance of spontaneous afferent firing by the turbulence of water flowing into the mouth of a fish was minimized by partitioning the chamber [18]. Nonstationarity was further minimized by choosing segments of data in which a moving average of the afferent firing rate over a 10 s window fluctuated less than ±2% from the mean firing rate.
Analyses of spike time sequences from paddlefish electroreceptor afferents were performed using MATLAB's Signal Processing and Statistics Toolboxes. A spike train, , was represented as a sequence of delta functions centered at spike times of an afferent, with the mean firing rate subtracted: . For the purpose of estimating the power spectral density (PSD), each delta function was approximated by a rectangular pulse of width and height , where the sampling interval was set to 1 ms. The PSD, defined as , where is the Fourier transform of the spike train, was estimated using the Welch periodogram method (function pwelch of MATLAB's Signal Processing Toolbox).
The following procedure was used to extract 4 parameters of the PIF model (, , , and ) from an experimental sequence of ISIs:
|
10.1371/journal.ppat.1000796 | Broadly Protective Monoclonal Antibodies against H3 Influenza Viruses following Sequential Immunization with Different Hemagglutinins | As targets of adaptive immunity, influenza viruses are characterized by the fluidity with which they respond to the selective pressure applied by neutralizing antibodies. This mutability of structural determinants of protective immunity is the obstacle in developing universal influenza vaccines. Towards the development of such vaccines and other immune therapies, our studies are designed to identify regions of influenza viruses that are conserved and that mediate virus neutralization. We have specifically focused on viruses of the H3N2 subtype, which have persisted as a principal source of influenza-related morbidity and mortality in humans since the pandemic of 1968. Three monoclonal antibodies have been identified that are broadly-neutralizing against H3 influenza viruses spanning 40 years. The antibodies react with the hemagglutinin glycoprotein and appear to bind in regions that are refractory to the structural variation required for viral escape from neutralization. The antibodies demonstrate therapeutic efficacy in mice against H3N2 virus infection and have potential for use in the treatment of human influenza disease. By mapping the binding region of one antibody, 12D1, we have identified a continuous region of the hemagglutinin that may act as an immunogen to elicit broadly protective immunity to H3 viruses. The anti-H3 monoclonal antibodies were identified after immunization of mice with the hemagglutinin of four different viruses (A/Hong Kong/1/1968, A/Alabama/1/1981, A/Beijing/47/1992, A/Wyoming/3/2003). This immunization schedule was designed to boost B cells specific for conserved regions of the hemagglutinin from distinct antigenic clusters. Importantly, our antibodies are of naturally occurring specificity rather than selected from cloned libraries, demonstrating that broad-spectrum humoral immunity to influenza viruses can be elicited in vivo.
| Influenza viruses remain a formidable public health threat. Because of a dramatic increase in drug resistant strains of influenza viruses and due to the semi-regular emergence of pandemic virus strains, the development of novel antibody-based therapies and influenza vaccine constructs is of great interest. Recently, monoclonal antibodies with broad neutralizing activity against an array of Group 1 influenza viruses (including H5 and H1 subtypes) were identified; studies using these antibodies have expanded our understanding of structural aspects of the viral hemagglutinin, the molecule mediating protective immunity to influenza viruses. We have identified the first broadly neutralizing antibodies against viruses in Group 2—specifically, they are active against H3 influenza viruses spanning 40 years. The antibodies react with the hemagglutinin and appear to bind in regions that are refractory to the structural variation required for viral escape from neutralization. The antibodies demonstrate therapeutic efficacy in mice against H3N2 virus infection and have potential for use in the treatment of human influenza disease. By mapping the binding region of one antibody, 12D1, we have identified a continuous region of the hemagglutinin that may act as an immunogen to elicit an immune response conferring broad protection against H3 viruses.
| Under non-pandemic conditions, the global mortality attributed to influenza virus infection is considerable, with 200,000–500,000 associated deaths occurring each year [1]. In the setting of the 1918 influenza pandemic, the global mortality reached 50 million people in one year, equivalent to twice the number of people killed by HIV/AIDS since its emergence almost thirty years ago [2]. Notably, in 1918 and in the current swine-origin influenza virus pandemic, the populations normally considered the fittest are observed to be among the most vulnerable [3],[4].
Four kinds of influenza viruses are circulating in the human population at this time: influenza A viruses of the hemagglutinin H3 and H1 subtypes (H1 viruses are further divided into those of human and swine origin) and influenza B viruses. Influenza A viruses are responsible for the bulk of seasonal disease, with H3 viruses dominating eight of the past twelve influenza seasons in the United States [5]. In 1968, an H3 virus caused one of the three major influenza pandemics of the twentieth century and H3 viruses have persisted since that time as a significant agent of human disease. In addition to humans, H3 influenza viruses commonly infect birds, swine, and horses. It is not known whether H3 viruses will persist as human pathogens or how they may evolve to become more or less virulent in humans.
Immunity to influenza viruses is currently achieved by vaccination with strains representing those predicted to circulate in the coming flu season. In a healthy person, the virus acts as a robust immunogen, eliciting neutralizing serum antibody that protects against influenza disease. Both the humoral and cell-mediated arms of the adaptive system are involved in resolution of active influenza infection, with neutralizing antibody titers correlating with protection in vivo [6].
The hemagglutinin glycoprotein is the primary target of antibodies that confer protective immunity to influenza viruses. Antibodies to other influenza proteins likely act in: Fc-receptor mediated uptake of virus particles, antibody-dependent cell cytotoxicity, delay of replication kinetics and, in aggregate, they may contribute to virus neutralization. On a monoclonal level, however, only antibodies specific for the viral hemagglutinin have been shown to block/neutralize infection [7].
Neutralizing monoclonal antibodies (mAbs) act by preventing either of the two functions of the hemagglutinin molecule: virus attachment or virus fusion with the host cell [8]. Antibodies that prevent attachment bind antigenic sites surrounding the receptor binding pocket in the membrane distal HA1 subunit of the hemagglutinin and restrict the association with host cell receptors (sialic acids) [9]. These antibodies drive the outgrowth of antigenic variants, resulting in a continuum of changes in the hemagglutinin structure known as ‘antigenic drift’. Relatively few examples of fusion-inhibiting mAbs are available, but they are most commonly described to interact with the membrane proximal HA2 portion of the hemagglutinin in the region of the fusion peptide [10],[11],[12].
The sixteen subtypes of the influenza hemagglutinin are divided broadly into two phylogenetic groups that correlate with two basic structures taken by the stalk of the molecule [13]. In 1993, mAb C179, an antibody with broad neutralizing activity against viruses in Group 1 (of H1 and H2 subtypes) was described [14]. More recently, several other monoclonal antibodies that neutralize a broad array of Group 1 viruses (including representative H1 and H5 viruses) were identified [11],[12],[15],[16]. These antibodies have consistently been shown to interact with the stalk of the hemagglutinin and neutralize virus by preventing fusion with the host cell. This report constitutes the first description of broadly neutralizing antibodies against viruses in Group 2.
In order to enhance the production of cross-reactive antibody specificities, we immunized mice by sequential administration with DNA coding for the hemagglutinin from H3 viruses arising approximately 10 years apart: A/Hong Kong/1/1968, A/Alabama/1/1981, A/Beijing/47/1992. Finally, three days prior to fusion, mice were boosted with the H3 virus A/Wyoming/3/2003. By performing the fusion rapidly after virus boost we ensured that only hemagglutinin-specific B cells were present in the spleen at time of fusion. The hemagglutinins chosen were from viruses that arose over several decades, thus representing multiple H3 antigenic clusters [17]. Post-fusion, hybridoma supernatants were screened for the ability to bind A/Hong Kong/1/1968 by western blot or by ELISA and successive rounds of subcloning were performed on positive supernatants until monoclonal hybridoma populations were isolated.
The immunization schedule we utilized successfully elicited the production of antibodies with broad reactivity against H3 viruses. Approximately 120 clones were isolated that reacted with A/Hong Kong/1/1968; of those, eight mAbs were cross-reactive against all of the H3 hemagglutinins tested. Interestingly, the particular immunization protocol also preferentially elicited the production of antibodies specific for the HA2 subunit of the hemagglutinin. Of the 8 mAbs identified, 5 mAbs react with HA2 and 1 mAb reacts with HA1 by western blot. The remaining 2 mAbs bind conformational epitopes present in the HA trimer as detected by western blot of purified H3 virus proteins separated under non-reducing gel conditions. All mAbs were reactive in a purified H3 virus ELISA. Three of the mAbs, 7A7, 12D1, 39A4, had the highest activity by ELISA and were selected for thorough characterization (Table 1, Figure 1).
Antibodies 7A7, 12D1 and 39A4 react by ELISA with purified A/Alabama/1/1981 and purified A/Hong Kong/1/1968 viruses (Figure 2). MAb XY102 is specific for the hemagglutinin of A/Hong Kong/1/1968 virus [18]. MAbs 7A7, 12D1 and 39A4 show broad reactivity by immunofluorescence against cells infected with all H3 viruses spanning 40 drift years. MAbs 7A7 and 39A4 also react by immunofluorescence with other influenza A viruses chosen at random, including representative H1, H2 and equine H3 viruses (Table 2).
The anti-H3 mAbs were first evaluated for their ability to neutralize H3 influenza viruses by microneutralization assay. Viruses used in this assay contain a gene segment coding for firefly luciferase in place of the viral hemagglutinin; a hemagglutinin is present on the viral envelope due to propagation of virus in cells stably expressing a particular H3 hemagglutinin protein (see methods). Luciferase viruses were generated that express the hemagglutinin of A/HK/1968 or A/Panama/99 viruses. Neutralization of viruses by anti-H3 mAbs was determined based on luciferase activity after single-cycle replication; mAbs 7A7, 12D1 and 39A4 were determined to neutralize the hemagglutinin of both A/HK/1968 and A/Pan/99 (Figure 3).
Next, we evaluated neutralization activity by plaque reduction assay. The anti-H3 mAbs were able to prevent infection (not simply reduce plaque size) of Madin Darby canine kidney cells by H3 viruses arising over 40 drift years: A/HK/1968, A/BJ/1992, A/Pan/99, A/Bris/07, A/NY/08 (Figure 4). We tested 7A7, 12D1 and 39A4 against representative H4 and H7 viruses (Group 2) as well as an H1 virus (Group 1) and found that they did not neutralize the non-H3 subtype viruses (Figure 4).
The three mAbs were tested in vivo for use as passive transfer therapies in disease caused by H3 virus infection. Mice were given 30mg/kg mAb intraperitoneally either 1 hour before, 24 hours post or 48 hours post challenge with 10 mouse LD50 reassortant H3 virus (the A/HK/68 reassortant virus contains the six non-hemagglutinin, non-neuraminidase segments from the mouse-adapted A/PR/8 virus). Mice were weighed daily and were sacrificed if they reached 75% of their starting weight. Treatment of mice with mAb 12D1 either prophylactically or therapeutically was 100% protective. mAb 39A4 was evaluated for efficacy by prophylactic treatment and was similarly 100% protective in vivo. Mice treated prophylactically with mAb 7A7 were only 40% protected against the A/HK/68 reassortant virus (Figure 5).
Next, the effect of prophylactic treatment with mAb 12D1 or 39A4 on lung damage caused by H3 viral pneumonia was assessed by histologic evaluation of tissue taken 4 days post infection with the A/HK/68 reassortant virus. Without treatment, lungs showed degenerative changes with focal hemorrhaging, dense neutrophilic infiltrates and diffuse alveolar damage with edema. Treatment with either anti-H3 mAb significantly diminished pathologic changes (Figure 6).
Having demonstrated protective activity in vivo against the A/HK/68 reassortant virus we sought to evaluate cross-protection mediated by mAbs 12D1 and 39A4 against a second H3 virus, A/Georgia/1981. MAbs 12D1 and 39A4 were administered as described above to BALB/c mice one hour prior to infection. Mice were then infected intranasally with 2700 pfu A/Georgia/1981 and lung titers were evaluated two days post infection. The anti-H3 mAbs were found to reduce lung titers by 97.75% (12D1) or 99.03% (39A4) (Figure 7).
In order to determine the mechanism of virus neutralization by our anti-H3 mAbs, we first looked at the ability of the mAbs to inhibit virus hemagglutination of chicken red blood cells. We found that none of the three mAbs had hemagglutination inhibition activity, suggesting that the mAbs did not act by obstructing the binding of virus to the host-cell.
Next, we tested the effect of the anti-H3 mabs on virus fusion. MAbs 7A7, 12D1 and 39A4 were determined to inhibit the low-pH fusion of A/HK/1968 virus with chicken red blood cells by at least 80% at 10ug/ml (Figure 8).
Finally, we aimed to identify the region of the H3 hemagglutinin that might elicit antibodies with fine specificities mirroring those of 12D1 or 39A4. Sixteen passages of A/HK/1968 virus in the presence of the anti-H3 mAbs 12D1 or 39A4 did not yield escape variants that might have assisted in identification of the binding epitopes. Also, the hemagglutinin of six plaques present after incubation of A/HK/1968 virus with 50ug/ml mAb 12D1 or 39A4 in a plaque assay was sequenced and we were surprised to find no changes from the wild-type hemagglutinin.
Because mAb 12D1 mediates protection against influenza disease in vivo and reacts with a continuous epitope of the viral hemagglutinin (no trimeric structure required), as evidenced by reactivity with the denatured hemagglutinin monomer by western blot (Figure 1), we focused on identification of the 12D1 binding epitope. Hemagglutinin truncation mutants consisting of hemagglutinin segments of varying length fused to GFP were generated. GFP expression was utilized to assess expression of the constructs in transfected 293T cells. By analysis of the truncation mutants, it was determined that the 12D1 paratope makes dominant interactions with the HA2 subunit in the region of amino acids 30–106. Diminished 12D1 binding without diminished GFP expression in the 76–184 and 91–184 truncations along with loss of binding with the 106–184 truncation suggested that 12D1 binding is dependent on contacts with amino acids in the HA2 76–106 region (Figure 9). These 30 amino acids fall within the membrane distal half of the long alpha-helix of HA2 (Figure 10). The 12D1 paratope may have additional contacts with amino acids outside of this region (in HA1 or HA2) that are not required for binding by western blot.
For this study, we developed an immunization schedule that elicited broadly-neutralizing antibodies against H3 influenza viruses in vivo. The finding that such antibody specificities can be elicited by vaccination of mice suggests that with the proper immunogen(s) and vaccination protocol, such a response might also be elicited in humans. Several recent studies describe antibodies isolated from human phage display libraries that have cross-neutralizing activity against Group 1 influenza viruses [11],[12],[15],[16]. Mabs isolated from human display libraries have proved extremely useful in the characterization of structural epitopes that mediate heterosubtypic neutralization. Caveats to this methodology exist, however, since the diversity of combinatorial display libraries is typically orders of magnitude greater than the diversity of the true human variable region repertoire [19]. Additionally, phage display libraries are generated by random combination of immunoglobulin VH and VL genes and are therefore not restricted, as the in vivo repertoire is, by mechanisms regulating the production of auto-reactive specificities.
Until now, broadly neutralizing antibodies reactive with H3 viruses have not been described. Interestingly, mAbs 7A7 and 39A4 react by immunofluorescence with the hemagglutinin of multiple subtypes, though neutralizing activity appears to be limited to H3 viruses. Binding by these mAbs to other subtypes may be of relatively low avidity such that they no longer mediate neutralization, or, they may simply bind an epitope of non-H3 hemagglutinins that does not mediate neutralization. The identification of anti-H3 mAbs 12D1 and 39A4 complements recent works describing antibodies F10 and CR6261 that neutralize an array of Group 1 viruses [11],[12],[16]. One might envision a passive transfer therapy consisting of multiple broadly neutralizing mAbs for general use against pandemic and seasonal influenza virus strains. With the increasing resistance of influenza virus isolates to available anti-viral drugs, such an antibody cocktail could be of great value in severe disease.
Mouse monoclonal antibodies such as the anti-H3 mabs described herein are commonly used in the development of therapeutic antibodies for use in humans. Once characterized, rodent antibodies are readily humanized by methods typically involving grafting of non-human complimentary determining regions into appropriate human frameworks followed by cloning of variable region segments into complete human immunoglobulin constructs [20].
The fact that escape mutants were not selected after multiple passages of virus in the presence of anti-H3 mabs 12D1 and 39A4 is intriguing. Sui et al. reported that they were similarly unable to isolate escape mutants using their fusion-inhibiting mAb F10 [11]. MAb F10 makes multiple interactions with the hydrophobic pocket of the hemagglutinin including with the fusion peptide itself and prevents the low-pH triggered conformation change required for fusion. Considering the rigid structural and electrostatic requirements involved in membrane fusion, the hemagglutinin might not readily accommodate mutations at the F10 binding epitope. Anti-H3 mAb 39A4 binds a conformational epitope of the hemagglutinin trimer; the region of binding may bridge two monomers, therefore interacting with two different portions/faces of each monomer. A mutation at one region of contact (that does not affect trimer formation) may not be sufficient to ablate 39A4 binding. Anti-H3 mAb 12D1 likely binds within the long alpha-helix of HA2. This region may not accommodate changes that would affect 12D1 binding due to required secondary helix structure and specific van der Waals interactions that stabilize the hemagglutinin trimer [21]. Generally, mutations in the stalk of the hemagglutinin are more likely to affect the architecture of the entire molecule than are mutations in the classical antigenic sites [9].
The development of HA2-based vaccine constructs is of significant interest given recent reports of anti-HA2 mAbs with broad neutralizing activity against influenza viruses. Original studies of immunogens consisting of virus particles lacking the HA1 subunit demonstrated that design of an effective construct, however, will likely not be straightforward [22]. This is in large part due to the difficulty involved in maintaining the native configuration of the hemagglutinin stalk, which has complex tertiary structure and incorporates a portion of HA1 in addition to the HA2 subunit. Recent reports of mAbs with broad neutralizing activity against influenza viruses that are not active by western blot and that make contacts with amino acids in both HA1 and HA2 underscore the importance of maintaining non-contiguous epitopes in HA2 vaccine contructs [12],[16],[20].
In contrast to these mAbs, anti-H3 mAb 12D1 does not rely on a structural/non-contiguous epitope of the hemagglutinin stalk for binding. The observation that 12D1 makes dominant contacts within a continuous segment of the HA2 subunit suggests the design of an immunogen, perhaps consisting of that HA2 segment coupled to a carrier protein, that would direct an immune response to the region. The identified region, HA2 76–106, is 100% conserved between the H3 viruses used in this study and all other H3 viruses that we have examined. In contrast, the H1 viruses A/New Caledonia/20/99 and A/PR/8/34 share only 56.7% identity with the equivalent region in the H3 hemagglutinin. A vaccine construct incorporating this region, therefore, would likely not provide protection against H1 influenza viruses. This study and other structural studies [11],[12],[14] of the influenza hemagglutinin provide groundwork for the design of novel vaccine constructs aimed at providing broad-spectrum immunity to influenza viruses.
6 week old female BALB/c mice from Jackson Laboratory were used for all experiments. All animal procedures performed in this study are in accordance with Institutional Animal Care and Use Committee (IACUC) guidelines, and have been approved by the IACUC of Mount Sinai School of Medicine.
Madin Darby canine kidney cells from ATCC were used for all cell based assays. Cells were maintained in minimum essential medium supplemented with 10% fetal bovine serum, and 100 units/ml of penicillin-100 µg/ml of streptomycin. All viruses were propagated in eggs. Viruses used in various studies: A/Hong Kong/1/1968 (HK/68) (H3), A/Alabama/1/1981 (AL/81) (H3), A/Georgia/1981 (H3), A/Beijing/47/1992 (BJ/92) (H3), A/Wyoming/3/2003 (H3), A/Wisconsin/67/2005 (WI/05) (H3), A/Brisbane/10/2007 (BR/07) (H3), A/New York/2008 (NY08) (H3), A/Texas/36/1991 (TX/91) (H1), A/New Caledonia/20/99 (N.Cal/99) (H1), A/Duck/England/1962 (Dk/62) (H4), A/Turkey/England/1963 (Tky/63) (H7), A/Equine/Kentucky/2002 (e/KY/02) (H3), A/Ann Arbor/6/1960 (AA/60) (H2), A/Fort Monmouth/1/1947 (FM/47) (H1). Purified virus was prepared by high speed centrifugation (43,000 rpm, 1 hour) of allantoic fluid through a 20% sucrose cushion.
Hybridoma supernatants were used for screening of mAbs for reactivity by enzyme-linked immunosorbent assay (ELISA) and by western blot. For other assays, purified monoclonal antibody or ascites preparations treated with receptor-destroying enzyme [23] were used. RDE –treated ascites was used for measurement of binding by ELISA, microneutralization, plaque reducion and fusion assays. Antibodies were purified by methods previously described [24]. Because of differences in isotypes, Protein A-agarose (Roche) was used for purification of mAbs 7A7 and 39A4 while protein G-agarose (Roche) was used for purification of mAb 12D1.
6-week old BALB/c mice were immunized with DNA constructs coding for the open-reading frame of influenza virus hemagglutinin in the pCAGGS plasmid [25]. Individual immunizations were given intramuscularly, 3-weeks apart and consisted of 100ug DNA in 100ul PBS. Hemagglutinins utilized in the immunization schedule were cloned from the following parental viruses - primary immunization: A/Hong Kong/1/1968, secondary immunization: A/Alabama/1/1981, tertiary immunization: A/Beijing/47/1992 HA. Three days prior to fusion, mice were boosted with 50ug purified A/Wyoming/3/2003 virus. B cell hybridomas were produced by methods previously described [26],[27].
Hybridoma supernatants were screened by blot and by ELISA for reactivity with A/Hong Kong/1/1968 virus. For the ELISA, direct binding to wells coated with 5ug/ml purified virus, 50ul per well was assessed. For the blot assay, 10ug purified virus was adsorbed onto nitrocellulose strips and individual strips were incubated with hybridoma supernatants. For the ELISA and blot assays, binding of antibody to virus was detected using goat anti-mouse γ-chain horse radish peroxidase secondary antibody (SouthernBiotech, Birmingham, Al). All wells that had activity in either assay against A/Hong Kong/1/1968 virus were subcloned repeatedly to ensure the monoclonality of the hybridoma populations.
Blots were produced by methods previously described [28]. Samples were boiled for 5 minutes at 100°C in loading buffer containing SDS and 0.6M DTT. SDS migration buffer was used for electrophoresis. For non-reducing gel conditions samples were prepared in loading buffer with SDS but without reducing agent and were not boiled.
MDCK cells were infected with virus at a multiplicity of infection of 1 and incubated for 6 hours at 37°C. Infected and uninfected cells were incubated with 1ug/ml mAb for 1 hour at room temperature. Goat anti-mouse fluorescein conjugate (SouthernBiotech) was used for detection of mAb binding.
Two stable cell lines were generated that expressed the HA of A/Hong Kong/1/1968 virus or A/Panama/2007/1999 virus. Pseudotyped viruses expressing the HA of either cell line were generated by infection of cells with a virus that carries seven segments from A/WSN/33 virus (all except the HA segment) and one segment encoding Renilla luciferase. Pseudotyped viruses expressing the HA of A/Hong Kong/1/1968 virus or A/Panama/2007/1999 virus were used as the neutralization target. Viruses were incubated with mAb at room temperature for 30 minutes, rocking. Purified polyclonal mouse IgG (Invitrogen) was used for the negative control. The mixture containing virus and mAb was then transferred to wells of a 96-well plate seeded to confluency with MDCK cells and incubated for 12 hours at 37°C. Individual determinants were performed in triplicate. After incubation, luciferase activity in cell-lysates was measured as a read-out of virus infection (Renilla luciferase assay system, Promega).
Antibody and virus (∼50 pfu/well) were co-incubated at room temperature for 30 minutes, rocking. 6 well plates seeded with MDCK cells were washed once with PBS and 200ul of virus and mAb was added to each well then incubated for 20 minutes, 37°C. Virus with mAb was aspirated from cells and an agar overlay containing antibody was added to each well. Plates were incubated for 3 days, 37°C and plaques were counted by crystal violet staining. Purified mouse IgG (Invitrogen) was used for the negative control.
Before infection, mice were anesthetized by intraperitoneal administration of a ketamine (75 mg/kg of body weight)/xylazine (15 mg/kg of body weight) mixture. 6 week old BALB/c mice were given 30mg/kg mAb intraperitoneally either one hour before, 24 hours after or 48 hours after challenge with 10 LD50 A/Hong Kong/1/1968, A/PR/8/34 reassortant virus or 2700 pfu A/Georgia/1981 virus (lung titer experiment). Purified mouse IgG (Invitrogen) was used for the negative control. Virus was suspended in PBS and administered intranasally in 50ul (25ul per nostril). Mice were weighed daily and sacrificed if they fell to 75% of starting weight. For the lung titer experiment, mouse lungs were harvested 4 days post infection with A/Georgia/1981 and virus titers in lung homogenates were determined by plaque assay. For histologic evaluation of lung damage, lungs were harvested 4 days post infection with A/Hong Kong/1/1968 - A/PR/8/34 reassortant virus. Tissues were imbedded in paraffin and sections were stained with hematoxylin and eosin.
MAbs were tested in a standard hemagglutination inhibition assay [29] using chicken red blood cells and A/Hong Kong/1/1968 virus. For the red blood cell fusion assay, virus was incubated with chicken red blood cells (2% final red cell concentration) on ice for 10 minutes. Dilutions of antibody were added and samples were incubated on ice for 30 minutes. Sodium citrate buffer, pH 4.6 was then added to bring the final pH to 5.0 and samples were incubated for 30 minutes at room temperature. Samples were centrifuged for 3 minutes at 3000rpm to pellet red blood cells and supernatants were then transferred to an ELISA plate for determination of NADPH content by optical density measurement (340nm). NADPH was present in the supernatant as a function of fusion-induced red blood cell lysis.
DNA constructs were generated in the pCAGGS plasmid that coded for truncations of the A/HK/1/68 virus hemaggluinin fused to green fluorescent protein. All constructs were sequenced and confirmed. 293T cells were then transfected using Lipofectamine 2000 (Invitrogen, Inc.) with the various pCAGGS encoding the HA-GFP fusion gene. Cell lysates were resolved in a 4–20% Tris-HCl SDS-PAGE gel (Bio-Rad Laboratories) and proteins were blotted onto a Protran nitrocellulose membrane (Whatman). GFP and truncated HA fragments were detected using rabbit anti-GFP (Santa Cruz Biotechnology, Inc.) and anti-H3 mAb 12D1 respectively. Secondary antibodies were anti-rabbit IgG HRP (Dako) and anti-mouse Ig HRP(GE Healthcare).
|
10.1371/journal.ppat.1005445 | Multiple Novel Functions of Henipavirus O-glycans: The First O-glycan Functions Identified in the Paramyxovirus Family | O-linked glycosylation is a ubiquitous protein modification in organisms belonging to several kingdoms. Both microbial and host protein glycans are used by many pathogens for host invasion and immune evasion, yet little is known about the roles of O-glycans in viral pathogenesis. Reportedly, there is no single function attributed to O-glycans for the significant paramyxovirus family. The paramyxovirus family includes many important pathogens, such as measles, mumps, parainfluenza, metapneumo- and the deadly Henipaviruses Nipah (NiV) and Hendra (HeV) viruses. Paramyxoviral cell entry requires the coordinated actions of two viral membrane glycoproteins: the attachment (HN/H/G) and fusion (F) glycoproteins. O-glycan sites in HeV G were recently identified, facilitating use of the attachment protein of this deadly paramyxovirus as a model to study O-glycan functions. We mutated the identified HeV G O-glycosylation sites and found mutants with altered cell-cell fusion, G conformation, G/F association, viral entry in a pseudotyped viral system, and, quite unexpectedly, pseudotyped viral F protein incorporation and processing phenotypes. These are all important functions of viral glycoproteins. These phenotypes were broadly conserved for equivalent NiV mutants. Thus our results identify multiple novel and pathologically important functions of paramyxoviral O-glycans, paving the way to study O-glycan functions in other paramyxoviruses and enveloped viruses.
| Glycosylation is a protein modification process that occurs inside cells, in which specific types of sugars (glycans) are added to certain amino acids in some proteins. Glycosylation happens for many organisms, from microbes to mammals, including many pathogens. Altered glycosylation is increasingly being associated with auto-immune diseases and cancer, highlighting the need to better understand glycosylation. There are two types of sugars added during the glycosylation process: N-glycans and O-glycans. While the roles of N-glycans have been extensively reported for many organisms, the roles of O-glycans remain largely unknown, particularly for viruses. The paramyxoviruses are a medically important family of viruses that include the highly lethal Hendra (HeV) and Nipah (NiV) viruses. Viral entry into host cells and spread from cell to cell relies on two viral proteins: G and F. Here we mutated known O-glycan locations in the HeV and NiV G proteins. Loss of O-glycans affected several viral processes crucial to viral entry and spread from cell to cell. Our results are the first reported functions for paramyxoviral O-glycans, contributing to the field of viral entry and spread, and helping pave the way for future functional studies in other pathogens.
| Many microbial pathogens utilize protein glycosylation for host invasion and immune evasion [1]. Although many N-glycan functions have been reported, relatively little is known about the roles of O-glycans in microbial pathogenesis or biology, particularly for viruses. O-linked glycosylation is a ubiquitous protein modification in organisms belonging to several kingdoms. For example, O-glycans play roles in protein trafficking, signaling, cell-cell interactions, and receptor binding for host proteins [2–4], and O-glycans are important for developmental processes and immune system functions [3]. Additionally, altered O-glycosylation has been linked to illnesses such as autoimmune diseases and cancer [5–7], as well as pathogen virulence [8–10]. Yet currently the specific functions of O-glycans on viral glycoproteins are not well understood, and to our knowledge there is no single function attributed to O-glycans for the important paramyxovirus family.
Conversely, viral glycoprotein N-glycans are known to be critical for proper protein folding and trafficking, receptor interactions, cell adhesion, and evasion of host immune responses [reviewed in [1]]. In addition, loss of paramyxoviral N-glycans reduces or increases membrane fusion capacity and processing of the F glycoproteins of measles virus (MeV), NiV, and Sendai virus [11–13], and N glycans on NiV F and G modulate membrane fusion and viral infectivity, and protect the virus from antibody neutralization [12,14–17]. Additionally, galectin-1, a lectin that binds specific N-glycans and O-glycans, inhibits NiV cell-cell fusion when added post-infection, but can enhance viral entry into endothelial cells by increasing viral attachment to target cells [18–20].
While potential N-glycosylation sites are marked by a distinctive N-X-S/T motif (where N is asparagine, X is any residue except proline, S is serine, and T is threonine), the determinants that cause addition of O-glycans to S or T residues are incompletely understood. Moreover, O-glycosylation of one S or T can affect O-glycosylation of other S or T residues [21]. O-glycans can act as an antibody shield for the gammaherpesvirus bovine herpes virus (BoHV-4) [22], affect binding of herpes simplex virus 1 (HSV1) gB to the paired immunoglobulin-like type 2 receptor alpha (PILRα) receptor and hence PILRα-dependent viral entry [23,24], and O-glycans on HSV1 gC-1 are thought to affect receptor binding and cell-cell spread [25]. For HSV-2, viral O-glycans have been recently shown to stimulate an antiviral immune response upstream from interferons [26]. Additionally, an O-glycan-rich region of Ebola virus GP plays a role in cell detachment [27], though this function has yet to be linked directly to the O-glycans themselves. O-glycans are present on the human immunodeficiency virus (HIV) and simian immunodeficiency virus (SIV) envelope, Marburg spike protein, mouse hepatitis virus (MHV) E1 protein, and vaccinia hemagglutinin (HA) [28–32], but the functions of these O-glycans are unknown. For the paramyxoviruses, it is known that O-glycans are not involved in respiratory syncytial virus (RSV) G oligomerization [33] nor Newcastle disease virus (NDV) receptor binding [34]. To our knowledge, however, no study has reported any functions for paramyxoviral O-glycans.
The Paramyxoviridae family is comprised of negative-sense single-stranded enveloped RNA viruses including measles (MeV), respiratory syncytial (RSV), Newcastle disease (NDV), parainfluenza (PIV), metapneumo- (MPV), and the deadly Henipaviruses (HNV) NiV and HeV among others [35]. HeV and NiV are biosafety level 4 (BSL4) pathogens with no approved vaccine or treatment for humans. Individuals infected with NiV or HeV suffer from respiratory distress and encephalitis, with 40–90% mortality rates [36]. A hallmark of HNV viral infections is extensive formation of syncytia (cell-cell fusion). These viruses can be transmitted from animal-to-animal, animal-to-human and human-to-human [37], highlighting a need for understanding the mechanism of HNV spread between individuals and within infected individuals.
Paramyxoviruses use two transmembrane glycoproteins to facilitate both viral-cell fusion during viral entry, and the pathognomonic syncytia formation. The attachment (H, G, or HN) protein binds the cell receptor, in turn activating the fusion protein (F) to insert its hydrophobic fusion peptide into the host cell membrane and execute viral-cell membrane fusion, resulting in viral entry into the host cell [38]. Via a similar mechanism, F and H/G/HN also cause cell-cell fusion of infected-healthy cells. For HNV, the attachment protein G binds the cell receptor ephrinB2 or ephrinB3 [39–41]. HNV G consists of an N-terminal cytoplasmic tail followed by a transmembrane region, a stalk domain, and a C-terminal globular head (Fig 1A), responsible for receptor binding [42].
The H/G/HN stalk plays an important role in paramyxovirus-induced membrane fusion, executed by the F glycoprotein. Disrupting the H/G/HN stalk inhibits fusion [43–45], and headless mutants are capable of causing fusion, underlining that the stalk itself is key in the fusion triggering process [46–49]. Recently, multiple specific O-glycan structures in a stalk domain of HeV G were identified [50], but the functions of these O-glycans are unknown. Since the HeV G stalk plays an important role in viral-cell and cell-cell membrane fusion, and O-glycan structural data is available for HeV, we used HeV G as a model to investigate the functions of O-glycans in henipaviral and paramyxoviral biology.
We mutated each O-glycosylated Ser or Thr in the heavily O-glycosylated HeV G stalk domain 103–137 to alanine (Ala), to prevent O-glycosylation at specific sites on the protein. Loss of these O-glycans affected membrane fusion levels, resulting in both hyper- and hypo-fusogenic phenotypes. Multiple mutants displayed altered HeV G/F association, viral entry capabilities in a pseudotyped virus system, and F incorporation into pseudotyped virions, and one mutant had an altered conformation and altered receptor-induced conformational changes important for membrane fusion. Furthermore, equivalent mutants in NiV G displayed highly-conserved roles, with most mutants replicating the HeV phenotypes, although some distinct photypes were observed for NiV, suggesting overall high conservation of O-glycan functions for this genus. This study is the first to uncover multiple functions of O-glycans in the paramyxovirus family, identifies the highest number of functions of O-glycans for any single virus, and points to differences between N- and O-glycan functions for the paramyxoviruses.
The specific locations and structures of O-glycans in HeV G have been identified via mass spectrometry [50], and except for one, all were found in the stalk. We mutated each O-glycosylated S or T residue in the heavily O-glycosylated stalk domain 103–137 to an alanine (A) to prevent O-glycosylation at each specific site (Fig 1A). Multiple adjacent O-glycosylation sites were mutated simultaneously to limit the number of target mutants, thus we initially constructed nine mutants. The S116A-T117A double mutant was then split into two single mutants, S116A and T117A, respectively, for a total of eleven mutants.
We first tested the ability of the mutant HeV G proteins to induce cell-cell fusion. For consistency, we used 293T cells to provide a similar cell background as that used for the HeV G O-glycan structural studies by Colgrave et. al. [50]. 293T cells were transfected with wild-type (wt) HeV F and either wt HeV G or an O-glycan mutant. After 18–24 hours post-transfection, nuclei in syncytia (fused cells) were quantified microscopically, with a syncytium defined as four or more nuclei within a single cell membrane, and compared to the wt G [49]. Interestingly, five of the eleven mutants showed differences in cell-cell fusion, with reduced fusion levels for T103A and T119A (2% and 0%, respectively) and increased fusion for T117A, S116-T117A, and S129A (232–316%). These mutants are the first to show that viral O-glycans can have an effect on modulating the extent of cell-cell fusion (Fig 1B). Interestingly the single mutant S116A did not yield an altered fusion phenotype, suggesting that the hyper-fusogenicity of the S116A-T117A double mutant is largely due to the S117A substitution (Fig 1B).
We then used flow cytometry to ask if these fusion phenotypes were due to altered cell surface expression (CSE) levels of the mutant HeV G proteins. 293T cells transfected with either wt or mutant G plasmids were collected 18–24 hours post-transfection and stained with a 1° antibody to the extracellular C-terminal hemagglutinin (HA) tag in the proteins, to avoid false conformational influences on the mutant’s real CSE levels, followed by a fluorescent 2° antibody. CSE of mutant HeV G proteins were normalized to the values of wt HeV G. All mutants were expressed at the cell surface at levels comparable to that of wt HeV G (87–115%, Fig 1B). Normalized fusion levels were then divided by normalized CSE values to obtain a fusion index, with hyperfusogenic mutants defined as having a fusion index ≥1.5, and hypofusogenic mutants having a fusion index ≤0.5 (Table 1) [14]. All mutants also bound soluble B2-hFC as detected by flow cytometry at levels similar to their respective CSE values (Fig 1B), which is expected since the G head, not the stalk, is responsible for ephrinB2 binding [42,51]. These results suggest that the mutants’ altered membrane fusion capabilities were not due to differences in CSE or receptor binding. In addition, Western blot analysis using an anti-HA antibody confirmed that all mutants were expressed in cell lysates (Fig 1C).
The stalk domains of HeV and NiV are highly conserved, with 90% amino acid identity between the two. After observing that loss of O-glycans in HeV G affected cell-cell fusion, we created equivalent mutants in NiV G. Thus we generated the same eleven NiV G mutants, with the exception of the HeV G double mutant S136A-S137A, which became a single mutant S137A in NiV, as residue 136 in wt NiV is already an alanine. Remarkably, we observed very similar overall fusion phenotypes for NiV, with the T103A and T119A mutants being hypo-fusogenic (6% and 0% fusion, respectively) and T117A, S116A-T117A, and S129A being hyper-fusogenic (212–350% fusion). CSE levels for these mutants were similar to HeV as well, with the exception that T119A expressed at the cell surface at only ~36% in NiV compared to ~87% in HeV, and S110A did not express at all for NiV (Fig 1D). Fusion indices for the NiV G mutants showed similar patterns to HeV as well (Table 1). All NiV G mutants bound soluble B2-hFC at levels similar to their CSE levels (Fig 1D). Western blot analysis of the NiV G mutants showed that, again, all with the exception of S110A were expressing within cell lysates (Fig 1E), including T119A, which was expressed poorly at the cell surface, suggesting that T119A’s reduced CSE may be due to an alteration in trafficking to the cell surface. Altogether these results are the first demonstration of a role for O-glycans in modulating paramyxovirus-induced membrane fusion, and suggest this function is highly conserved among the henipaviruses.
Though the specifics of the HNV F and G spatiotemporal interactions are still largely unknown, the available evidence suggests that HNV F and G interact on the cell or viral membrane prior to receptor binding, after which F and G dissociate from each other, triggering F to execute fusion (dissociation model) [38,52,53]. We tested the interactions of the HeV G and NiV G O-glycan mutants with HeV or NiV F, respectively, through a co-immunoprecipitation (co-IP) assay. 293T cells were co-transfected with wt HNV F and wt or mutant HNV G. After 16–20 hours cells were collected in lysis buffer and applied to a co-IP column (Miltenyi Biotec). Cell lysates and co-IP eluates were separated by SDS-PAGE and blotted using rabbit anti-HA (to detect G) and mouse anti-AU1 (to detect F) 1° antibodies, followed by fluorescent 2° antibodies. HNV F is produced as an uncleaved precursor, F0, that is cleaved by cathepsin L upon endocytosis into F1 and F2 subunits, linked by a disulfide bond [54,55]. The F0 (precursor) and F1 (processed) bands are both AU1-tagged and thus visible in Western blots, resulting in two bands for F and one for G (Fig 2). Co-IP values calculated via quantitative Western blot analyses (Fig 2E) revealed that the hyper-fusogenic mutants all had decreased F0 association (0.64–0.74 co-IP values), though this decrease in association was not observed for the F1 band, suggesting that removing O-glycans from these residues may alter intracellular F/G interactions, and to a lesser extent or not at all surface interactions (Fig 2). The nature of co-immunoprecipitation assays to study membrane protein interactions, including the use of detergents and the artificial association of membrane proteins via micelle formation, may account for the lack of differences in G/F1 interactions for the mutants, as further discussed in our Discussion section. Thus, even though the G/F1 interactions do not seem to be affected, it is possible that the altered F0/G interactions suggest overall effects in the G/F interactions, which generally support the dissociation model for HNV. These results were consistent for both the HeV G and NiV G mutants. Most likely, however, our G/F1 interaction data is consistent with the hyper- or hypo-fusogenic phenotypes of the O-glycan mutants not being a result of the effects of the O-glycan mutants on F/G interactions.
It is possible that loss of O-glycans could affect glycoprotein conformations, so we next tested if differing conformations correlated with the altered fusion phenotypes we observed. We have previously generated conformational antibodies to NiV G, some of which also bind HeV G. The first, Mab45, binds at the base of the head near the stalk, while the second, Mab26, binds the top of the head near the receptor binding site of both NiV and HeV [49,56]. We tested the binding of these antibodies to HNV G via flow cytometry and found that, while the majority of the mutants had antibody binding profiles similar to those of the wt HNV G, the fusion dead T119A mutant showed decreased Mab26 binding, suggesting that this mutant has an altered conformation for both viruses (Fig 3A and 3B).
Mab45 and Mab26 can also measure receptor-induced conformational changes in HNV G that are important to trigger F. We previously showed that upon receptor binding a conformational change occurs in NiV G region 9 (bound by Mab26), followed by a conformational change in region 4 (bound by Mab45), leading to uncovering of a specific domain at the C-terminus of the stalk, which triggers F. Moreover, mutants unable to fully undergo these conformational changes were hypo-fusogenic, demonstrating that these conformational changes in G are important for fusion [49]. These conformational changes can be measured by comparing antibody binding in the absence vs. presence of soluble ephrinB2 receptor binding. With no receptor present, Mab26 binding to G is relatively higher, but decreases as more ephrinB2 is added. Mab45 exhibits the opposite trend, with binding of Mab45 to G increasing upon the addition of soluble ephrinB2 [49].
Thus we tested Mab26 and Mab45 binding to wt or mutant HNV G at 0nM and 100nM ephrinB2. 100nM values were normalized to those at 0nM to represent the amount of increased or decreased antibody binding in the presence of receptor. All HeV mutants showed a decrease in Mab26 binding and an increase in Mab45 binding to some degree, suggesting the mutants are all still able to make the conformational changes involved in fusion (Fig 3C). Interestingly however, when the same assays were performed with the NiV G mutants, the T119A mutant showed decreased Mab45 and Mab26 binding in 100nM ephrinB2 compared to wt G (Fig 3D), suggesting that the hypo-fusogenicity of this mutant may in part be due to an inability to undergo the full range of conformational changes required for fusion. These data, combined with T119A’s decreased Mab26 binding in the absence of receptor, suggest that this mutant is in a near post-receptor binding conformation, preventing it from undergoing the Mab45 binding enhancement indicative of a full-range conformational switch from pre to post receptor binding conformations.
In addition, all mutants co-migrated with wt HNV G when analyzed by semi-denaturing SDS PAGE, with all relative levels of tetramers, dimers, and monomers not significantly affected by the loss of an O-glycan as determined by densitometry (S1 Fig), suggesting that O-glycan loss does not affect G oligomerization. This is important as proper oligomerization is critical for glycoprotein function, and the G stalk is an important determinant of G oligomerization [16,45]. Overall, the majority of the O-glycan mutants’ fusion phenotypes were not due to aberrant G conformations or abilities to undergo receptor-induced conformational changes, with the exception of mutant T119A.
As we found that loss of O-glycans modulated cell-cell fusion, we next tested if this affected viral entry, using a validated BSL2 pseudotyped virus infection assay [14]. Vesicular stomatitis virus (VSV) virions containing a Renilla luciferase gene in place of the VSV glycoprotein G gene, and expressing HNV wt F with either wt or mutant HNV G on the outer membranes were created as described previously [14]. These pseudotyped virions were tested for viral entry using a Renilla luciferase reporter assay. Vero cells were infected with the pseudotyped virions and incubated for 2 hours at 37°C. Growth media was then added, and at 18–24 hours post infection the cells were collected, lysed, and luminescence was determined. Relative light units (RLU), representing viral entry for each of the mutants, were compared to those of wt HNV G. Western blot analysis of pseudotyped virion lysates were also performed to assess protein incorporation levels.
Remarkably, Western blot analysis showed that, although G incorporation for each of the HeV G mutants was similar to wt G, F incorporation was reduced and F processing was enhanced for all G mutants except T119A (Fig 4A). Densitometry confirmed this observation; all mutants except for T119A incorporated HeV F at levels only 19–24% of that observed for wt pseudotyped virions and increased processing of F, as determined by the ratio of F1/total F, which was 86–99% for the mutants as compared to 54% for wt pseudotype virions (Fig 4B). Viral entry assays showed that all G mutants displayed reduced entry levels as compared to wt G, once again with the exception of T119A (Fig 4C), even though we would expect the three hyper-fusogenic mutants to enter at levels higher than that of wt HeV G. Reduced pseudotyped virion HeV F incorporation levels likely contribute to the decreased viral entry levels. To our knowledge, this is the first report of a paramyxovirus HN, H, or G mutant affecting F incorporation into virions and thus viral entry, or affecting F processing. It is possible that loss of an O-glycan therefore affects intracellular F/G interactions and/or glycoprotein trafficking, as well as F processing, consistent with our co-IP results (Fig 2). It is noteworthy that the effect of G O-glycan mutants on F processing was not observed in cell lysates (Fig 2). Additionally, flow cytometry using a variety of anti-F antibodies revealed no significant difference in F cell surface expression in the presence of HeV or NiV G (S2 Fig), a finding that also suggests that the altered interactions observed by co-IP (Fig 2) were not due to differences in F cell surface expression.
Interestingly, Western blot and viral entry assays of the NiV pseudotyped virions did not reveal the same phenotypes. All pseudotyped virions had similar levels of NiV F as assessed by Western blot analysis (Fig 5A) and by densitometry (Fig 5B), and with the exception of T119A, which had less G incorporation, had viral entry values similar to that observed for wt G (Fig 5C). F processing, however, was still altered (Fig 5A and 5B). The hyper-fusogenic mutants T117A, S116A-T117A, and S129A all had higher processing (F1) levels (69%, 55%, and 55%, respectively), whereas the wt (39%) and hypo-fusogenic mutants T103A and T119A had lower (27% and 29%) processing (F1) levels (Fig 5B). These data further corroborate that HNV G O-glycan loss can affect HNV F in pseudotyped virions, and implies that the mechanisms may differ between HeV and NiV, since F incorporation was affected more for HeV than for NiV. Since equivalent mutant NiV pseudotyped virions are able to enter cells at wt levels with proper F incorporation, the data from NiV pseudotyped virions also further supports the explanation that reduced F incorporation is responsible for the reduced pseudotyped viral entry levels observed in the HeV mutants, at least to some degree. Other potential mechanisms, such as the contribution of O-glycans to the stability of virions, are further explored in the Discussion. Interestingly, the S129A mutant also had less G incorporation (Fig 5A), though it was still able to enter cells at levels similar to wt G, suggesting that with full G incorporation this mutant may have had entry levels greater than the wt G pseudotyped virions.
In summary, we have found that HNV O-glycans had roles in cell-cell fusion, G/F interactions, G conformation, pseudotyped viral entry, and F incorporation and processing in pseudotyped virions. These important viral glycoprotein phenotypes were mostly conserved among HeV and NiV, but some important differences were observed between NiV and HeV O-glycan functions.
Viral O-glycans remain understudied, and the present study is the first to show functions for paramyxoviral O-glycans. In mammalian cells, O-glycans have functions in receptor binding, cell-cell interactions, cellular trafficking, and immune system regulation [2–4]. Viruses use glycosylation for protein trafficking, signaling, folding, immune evasion, and receptor interactions [1], but the majority of these functions have been described for N-glycans. The lack of clear O-glycosylation prediction motifs, and the fact that utilization of some O-glycan sites can affect O-glycosylation of neighboring sites [21], makes analysis of specific O-glycan functions much more complex. O-glycans are present on HIV, SIV, Marburg, MHV, and vaccinia virus glycoproteins [28–32] but their functions are still unknown. While some studies have shown that O-glycans are involved in immune evasion, receptor binding, plaque formation, and possibly cell detachment [22,25,27], this has not been described for paramyxoviruses, although prior work has shown that O-glycans do not play roles in RSV-G oligomerization [33] nor NDV receptor binding [34]. The present work is the first to show functions for paramyxoviral O-glycans, and one of only a handful of studies demonstrating functions of O-glycans for any virus.
Interestingly, with one exception, all the O-glycans identified in HeV were found in the G stalk domain [50]. Though it is still unclear which regions of HNV G are necessary for fusion, there is strong evidence that the HNV G stalk plays a central role in F triggering. Headless mutants can trigger fusion in PIV5, MeV, NDV, MuV, and NiV [46–49], and disruption of the stalk using various methods perturbs fusion in NiV, MeV, PIV5, and NDV [43–45,57–60]. Here we demonstrated that loss of O-glycans from the stalk of a paramyxovirus attachment protein affects fusion, and in all cases but one, this was not due to altered CSE or receptor binding levels (Fig 1B and 1D). Interestingly, previous studies showed that removal of certain N-glycans in HNV F or the G head resulted in hyper-fusogenicity [12,14–16], while removal of one N-glycan from the G stalk resulted in hypo-fusogenicity [15,16], without affecting receptor binding. Our study demonstrates that O-glycans in the stalk are capable of both up- or down-modulating fusion.
There are currently two different interaction models for the two paramyxovirus glycoproteins. In the association model proposed for PIV5 and NDV, for example, F and HN do not interact at the cell surface until receptor binding occurs, after which HN/F association occurs and F is triggered. In contrast, in the dissociation model proposed for MeV, NiV, and HeV, H/F or G/F interact prior to receptor binding; after which the glycoproteins dissociate from each other, triggering F and allowing fusion [38,52]. Reportedly, for HNV hyper-fusogenic and hypo-fusogenic mutants have shown decreased or increased F association, respectively, while receptor binding resulted in partial G/F dissociation [14,56,61], findings that support the dissociation model. Our co-IP data further support this dissociation model for HNV. The hyper-fusogenic mutants had decreased F interactions with F0, though this decrease did not extend to the F1 subunit (Fig 2), suggesting that the altered interactions may only be with the immature F protein. These results present the possibility that the O-glycans themselves are directly involved in the association between G and F, though these altered interactions are unlikely to account completely, if at all, for the observed fusion phenotypes since G/mature F (F1) interactions were not significantly altered. Additionally, the lack of difference in G/F1 association between mutants suggests that fusion phenotypes may be independent of F/G association levels. It is also possible, however, that G/F1 interactions are altered but not detectable via co-IP. Since co-IP relies on the use of detergents that may alter protein conformations and therefore interactions, results obtained by co-IP are not entirely conclusive. In addition, co-IP can’t distinguish pre vs post-fusion forms of the F protein. Our results, therefore, present the possibility that the observed fusion phenotypes are independent of F/G associations, but it is also possible that differences in mature F/G surface interactions are too small or transient to be observed by co-IP. New methodologies may be needed assess such interaction differences.
Mutation of a single O-glycan addition site in the majority of mutants did not alter protein conformation nor the conformational change capabilities necessary for fusion triggering [49]. The exception was mutant T119A, for which both HeV and NiV G bound Mab26, which detects a pre-receptor binding conformation, at reduced levels (Fig 3). This same mutant in NiV also failed to show enhancement of Mab45 binding upon addition of ephrinB2, suggesting that this mutant may already be in a post-receptor binding conformation at least partially, contributing to the loss-of-fusion phenotype. These data also suggest that retention of G in a pre-receptor binding conformation is important for its F-triggering capability. The HeV T119A mutant still showed enhanced, albeit less than wt HeV G, Mab45 binding after ephrinB2 addition, suggesting that, while HeV T119A may have an altered conformation, implied by decreased Mab26 binding, this conformational change may not be as functionally detrimental as that of NiV T119A. Even though HeV and NiV G are highly homologous, this hints at conformational differences between the two G glycoproteins. This may also relate to the differing CSE levels of these two mutant proteins (87% for HeV and 36% for NiV T119A) (Fig 1B and 1D). It is also noteworthy that loss of an O-glycan did not affect oligomerization for any of these mutants, an interesting finding since O-glycans are in the stalk domain of HNV G, which is crucial for G oligomerization [45].
Although cell-based assays, such as cell-cell fusion, CSE, ephrinB2 binding, F interactions, and conformational change capabilities, yielded similar results for HeV and NiV, we observed more marked differences in pseudotyped virion infections when the O-glycan mutant G proteins were expressed. The NiV G mutant virions all displayed pseudotyped viral entry of Vero cells similar to that observed for wt NiV G (Fig 5C), yet most HeV G mutant pseudotyped virions showed reduced viral entry (Fig 4C). Additionally, pseudotyped virions with mutant HeV G had reduced HeV F incorporation and altered F processing, evidenced by little to no F0 present in the viral lysates (Fig 4A and 4B). It is remarkable that loss of O-glycans from HeV G affects F incorporation and processing, and therefore viral entry. Since F may be a driver of viral budding [62], this finding has implications not only for fusion triggering but also for viral assembly and budding, viral particle release, and therefore viral infectivity. O-glycans have been shown to play roles in protein-protein interactions for non-viral proteins [4]. Therefore, it is possible that loss of G O-glycans affects intracellular G/F interactions, thus affecting viral incorporation of both glycoproteins into pseudotyped virions. This is particularly interesting because HeV G and F have been shown to traffic to the cell surface at different rates [63]. Further investigation is needed to determine exactly how mutations in HeV G have such a pronounced effect on HeV F incorporation, particularly in pseudotyped virions, and why this same phenomenon was not observed for NiV G.
Unlike their HeV counterparts, NiV mutant pseudotyped virions incorporated NiV F and also entered cells at wt levels (Fig 5A and 5B), yet F processing was still altered. The hyper-fusogenic mutants, especially T117A and S116A-T117A, showed more efficient F processing (Fig 5A and 5B). Since F1 is the mature form of F, more efficient F processing may induce hyper-fusogenicity for these mutants. Since F must be re-internalized for processing and cleavage, loss of specific O-glycans could affect endocytosis of F [54,55]. The hyperfusogenic mutants, for example, showed less interaction with F0 via co-IP (Fig 2), which may enable F endocytosis for cleavage into F1 and F2 more easily than the wt G or the hypo-fusogenic mutants with stronger F interactions, rationalizing the increase in F processing in the hyperfusogenic mutant virions. Our results demonstrating that HNV G mutants affect F processing are unprecedented and unexpected, warranting further mechanistic studies.
Furthermore, premature F triggering on viral surfaces may have also occurred with the HNV pseudotyped virions. Even though the NiV hyper-fusogenic pseudotyped virions entered cells at wt levels (Fig 5C), we would expect increased entry for the hyper-fusogenic mutants compared to wt G. Premature F triggering has been observed and offered as an explanation for differences between cell-cell fusion and pseudotyped viral entry levels for a headless NiV G mutant [49]. As virions are incubated at 37°C in cell supernatants prior to collection, F and G have the opportunity to interact prematurely, triggering F before a target membrane is present. The prematurely triggered F is then unable to trigger fusion once the virions attach to target cells. Since our co-IP results show that hyper-fusogenic mutants already have a lowered association with F, potentially releasing F more readily for fusion, this may explain why the hyper-fusogenic mutant virions would be most affected by premature F triggering. Supporting this notion, pseudotyped virions produced at 32°C instead of 37°C to hinder premature F triggering regained F incorporation and viral entry levels similar to those of the wt pseudotyped virions (S3 Fig), though we acknowledge that there are other potential cellular factors that may have been altered at the lower temperature.
It should be noted, however, that here we have used a pseudotyped viral system instead of actual henipaviruses. While this pseudotype system has been used extensively in previous studies [14–16,49,64], there are limitations to this approach. Though the pseudotyped virions are expressing the HNV glycoproteins on the viral membrane, they still contain the VSV genome and other VSV proteins. The presence of the VSV matrix protein in place of the HNV matrix protein, a protein known to affect viral processes such as budding [65], may alter virion characteristics. It is also possible that interactions occur between the VSV matrix protein and HNV F/G, since it has been suggested that the HNV matrix protein may interact with each of these glycoproteins [62,66]. The viral morphology of the pseudotyped virions is that of the bullet-shaped VSV, instead of the pleomorphic shaped paramyxoviruses, a factor that may also affect viral particle formation and entry. Thus it is possible that the phenotypes observed in the pseudotyped NiV/VSV and HeV/VSV systems would not translate 100% to actual HNV virions. Due to the BSL4 and reverse genetics requirements to assess these differences with actual HeV or NiV, however, the pseudotyped viral system is an effective surrogate system that provides valuable insight, and further BSL4/reverse genetics studies are warranted.
Both N and O-glycans have been shown to shield viruses against antibody neutralization. O-glycans play a role in shielding the gammaherpesvirus BoHV-4 [22], while loss of specific HeV and NiV N-glycans increased virus susceptibility to antibody neutralization [14–16]. Based on these findings, we hypothesized that loss of O-glycans from HNV G would enhance antibody neutralization sensitivity. Antibody neutralization studies, however, showed no antibody neutralization shielding effects (S4 Fig), highlighting a difference in function between N and O-glycans at least for HNV. It is possible that this functional difference is due to the difference in glycan size, as O-glycans are typically smaller than N-glycans [67]. It is also possible that loss of all O-glycans simultaneously, as performed for BoHV-4 [22] would result in increased antibody neutralization susceptibility, but that loss of a single (or two) O-glycan alone is insufficient to have this effect. Alternatively, this part of the G stalk may not be critical for HNV antibody neutralization.
Since it is known that O-glycosylation at one site can affect the glycosylation of neighboring sites [21], mass spectrometry would be needed to determine if loss of an O-glycan at one site affects addition of O-glycans at other sites for our set of O-glycan mutants. Although these studies require high levels of purified recombinant viral glycoproteins, glycobiology expertise, and time of analysis (likely a 1–2 year project), as evidenced by the scarcity of O-glycan studies currently published for viral glycoproteins, this would be a worthwhile future study. The functions of O-glycans observed in the present study, however, are accurate regardless of whether or not neighboring O-glycan sites are affected by loss of O-glycans at our specific mutated sites.
In summary, we have identified multiple novel functions of paramyxovirus O-glycans in HNV, including strong modulation of cell-cell fusion, and effects on G/F interactions, G conformation, receptor-induced G conformational changes, and quite remarkably, F processing and F incorporation into pseudotyped virions. These novel O-glycan functions shed light on the paramyxovirus-induced membrane fusion mechanisms, reveal a new function for G in F interactions, incorporation into virions, and processing, and contribute to our limited knowledge of the functions of viral O-glycans. It remains to be determined if the high densities of O-glycans and their multiple functions on the stalks of the attachment glycoproteins are conserved among all paramyxovirus genera. Moreover, differences in host cell glycosylation machinery may affect the extent of O-glycosylation or O-glycan structures, so that virions produced in different types of host cells may be differentially glycosylated. The current work paves the way for future functional studies of O-glycans for other paramyxoviruses and other viral families.
Codon optimized HNV G and HNV F genes tagged with HA or AU1 tags, respectively, were expressed in PCDNA3.1 or PCAGGS plasmids as previously described [18]. Alanine substitution mutants were created by site-directed mutagenesis of HNV G using a QuikChange kit (Stratagene). Mutations were confirmed by sequencing the entire open reading frame.
293T (ATCC) and PK13 (ATCC) cells were cultured in Dulbecco’s modified Eagle’s medium with 10% fetal bovine serum (FBS). Vero (ATCC) cells were cultured in minimal essential medium alpha with 10% FBS.
Production of rabbit anti-NiV G monoclonal antibodies has been previously described [56]. Binding of rabbit monoclonal antibodies, mouse anti-HA, or soluble B2-hFc to HNV G wild type (wt) or mutants was measured by flow cytometry. 293T cells were transfected with 2 μg wt or mutant HNV G expression plasmids and collected 20–24 hours post-transfection. Collected cells were incubated 1 hour at 4°C with 1° antibodies diluted 1:100 to 1:1000 and washed three times in FACS buffer (1% FBS with PBS). Cells were next incubated with fluorescent anti-human, anti-mouse, or anti-rabbit Alexa Fluor 647 or 488 antibodies (Life Technologies, NY) diluted 1:200 for 30 min at 4°C, followed by an additional two washes. Cells were fixed in 0.5% PFA and read on a flow cytometer (Guava easyCyte8 HT, EMD Millipore, MA). B2-hFc was used at a 100 μM concentration. For conformational change assays PK13 cells were transfected with wt or mutant HNV G expression plasmids. 0 or 100 μM soluble ephrinB2 was added and cells were incubated for 15 min at 4°C. Mab45 or 26 was added at concentrations specified above and cells were incubated 1 hour at 37°C. 2° antibody, fixation, and cytometer reading were as listed above [49,56].
293T cells grown in 6 well plates were transfected at 70–90% confluency with HNV F and either HNV G or G mutant expression plasmids (1:1 ratio, 2 μg total DNA) using BioT or turbofectamine transfection reagent. 18 hours post-transfection cells were fixed in 0.5% paraformaldehyde and syncytia counts were performed under the microscope (200x), with a syncytium being defined as four or more nuclei within a single cell. Five fields per well were counted [14,49].
Transfected cells or pseudotyped virions expressing HNV G wt or mutants with or without HNV F were lysed in RIPA buffer (Millipore) supplemented with complete protease inhibitor (cOmplete Mini, Roche). Lysates were heated at 65°C for 10 minutes and run on a 10% gel for SDS-PAGE analysis. Immunoblots were performed using mouse anti-AU1 or rabbit anti-HA at 1:250 to 1:2000 dilutions. Fluorescent 2° antibodies were diluted 1:1000–1:2000 and blots were imaged on a Li-Cor Odyssey fluorimager.
Cells transfected with HNV wt or mutant G expression plasmids were lysed in lysis buffer (Miltenyi Biotec). Cell lysates were incubated with 40ul of anti-HA microbeads at 4°C for 30 minutes with rotation. Lysates were purified and eluted over μ columns (Miltenyi Biotec). PAGE was used for cell lysates and column elutions using a 10% gel. F and G proteins were detected by Western blot analysis, as detailed above.
Pseudotyped virions containing HNV F with wt or mutant HNV G were manufactured as previously described [14,39]. Briefly, 15cm plates of 293T cells were transfected at 37°C with HNV F and wt or mutant HNV G expression plasmids at a 1:1 ratio. After 8 hours the transfection media was switched to growth media. After an additional 16 hours the cells were infected with recombinant VSV-ΔG-rLuc. 2 hours later the infection media was removed and replaced with growth media. 24 hours after infection virions were harvested from cell supernatants using ultracentrifugation, resuspended in NTE with 5% sucrose, and stored at -80°C. Viral RNA was extracted using a QIAamp viral RNA mini kit (Qiagen, CA) and the resulting viral RNA was reverse transcribed using a SuperScriptIII First-Strand Synthesis System for RT-PCR (Invitrogen, NY). Quantitative PCR (qPCR) was performed using a Taqman probe for the VSV genome to quantify viral copy number. For virions produced at 32°C, all transfections and viral infections were carried out at 37°C, and plates were switched to 32°C upon addition of new growth media.
Vero cells were infected with tenfold dilutions of pseudotyped virus particles in infection buffer (PBS + 1% FBS) and incubated for 2 hours at 37°C. After 2 hours growth media was added. 18–24 hours after infection cells were lysed and an Infinite M100 microplate reader (Tecan Ltd) was used to measure luciferase activity.
Pseudotyped virions were incubated for 20 min in infection buffer (PBS + 1% FBS) in the presence of varying dilutions (10-2 to 10-7) of anti HNV G polyclonal antibody. The virus/antibody mixture was added to Vero cells and incubated for 2 hours, after which growth media was added and the cells were incubated an additional 18–24 hours. Viral infection was then measured as described above.
|
10.1371/journal.pntd.0006400 | Scabies and impetigo in Timor-Leste: A school screening study in two districts | Scabies and impetigo are common and important skin conditions which are often neglected in developing countries. Limited data have been published on the prevalence of scabies and impetigo in Timor-Leste. Sequelae including cellulitis, bacteraemia, nephritis, acute rheumatic fever and rheumatic heart disease contribute significantly to the burden of disease.
School students were recruited from schools in Dili (urban) and Ermera (rural) in Timor-Leste for an epidemiological study in October 2016. A standard questionnaire was used to record demographics, anthropometry and skin examination results. Impetigo and scabies were diagnosed based on clinical examination of exposed surfaces, and clinical photographs were reviewed for correlation by an infectious diseases paediatrician. Prevalence of scabies and impetigo were calculated and binary risk factor associations were described using relative risks and 95% confidence intervals. Adjusted odds ratios were calculated using logistic regression multivariate analysis. Continuous variables were analysed for associations using the Mann-Whitney Rank Sum test.
The study enrolled 1396 students; median age 11 years (interquartile range (IQR) 9–15). The prevalence of scabies was 22.4% (95% CI 20.2–24.7%) and active impetigo 9.7% (95% CI 8.3–11.4%); 68.2% of students had evidence of either active or healed impetigo. Students in Ermera were more likely than those in Dili to have scabies (prevalence 32.0% vs 5.2%, aOR 8.1 (95% CI 5.2–12.4), p<0.01). There was no difference in the prevalence of active impetigo between urban and rural sites. More than a third of participants were moderately or severely underweight. Stunting was markedly more common in the rural district of Ermera.
Scabies and impetigo are common in Timor-Leste, with very high prevalence of scabies in the rural district of Ermera. Improvements in prevention and treatment are needed, with prioritised activities in the rural areas where prevalence is highest.
| Scabies and impetigo are common and important skin conditions which are often neglected in developing countries. Scabies affects more than 200 million people globally. There are limited data on the prevalence of these conditions in Timor-Leste. Scabies is a parasitic infection which causes an intensely itchy rash in a characteristic distribution. Secondary bacterial infection with group A Streptococcus (GAS) and/or Staphylococcus aureus causing impetigo is common and may lead to complications. GAS skin infection is known to be associated with the development of acute post streptococcal glomerulonephritis, and may also have a role in the aetiology of acute rheumatic fever (ARF) and rheumatic heart disease (RHD. We assessed almost 1400 school students in Timor-Leste from both urban and rural settings and showed a high prevalence of scabies (22.4%) and active impetigo (9.7%). Two-thirds of students had evidence of either active or healed impetigo. Our study shows that scabies and impetigo are common in Timor-Leste, with very high prevalence of scabies in the rural district of Ermera. Improvements in prevention and treatment are needed, with prioritised activities in the rural areas where prevalence is highest.
| Scabies and impetigo are common and important skin conditions which are often neglected in developing countries[1,2]. The global prevalence of scabies was estimated to be over 204 million (in thousands: 204 152 [177 534–237 466])[3], accounting for 0.21% of disability-adjusted life-years from all conditions in the global burden of disease study conducted in 2015[4]. Scabies has recently been adopted as a Category A World Health Organisation (WHO) Neglected Tropical Disease (NTD)[5,6] highlighting the priority of this condition in developing countries. Scabies is a skin infestation caused by the parasite Sarcoptes scabiei. Scabies presents with an intensely pruritic rash with a characteristic distribution pattern[7]. Clinical manifestations include papules, burrows and pruritus [1]. Distribution varies with age and often includes involvement of the webs of the fingers, flexor aspect of the wrist, feet and torso. Transmission is predominantly by direct contact with infected skin, but fomites including bedding and clothing can also play a role [1,8]. Secondary infection by group A Streptococcus (GAS) and/or Staphylococcus aureus causing impetigo is common and may lead to complications including cellulitis, abscess, septic arthritis, osteomyelitis and septicaemia [6]. Impetigo may also occur in the absence of scabies infection due to minor trauma, insect bites and dry skin. GAS skin infection is known to be associated with the development of acute post streptococcal glomerulonephritis, and a role in the aetiology of acute rheumatic fever (ARF) and rheumatic heart disease (RHD) has also been hypothesised [1].
Scabies is endemic in tropical regions globally, with prevalence rates between 5–10% commonly reported in children [9,10]. Timor-Leste is in Southeast Asia and lies northwest of Australia at the eastern end of the Indonesian archipelago. The island is semi-arid with a mountainous terrain and tropical climate. The population of Timor-Leste at the 2015 census was 1.2 million of which 39% were less than 15 years of age. The average household size for Timor-Leste is 5.7 people but is higher in the municipalities of Dili, Alieu, Ermera and Ainaro [11]. Epidemiological data regarding skin infections in Timor-Leste are limited, however previous population screening in Timor-Leste identified scabies in more than a third of children aged under 10 years, as well as high rates of pyoderma coinfection [12].
The aim of this study was to determine the prevalence of scabies and impetigo in school students in urban and rural settings in Timor-Leste and to investigate epidemiological associations of skin disease affecting this cohort. Improved epidemiological evidence is needed in order to design and implement appropriate treatment and prevention strategies at a community level [13].
Ethics approval for the study was obtained from the Human Research Ethics Committee of the Northern Territory (NT) Department of Health and Menzies School of Health Research (2016–2546) and the Institute Nacional de Saude in Timor-Leste (MS-INS/DF/DP/V/2016/220). Permission to undertake screening was granted by the Ministry of Education in Timor-Leste and by the principals of the schools involved. A plain language information sheet in Tetum (the lingua franca in Timor-Leste) was distributed to parents and families prior to commencement of screening, and they were given the opportunity to opt-out of having their child(ren) screened. Use of an opt-out approach to consent occurred because of the strong preference of school principals, clinical staff and other community leaders, identified during informal discussions conducted during consultation visits four months prior to commencement of screening. This is a novel approach to consent in Timor-Leste but has been utilised in similar research in other low and middle-income countries [14,15]. Based on anticipated difficulties with obtaining written consent due to low literacy levels, as well as the low or negligible risk associated with the screening process itself, this approach was approved by both Australian and Timorese human research ethics committees.
Students from schools in the municipalities of Dili (urban) and Ermera (rural) were enrolled in a school screening project that included echocardiography screening for RHD and limited skin examination for scabies and impetigo. The primary outcomes in the skin arm of the study, were presence or absence of scabies, and/or active impetigo, based on limited clinical examination. Baseline demographics including date of birth, age, sex, name, school, address, number of people and number of rooms in the home were collected for all participants. Weight, height and previous known allergy to penicillin were also recorded. Students who attended school on the days of screening were eligible to participate in the study. Children aged less than 5 years and people aged 25 years or older were excluded. Given multiple disruptions to schooling during recent years in Timor-Leste, it is common for secondary students to be aged up to 25 years.
Skin examination was conducted in a school classroom by six medical practitioners who had undergone specific training and validation (using standardised clinical photographs) in the diagnosis of scabies, impetigo and other childhood skin diseases prior to the study. Training was conducted in the week prior to screening. It incorporated teaching about the pathophysiology of childhood skin diseases, descriptions of typical clinical findings, and revision of clinical photographs of varying presentations. Validation involved assessment of 25 clinical photographs of childhood skin diseases, for which candidates were asked to describe their findings and make a diagnosis. Candidates were required to make the correct diagnosis in a minimum of 80% in order to successfully complete training.
Only exposed regions of skin (upper limbs, lower limbs, scalp, face and neck) were examined for reasons of modesty. Students were also asked if they had skin lesions elsewhere on the body hidden under clothing and this was documented but not examined.
Diagnoses of clinical scabies and impetigo were recorded (Box 1). The number of scabies and impetigo lesions were quantified and recorded as ‘none’, 1–10 (mild), 11–49 (moderate), or ≥50 (severe) and body regions involved were documented. [16] The number of inactive (flat, dry) or healed lesions were not quantified. Given the lack of laboratory resources in Timor-Leste it was not feasible to use skin swabs or scrapings for confirmatory diagnosis of aetiological agents.
Other skin conditions were also recorded including fungal skin infections, eczema and wounds. Study participants identified as having a skin condition requiring treatment (including scabies and impetigo) were counselled regarding the diagnosis and provided with an information sheet and referral letter to take to the local health clinic for treatment as per standard treatment guidelines.
Anthropometric data were analysed based on WHO normal growth parameters. Weight-for-age (WFA) z-scores were determined for participants aged 5–10 years and height-for-age (HFA) z-score for participants aged 5–20 years. Body-mass-index-for-age (BMIFA) z-scores were calculated for all participants.
Non-identifying clinical photographs were taken of skin lesions to enable later review of clinical diagnoses. Clinical photographs were reviewed by a member of the study team (AB) who is an infectious diseases paediatrician with extensive experience in the diagnosis of scabies and impetigo, who was not present for screening and was blinded to the results of the clinical examination findings obtained during the study.
Data were recorded in an electronic database (Microsoft Access (2016), and analysed using STATA 14.2 (College Station, TX). Prevalence rates of scabies and impetigo were calculated and binary risk factors described using relative risks and 95% confidence intervals (CI). Logistic regression was performed with the following variables in our model: sex, site, age, nutritional status and scabies/impetigo to calculate adjusted odds ratios (aOR) based on multivariate analysis. Continuous variables not normally distributed were analysed for associations using the Mann-Whitney Rank Sum test. Results were considered significant if p<0.05. Subjects with missing data were excluded from analyses of the missing variable.
A total of 1396 students were enrolled (Table 1). Two-thirds of participants were from the rural area (64.0%) compared to the urban area (36.0%). There were more females (52.8%) than males (47.2%). The median student age was 11 years (interquartile range (IQR) 9–15). The median household size was 7 people (IQR 6–9).
Most students were below average for weight, height and body mass index (BMI). Participants from Ermera were more likely to be moderately or severely underweight (WFA z-score <–2) than students from Dili, 40% vs 32%, RR = 1.3 (95% CI 1.0–1.6), p = 0.04. They were also more likely to be moderately or severely stunted (HFA z-score <-2) than participants from Dili, 37% vs 10%, RR = 3.8 (95% CI 2.9–5.1), p<0.01.
Scabies was detected in 312/1396 (22.3%) students screened (Table 2). Of those with scabies, 26.4% had fewer than 10 lesions, 47.9% had 10–49 lesions, and 25.7% had 50 or more lesions. There was one case of crusted scabies identified diagnosed based on clinical findings. Dermatoscopy and skin scrapings were not performed.
For the participants documented to have scabies lesions most had more than one body region affected. Scabies lesions were reported on the face/scalp in 14/312 (4.5%) of students, neck 52/312 (16.7%), arms 209/312 (67.0%), hands 271/312 (86.9%), legs 155/312 (49.7%) and feet 145/312 (46.5%), with no significant differences in distribution for the different age groups. Scabies in other regions including torso, buttocks, breasts and groin were self-reported with data recorded for less than half the participants.
Multivariate analysis showed that students in Emera were seven times more likely than those in Dili to have scabies (aOR 7.3 (95% CI 4.6–11.7)). Although more males had scabies, this was not statistically significant (aOR 1.3 (95% CI 1.0–1.8)). The median number of people per house for those with scabies was 8 (range 2–16) versus a median of 7 in the houses without scabies (range 2–13), p<0.01. Students aged 20–24 years old were twice as likely to have scabies than those aged 5–9 years (Table 2). All 20–24 year old students were from Ermera.
Purulent or crusted impetigo was diagnosed in 136/1396 (9.7%) of students screened (Table 3). Of those with impetigo, the vast majority (91.3%) had fewer than 10 lesions.
The presence of scabies increased the risk of impetigo infection on univariate and multivariate analyses (RR 2.5 (95% CI 2.1–3.1), p<0.01, aOR 4.4 (95% CI 2.9–6.8), p<0.01). There were no significant differences in risk of scabies for students of different ages or screened at different sites.
There was no significant difference in impetigo severity or distribution across the age categories (Fig 1). The distribution of impetigo involving face/scalp and neck was 15/136 (11.0%), arms 29/136 (21.3%), hands 38/136 (28.0%), legs 86/136 (63.2%) and feet 42/136 (30.9%). Most students (68.2%) had evidence of recent impetigo with either active or healed lesions.
Non-identifying clinical photographs of skin lesions were taken from 266 participants (19.1%). Some participants had multiple photos of different types of lesions. Photos of poor quality, without identifying labels, or multiple pictures from the same participant were excluded, with 208 photos available to be reviewed by a paediatric infectious disease specialist with expertise in skin health (AB), who was blinded to the initial screening diagnosis. The sample was representative of the cohort. Concordance of diagnosis was 165/208 (79%) for presence or absence of scabies (Kappa coefficient 0.57 (95% CI 0.46-.067)) and 167/208 (80%) for presence or absence of any impetigo (Kappa 0.55 (95% CI 0.44–0.66)). Agreement was moderate with discordance for impetigo mostly from cases labelled impetigo by the reviewer but not diagnosed clinically, suggesting that clinical examiners may have underdiagnosed impetigo. Disconcordance for scabies was from cases diagnosed clinically but not confirmed by photography, which may relate to difficulties photographing some of the more subtle features of scabies infestation, including burrows.
Other skin lesions were incidentally identified and reported for 168 students. These included warts 59/1396 (4.2%), fungal skin infections 26/1396 (1.9%), injury including lacerations, abrasions, burns and dog bites 24/1396 (1.7%), dermatitis/eczema 19/1396 (1.4%), and cellulitis or abscess 9/1396 (0.6%).
Of the participants identified as having scabies or active impetigo, 95% were referred for treatment at a local clinic. Limited resources were available to treat students with identified severe infection during the screening process. Two students received intramuscular benzathine penicillin G (BPG) for severe impetigo and oral ivermectin was given to 7 students with severe or crusted scabies.
Our estimates of the prevalence of scabies (22.4%, 95% CI 20.2 to 24.7%) and active impetigo (9.7%, 95% CI 8.3 to 11.4%) in Timor-Leste are higher than those reported for school aged children in an earlier study, which found scabies in 159/1114 (14.3%) and pyoderma in 74/1114 (6.6%) [12], and higher than regional norms excluding the Northern Territory of Australia [1,2,9,17,18]. Recent studies in Fiji have also reported very high population prevalence of scabies (36.4%) and impetigo (23.4%).[19]. Recognising this large burden of disease is a necessary first step in developing and implementing strategies to address it.
Scabies is a disease of poverty and associated with household size, low socioeconomic groups and poor access to healthcare have been well documented [1]. Rural students in Ermera were more likely than those in Dili to have scabies. The median number of people per house was 7 in this study. It is likely that scabies infection and household size contribute to ongoing GAS transmission, as suggested by the presence of healed impetigo in the majority of students. Anthropometric data identified that many participants were underweight and/or stunted, particularly in Ermera. These findings are suggestive of poor nutrition and may be indicative of poor socioeconomic conditions.
Scabies and impetigo contribute to a significant burden of disease for children and adults in low resource countries [4,9]. Scabies infection can predispose to secondary bacterial infection with GAS and S. aureus. GAS is also implicated in the aetiology of glomerulonephritis, ARF and RHD, which can develop in the context of mild infections or carriage. Echocardiography screening, conducted in parallel with the skin screening performed for this study, demonstrated a very high prevalence of definite and borderline RHD (3.5%) amongst school students in these two districts of Timor-Leste [20].
In Timor-Leste, there are limited options for treatment of scabies and impetigo, and consideration should be given to community treatment strategies. First line treatment for scabies is typically with topical agents. Permethrin is effective [21] and well tolerated, however, it is expensive and not widely available in Timor-Leste. In low resource settings sulfur containing preparations and benzyl benzoate 10–25% are often used [1], but regular supply of these is also unreliable. Ivermectin is an oral alternative that has previously been used for refractory cases or crusted scabies and has been trialled successfully for mass drug administration in settings with high prevalence of scabies [13]. Ivermectin has not been used routinely for treatment of scabies in Timor-Leste and is rarely available. Given the high prevalence of scabies in this sample, consideration for mass drug administration is urgently needed. However, the cost, logistics and needed infrastructure make this challenging. Scabies treatment is further complicated by the need to treat household contacts and implement eradication techniques for contaminated clothing and bedding[22].
Oral antibiotics for bacterial skin infection are available at local clinics in Timor-Leste although supply is unreliable. Evidence shows that short course oral trimethoprim/sulfamethoxazole is an effective treatment for impetigo in endemic settings [23], and offers some advantages over BPG injections in terms of tolerability [23,24]. Both antibiotics are cheap, intermittently available and could be used interchangeably in this context based on the high-quality evidence. The impact of this approach on subsequent post streptococcal sequelae such as glomerulonephritis and ARF is unknown.
High rates of skin infection in school students are suggestive of high rates community wide, including vulnerable groups such as infants, preschool children and the elderly. Whilst improving access to clinical treatment is important, consideration should also be given to implementing programs targeted at limiting or eradicating endemic scabies and impetigo. Possible strategies could include community education, water, sanitation and hygiene programs, and community wide treatment with mass drug administration [13].
Strengths of our study include low levels of non-participation. This was likely due to the opt-out consent approach; we were not aware of any families choosing to opt out. Thus, we believe that we have described a representative population of school-attenders in Dili and Ermera districts. The study is obviously limited by not including school aged children and young people who do not attend school which may lead to an underestimate of the prevalence of scabies and impetigo. By limiting the study to these two districts, it is not possible to confidently estimate the prevalence of scabies and impetigo in the rest of the country, but it is likely that rates in other districts are also high. Significantly, more rural than urban students were enrolled in the study. Urban students were enrolled from a single site thus we may not have captured students from all socioeconomic groups in Dili, some of which may have had higher or lower rates of skin infections. Another significant limitation of our study was the limited skin examination as due to the constraints of the facilities and need for privacy only visible skin was examined. This may have resulted in a measurement bias due to the potential under-diagnosis of scabies and impetigo, and the number of lesions on individual students. Clinicians performing skin examination were trained in the identification of scabies and impetigo, but did not receive specific training in the identification of other skin lesions, so these data may not reliably represent the prevalence of other skin diseases in the population. There was no dermatologist on the study team but paediatricians with expertise in the diagnosis and management of skin infections in children were involved.
Scabies and impetigo are highly prevalent in school students in Timor-Leste particularly in the rural district of Ermera. The burden is greater now than it was recognised to be previously. A coordinated approach to improving prevention and treatment are needed, and consideration should be given for implementing strategies at a community level, focusing on rural areas.
|
10.1371/journal.pgen.1005864 | Ingression Progression Complexes Control Extracellular Matrix Remodelling during Cytokinesis in Budding Yeast | Eukaryotic cells must coordinate contraction of the actomyosin ring at the division site together with ingression of the plasma membrane and remodelling of the extracellular matrix (ECM) to support cytokinesis, but the underlying mechanisms are still poorly understood. In eukaryotes, glycosyltransferases that synthesise ECM polysaccharides are emerging as key factors during cytokinesis. The budding yeast chitin synthase Chs2 makes the primary septum, a special layer of the ECM, which is an essential process during cell division. Here we isolated a group of actomyosin ring components that form complexes together with Chs2 at the cleavage site at the end of the cell cycle, which we named ‘ingression progression complexes’ (IPCs). In addition to type II myosin, the IQGAP protein Iqg1 and Chs2, IPCs contain the F-BAR protein Hof1, and the cytokinesis regulators Inn1 and Cyk3. We describe the molecular mechanism by which chitin synthase is activated by direct association of the C2 domain of Inn1, and the transglutaminase-like domain of Cyk3, with the catalytic domain of Chs2. We used an experimental system to find a previously unanticipated role for the C-terminus of Inn1 in preventing the untimely activation of Chs2 at the cleavage site until Cyk3 releases the block on Chs2 activity during late mitosis. These findings support a model for the co-ordinated regulation of cell division in budding yeast, in which IPCs play a central role.
| Cytokinesis is the process by which a cell divides in two and occurs once cells have replicated and segregated their chromosomes. Eukaryotic cells assemble a molecular machine called the actomyosin ring that drives cytokinesis. Contraction of the actomyosin ring is coupled to ingression of the plasma membrane and extracellular matrix remodelling. In eukaryotes, glycosyltransferases that synthesise polysaccharides of the extracellular matrix are emerging as essential factors during cytokinesis. Defects associated with the function of those glycosyltransferases induce the failure of cell division, which promotes the formation of genetically unstable tetraploid cells. Budding yeast cells contain a glycosyltransferase called Chs2 that makes a special layer of extracellular matrix and is essential during cell division. Our findings provide new insights into the molecular mechanism by which the cytokinesis regulators Inn1 and Cyk3 finely regulate the activity of glycosyltransferase Chs2 at the end of mitosis. In addition we isolated a group of actomyosin ring components that form complexes together with Chs2 and Inn1 at the cleavage site, which we have named ‘ingression progression complexes’. These complexes coordinate the contraction of the actomyosin ring, ingression of the plasma membrane and extracellular matrix remodelling in a precise manner. Chs2 is indeed a key factor for coordinating these events. It appears that similar principles could apply to other eukaryotic species, such as fission yeast even if the identity of the relevant glycosyltransferase has changed over the evolution. Taking into account the conservation of the basic cytokinetic mechanisms future studies should try to determine whether a glycosyltransferase similar to Chs2 plays a key role during cytokinesis in human cells.
| Eukaryotic cells divide their cytoplasm at the end of mitosis in a highly regulated process called cytokinesis, which safeguards inheritance of the genome and organelles by the two daughter cells. The failure of cell division results in the formation of genetically unstable tetraploid cells, which may give rise to cancer [1] [2]. The successful completion of cytokinesis requires the precise coordination between an actomyosin-based contractile ring, which drives the ingression of the plasma membrane, and the remodelling of the extracellular matrix (ECM) [3] [4] [5] [6]. Yeast cells are surrounded by rigid ECM known as the cell wall, which provides the structural support and protection necessary to survive as unicellular organisms. The ECM is composed of a collection of biochemically distinct components, among which polysaccharides are emerging as key factors during cytokinesis, as shown by the failure in cytokinesis caused by defects associated with their synthesis in evolutionary distant organisms such as the budding yeast Saccharomyces cerevisiae [7] [8], the fission yeast Schizosaccharomyces pombe [9] [10] [11], the nematode Caenorhabditis elegans [12] and the mouse [13]. In these four examples, the impairment of a glycosyltransferase determines clear cell division defects. In budding yeast, it is the glycosyltransferase chitin synthase II, a transmembrane protein encoded by CHS2, which centripetally produces a distinct layer of chitin between mother and daughter cells during cytokinesis, called the primary septum, which is essential for life [6]. Chitin is a polymer of N-acetylglucosamine (Glc-NAc), which is synthesised from an activated nucleotide substrate UDP-N-acetylglucosamine (UDP-GlcNAc), and chitin chains are subsequently secreted outside the cells, assembled into microfibrils and organised in the extracellular matrix [14] [15].
In yeast cells, primary septum formation is tightly coupled to actomyosin ring contraction and ingression of the plasma membrane at the cleavage site [6]. In fact, defects associated with one of those processes perturb the others, although the underlying mechanisms linking them together remain unclear [16] [17] [18] [8] [19] [20]. The primary septum, which is later flanked by secondary septa, is finally digested to allow separation of the two daughter cells [21].
The core components and mechanisms of cytokinesis are largely conserved from yeast to humans, which makes the budding yeast cells an attractive model for studying the process of eukaryotic cytokinesis and for identifying how cells coordinate such processes [22] [5] [6]. Successful cytokinesis requires mechanisms that timely and effectively orchestrate the completion of the different steps along the cell cycle. First, cells need to assemble a contractile ring containing type II myosin and many other factors at the cleavage site, in a sequential and highly regulated process. At the early stages of the cell cycle, the type II myosin Myo1 forms a ring at the place that will later become the division site [16] [23]. Myo1 plays a scaffolding role in the assembly of the cytokinetic machinery [24] and associates with other factors during mitosis. These include actin-nucleating and bundling factors such as formins and the IQGAP protein Iqg1, leading to the assembly of a functional contractile actomyosin ring at the end of anaphase [25] [26] [22] [5]. Iqg1 contains an amino terminal calponin homology domain, which is thought to crosslink actin filaments, followed by IQ repeats that interact with Hof1 [27]. Interestingly, Hof1 interacts directly with type II myosin Myo1 and localises at the cleavage site in a complex manner, which depends upon Myo1 [28] [29]. In addition, it has recently been described that Hof1 shares a role with Rvs167 in actin ring assembly and Iqg1 recruitment to the bud-neck [30]. Hof1 contains an F-BAR domain in its N-terminal region and an SH3 domain in its C-terminus, both of which have been shown to be important for dynamics and function of the Hof1 protein [28] [29]. The SH3 domain of Hof1 is known to interact with proline-rich motifs (PXXP) located at the C-terminus of Inn1 [19] [31] [30]. Cells depleted for Inn1 still allow contraction of the actomyosin ring, but membrane ingression fails and the primary septum is not formed, despite the presence of Chs2 [19] [31]. In addition to Hof1, Inn1 interacts with Iqg1 [19] and with Cyk3, through the SH3 domain of Cyk3 located at its N-terminus [32] [31] [33] [30]. Furthermore, Hof1 SH3 binds to a proline-rich stretch of Cyk3 [34]. Taken together, it seems that multiple actomyosin components share binary interactions, but until now there has been no evidence that they all interact together to form large complexes in cells, in order to perform coordinated functions during cytokinesis.
Following full assembly of the contractile ring, primary septum formation occurs when cells have segregated their chromosomes and actomyosin ring contraction initiates. The expression, localisation and enzymatic activity of chitin synthase Chs2 are temporally and spatially regulated [35] [36] [8] [37] [38] [39]. Recent findings suggest that Hof1, Inn1 and Cyk3 regulate chitin synthase during cytokinesis in budding yeast, although the molecular mechanism is poorly understood [19] [31] [32] [40] [20]. Hof1 interacts directly with Chs2 and stabilises the chitin synthase at the cleavage site [29]. It also appears that Cyk3 could regulate Chs2 activity, since an increased dosage of Cyk3 stimulates Chs2-dependent chitin synthesis and the formation of primary-septum-like structures at the bud neck [41] [42]. Moreover, we found genetic evidence that enhanced chitin synthase activity associated with a hypermorphic allele of CHS2, CHS2-V377I, suppresses the defects associated with an inactive form of the C2 domain of Inn1 (first 134 amino acids of Inn1) and deficiencies associated with the lack of Cyk3 in budding yeast cells [20].
Here, we have isolated complexes containing the actomyosin ring components Myo1, Iqg1, Hof1, Inn1 and Cyk3 all together with chitin synthase Chs2 from cells undergoing cytokinesis, which we named ‘ingression progression complexes’ or IPCs. We show that IPCs are assembled at the end of the cell cycle and we propose that IPCs coordinate contraction of the actomyosin ring, plasma membrane ingression and primary septum deposition in budding yeast. We find that IPC components co-operate to recruit Chs2 to the division site. Moreover, we provide evidence that Inn1 and Cyk3 interact directly with the catalytic domain of Chs2. Our data indicate that the C2 domain of Inn1 and the transglutaminase-like domain of Cyk3 increase the chitin synthase activity associated with Chs2. We used an experimental system to find a previously unanticipated role for the C-terminus of Inn1 in preventing the untimely activation of Chs2 at the cleavage site until Cyk3 releases the block on Chs2 activity, when cells reach the end of the cell cycle
We previously found that Inn1 co-purified with Chs2 when studying cells that had been released into mitosis from a G2-M block to allow them to undergo cytokinesis synchronously [20]. To study further the interaction between Inn1 and Chs2, we used several approaches. First, we used the yeast two-hybrid assay to show that a fragment of Chs2 that contains its catalytic domain (Chs2-215-629) was able to interact with full-length Inn1 (Fig 1A). We then determined whether these factors were able to interact directly in an extract of E. coli cells. We generated an E. coli strain that expressed 6His-tagged Inn1 and, in parallel, another strain that expressed a truncated version of Chs2 fused to Streptag (Streptag-Chs2-215-629), as indicated in S1A Fig. We then mixed the cultures and generated a single cell extract containing Inn1, Chs2 and all the native E. coli proteins (S1A Fig). We initially purified the truncated version of Chs2 from the cell extracts, and subsequently isolated 6His-Inn1 from the purified material. In this way, we found that Chs2 co-purified specifically with Inn1 (Fig 1B). Note that both Inn1 and Chs2-215-629 migrate similarly in SDS-PAGE gels, and so their presence was confirmed by mass spectrometry and immunoblotting analysis (Fig 1Bii and 1Biii). Following the same purification procedure described above, we found that a fragment of Chs2 that contains its CDK-regulated N-terminal domain together with its catalytic domain (Chs2-1-629, which only lacks the transmembrane domain) co-purified specifically with Inn1 (Fig 1C). Furthermore, we determined that formation of Chs2-Inn1 complex was not abolished by the Inn1-K31A mutation, which disrupts the function of the Inn1 C2 domain, or by a hypermorphic mutation in the catalytic domain of Chs2 (Chs2-V377I), which enhances its activity in vitro (Fig 1D).
Since it has been shown that inactivation of the C2 domain of Inn1 can be rescued by specific mutations in the catalytic domain of Chs2 that increase its activity [20], we tested whether the C2 domain of Inn1 directly associates with and regulates chitin synthase Chs2 in vivo. A yeast strain was generated in which either wild-type Inn1 or the Inn1 C2 domain were fused to the tandem affinity purification (TAP) tag and expressed under the control of the INN1 promoter. INN1-TAP, C2-TAP and control strains were grown at 24°C, synchronised in G1 phase of the cell cycle by the addition of mating pheromone and cells were then released from G1 arrest. The resultant fusion proteins were isolated from cells going through cytokinesis synchronously 105 minutes after the release from G1 block, when localisation of Inn1 and Chs2 at the site of division peaks. We found that Chs2 co-purified specifically with the C2 domain of Inn1, equivalent to full-length Inn1 (Fig 2Ai). To test whether the C2 domain of Inn1 could interact directly with Chs2, we generated E.coli strains that produced 6His-tagged-Inn1-C2 and Strep-tag-Chs2-215-629 and proceeded as above. We found that 6His-C2 co-purified over two purification steps with Strep-tag-Chs2-215-629 (Fig 2Aii), indicating that the Inn1 C2 domain interacts directly with a fragment of Chs2 that contains its catalytic domain. In addition, we determined that interaction between Chs2 and the Inn1 C2 domain was not disrupted by Inn1-K31A mutation (Fig 2Aiii).
To investigate whether the Inn1 C2 could induce chitin synthase activity in vivo, we monitored the chitin level at the division site by calcofluor staining [43] in cells that overexpressed the C2 domain and lacked Chs3 (studies focused on Chs2 activity require the use of chs3Δ cells, because Chs3 is responsible for the synthesis of the vast majority of the chitin content in budding yeast cells [44]). Asynchronous cultures were grown at 24°C and cells were synchronised in G1 phase with mating pheromone. Subsequently, we released cells from G1 block into medium containing calcofluor to stain primary septa and galactose to allow overexpression of the C2 domain. Progression through cytokinesis and localisation of Chs2 at the site of division were similar in both control and GAL-C2 cells (S1B Fig). To observe calcofluor-stained chitin in cells completing mitosis, cells were collected 135 minutes after release from G1 block (Fig 2B) when the percentage of cells containing primary septa peaks. We found that the signal intensity associated with calcofluor-stained chitin at the division site was higher in cells overexpressing C2 as compared to control cells (Fig 2B), indicating that the C2 domain of Inn1 is able to induce septum formation.
To test whether the C2 of Inn1 positively regulates the chitin synthase Chs2, two different in vitro approaches were used. First, two chs3Δ yeast strains were generated to overexpress either CHS2 or CHS2 together with the C2 domain. We grew control, GAL-CHS2 and GAL-CHS2 GAL-C2 strains asynchronously in the presence of raffinose and then switched to medium containing galactose to induce the expression of Chs2 and the C2 domain (Fig 2Ci). After two hours we isolated membranes to perform a chitin assay, as previously reported [45] [20]. We found that the Inn1 C2 domain had an effect on Chs2 activity, since overexpression of C2 at the same time as Chs2 caused a 30% increase in the percentage of active chitin synthase (Fig 2Cii, compare CHS2 with CHS2 C2). Consistently, we observed that cells overproducing the C2 domain and Chs2 induced thicker primary septum deposition (Fig 2Ciii and 2Civ).
Second, we fused the Inn1 C2 domain to Chs2 and measured the enzymatic activity associated with the C2-Chs2 fusion protein. We have previously shown that C2-CHS2 fusion fully supports cytokinesis in inn1Δ cells [20]. We grew cells asynchronously, isolated membranes and performed an in vitro chitin assay. Subsequently, we calculated the percentage of active chitin synthase associated with C2-Chs2 and found that it increased 3-fold in comparison with Chs2 activity in control cells (Fig 2D, compare 1–2). This increase was significantly reduced when an inactive version of the C2 (C2-K31A) was fused to Chs2 (Fig 2D, compare 1–3). Taken together, these findings show that the Inn1 C2 domain directly binds to and regulates the catalytic domain of Chs2, which is required to form the primary septum during cytokinesis. Interestingly, the percentage of active chitin synthase associated with Chs2-V377I, increased 4.3 fold when compared with control Chs2 under the same conditions described above (Fig 2D(i), compare 1–4) while fusion proteins expression levels were similar (Fig 2D(ii)). This suggests that there might be other factors, in addition to the Inn1 C2, that contribute to Chs2 activation (Fig 2D, compare the difference between 1–2, 1–4 and 2–4).
To understand how cells control the activity of Chs2 at the division site during cytokinesis, we aimed to isolate Inn1-Chs2 complexes specifically and subsequently identify their protein composition by mass spectrometry. We grew a five-litres culture of INN1-TAP CHS2-9MYC cells, together with INN1 CHS2-9MYC control cells that expressed the TAP tag under the control of the TET promoter. Both cultures were grown at 24°C, synchronised in G1 phase of the cell cycle by the addition of mating pheromone and cells were then released from G1 arrest for 105 minutes to focus on the time when the localisation of Inn1 and Chs2 at the cleavage site peaks. Initially, after making cell extracts, Inn1-TAP (or TAP tag in the control) were pulled down and subsequently Chs2-9MYC was immunoprecipitated from the material generated in the first step. This method facilitated the specific enrichment of Inn1-Chs2 complexes, as well as any proteins interacting with them at this point during cell division [46]. First, we confirmed by immunoblotting the presence of both Inn1 and Chs2 in our final purified material (Fig 3Ai). To identify in an unbiased fashion other factors that might regulate Chs2 activity, both purified samples were run in polyacrylamide gels and the lanes were cut into 10 bands and analysed by mass spectrometry (Fig 3Aii). A specific set of proteins that interact with Inn1-Chs2 complexes and are known core components of the budding yeast actomyosin ring was found: the sole and essential IQGAP protein Iqg1; the F-BAR domain containing protein Hof1; the type II myosin, Myo1 and Cyk3 protein, which contains a transglutaminase-like domain and an SH3 domain (Fig 3Aii). The interactions were subsequently confirmed by immunoblotting (Fig 3Aiii), using antibodies we raised against Cyk3 (S1C Fig) and Inn1 [20]. In addition, to test whether this set of proteins could be isolated immunoprecipitating another component of that newly identified complex, we pulled down protein Iqg1 fused to HA from cells going through cytokinesis synchronously, as explained above. Then, we used antibodies against Chs2 that we raised (S1D Fig), together with antibodies against Inn1 and Cyk3, to confirm that Iqg1 indeed interacted with proteins isolated in our systematic analysis (Fig 3B), in agreement with past observations of binary interactions amongst these factors [25] [19] [31] [32] [33] [34] [29] [30] [27]. These findings suggest that Inn1, Chs2, Iqg1, Hof1, Myo1 and Cyk3 interact during cytokinesis, to form complexes that coordinate actomyosin ring contraction, plasma membrane ingression and primary septum formation. We propose to name these complexes ‘ingression progression complexes’ or IPCs. To determine when during the cell cycle IPC components interact, the type II myosin Myo1 was immunoprecipitated from extracts of cells that had been arrested in G1 phase, S phase or were going through cytokinesis synchronously (Fig 3C). We found that Myo1 only interacted with IPC components at the end of the cell cycle, which is consistent with a key role of IPCs during cytokinesis (Fig 3C).
Amongst the components of the IPCs, the Cyk3 protein is poorly characterised and might play a direct role in the regulation of Chs2 chitin synthase activity associated with Chs2, although the molecular details are unclear. Genetic studies have shown that increased doses of Cyk3 complemented defects associated with cytokinesis mutants myo1, iqg1, inn1 and hof1 [47] [31] [32] [42] but failed to rescue chs2Δ cells (S2 Fig) [42]. In addition, the overexpression of Cyk3 stimulated chitin synthesis at the division site (Fig 4B) [41] [42] and a hypermorphic allele of CHS2 rescued defects produced by the lack of the Cyk3 protein [20]. Thus, we aimed to explore further the role of the Cyk3 subunit of IPCs in the regulation of the Chs2 chitin synthase.
Cyk3 contains two domains, an N-terminal SH3 domain and a transglutaminase-like domain located in the second half of the protein. We initially used the yeast two-hybrid assay to determine whether the Cyk3 SH3 interacted with a truncated version of Chs2 lacking the transmembrane domain (Chs2-1-629; S3 Fig). We found that the Cyk3 SH3 domain did not interact with chitin synthase Chs2, although it did interact with the C-terminus of Inn1 (S3 Fig) [31] [32] [33] [30].
The catalytic core of the active transglutaminase domains has three conserved active residues that form the catalytic triad: cysteine, histidine, and aspartic acid. The catalytic triad of the fungal Cyk3 proteins is unusual since it contains the conserved histidine and aspartic acid, but it lacks the conserved catalytic cysteine (S6A Fig) [48] [49]. To examine the role of the transglutaminase-like domain of Cyk3 in more detail, the conserved histidine and aspartic acid, which are well conserved in orthologues of Cyk3 in other eukaryotic species, were mutated to alanines (H563A and D578A; hereafter called cyk3-2A) (Figs 4 and S6A). To analyse the function of the Cyk3 transglutaminase-like domain in budding yeast we used cells in which the C2 domain of Inn1 was fused to the actomyosin ring component Hof1, since we have previously reported that CYK3 becomes essential in these cells [20]. C2-HOF1 strain grew as rapidly as a wild-type strain and did not display any detectable defects in cell division [19]. The meiotic progeny of C2-HOF1 cyk3-2A diploid cells was then analysed by tetrad analysis. We found that cyk3-2A was synthetically lethal with C2-HOF1, which suggested that the conserved residues H563 and D578 in the tranglutaminase-like domain are important for maintaining the function of Cyk3 (Fig 4A).
Overexpression of Cyk3 stimulated chitin synthesis at the division site (Fig 4B) [41] [42] and seemed to have no effects on cell cycle progression and Chs2 localisation (S4 Fig). To investigate whether the transglutaminase-like domain mutant cyk3-2A conserved the ability to increase the primary septum formation, chs3Δ strains overexpressing CYK3 or cyk3-2A, together with control were grown at 24°C and the cells were synchronised in G1 phase. They were then released from G1 arrest into medium containing calcofluor to stain the primary septa and galactose to allow the overexpression of either Cyk3 or Cyk3-2A. Cells were collected 135 minutes after the release when the percentage of cells containing primary septa peaks. Subsequently, samples were used to examine the presence of primary septum at the division site by fluorescence microscopy. Cells overexpressing Cyk3-2A contained similar levels of primary septum as control cells (Fig 4B) unlike cells overproducing Cyk3, whose primary septa were 3 fold more intense (Fig 4B). Taken together, these observations suggest that the transglutaminase-like domain of Cyk3 is important to stimulate chitin synthesis during cell division in budding yeast.
To explore the possibility that Cyk3 might interact with Chs2 and therefore could be important for its chitin synthase activity, we performed a yeast two-hybrid assay with two different fragments that contained the transglutaminase-like domain of Cyk3 (Cyk3-1-594 and Cyk3-475-885) against the Chs2 truncation mentioned above that includes its catalytic domain (Chs2-215-629) (Fig 5Ai) or the fragment of Chs2 that only lacks the transmembrane domain (Chs2-1-629) (S5 Fig). We determined that both Cyk3 truncations were clearly able to interact with Chs2 (Fig 5Ai; S5 Fig). To study whether the transglutaminase-like domain mutant cyk3-2A conserved the ability to interact with Chs2, we carried out a yeast two-hybrid assay as above (Fig 5Aii; S5 Fig). We found that interaction was not abolished by mutations in the transglutaminase-like domain of Cyk3 (Fig 5Aii; S5 Fig). Thus, we aimed to determine whether the transglutaminase-like domain of Cyk3 interacts with Chs2, so we performed a yeast two-hybrid assay and found that a fragment of Cyk3 that contains precisely the transglutaminase-like domain (Cyk3-475-594) was unable to interact with Chs2-215-629 (Fig 5Aiii) or Chs2-1-629 (S5 Fig). Whereas a slightly bigger fragment of Cyk3 containing the transglutaminase-like domain (Cyk3-475-764), interacted with Chs2, which would indicate that the transglutaminase-like domain is not enough to bind to Chs2 (Fig 5Aiii). In addition, we observed that the two versions of Cyk3 that lack the transglutaminase-like domain (Cyk3-1-475 and Cyk3-765-885) are able to interact with Chs2, which showed that different domains within Cyk3 protein structure are responsible for the interaction between Cyk3 and Chs2 (Fig 5Aiii; S5 Fig). Interestingly we found that interactions are the same whether we performed the yeast two-hybrid assay with either Chs2-215-629 (Fig 5A) or Chs2-1-629 (S5 Fig), except for Cyk3-475-764 fragment. We showed that Cyk3-475-764 interacted with the fragment of Chs2 that lacks the N-terminal domain but not with Chs2-1-629, which would suggest that the N-terminal tail of Chs2 could play a role in regulating the interaction between Chs2 and Cyk3. Our findings would indicate that Cyk3 protein uses different domains to bind to chitin synthase Chs2.
We then examine whether artificial recruitment of the transglutaminase-like domain of Cyk3 to the actomyosin ring was sufficient to supply Cyk3 function. We have previously reported that CYK3 becomes essential in cells in which the C2 domain of Inn1 was fused to the chitin synthase Chs2 (C2-CHS2) in the same way as when the C2 domain is fused to HOF1 [20]. A diploid strain was created in which a copy of CYK3 was inactivated (cyk3-2A) and one copy of HOF1 had been modified so that the encoded protein was fused to the transglutaminase-like domain of Cyk3 (TG-HOF1). The meiotic progeny of the resultant strain was then analysed by tetrad analysis. We found that expression of the TG-Hof1 fusion protein rescued the lethal effects associated to C2-CHS2 cyk3-2A cells (S6B Fig), which shows that the fusion protein is able to bring the function of the transglutaminase-like domain to the site of division. Nevertheless, the transglutaminase-like domain is not the only essential function of Cyk3 in C2-CHS2 cells, since artificial recruitment of the transglutaminase-like domain of Cyk3 to the actomyosin ring was insufficient to provide Cyk3 function in cells lacking the CYK3 gene (S6C Fig).
To determine whether Cyk3 and Chs2 did indeed bind each other directly, we studied whether these factors were able to interact in E.coli extracts. We used E. coli cells to express 6His-Cyk3, in parallel with another strain that expressed Strep-tag-Chs2-215-629. After two consecutive purification steps, as described previously, we were able to observe that both proteins formed a stable complex (Fig 5B). Overall, these data suggest that Cyk3 interacts directly with chitin synthase Chs2 and regulates its enzymatic activity during cytokinesis, in which the transglutaminase-like domain of Cyk3 plays an important role.
The next step was to determine whether a protein fragment of Chs2 that contained the catalytic domain (Chs2-215-629) together with Inn1 and Cyk3 could form a stable complex in the absence of other eukaryotic proteins by using an E. coli expression system. We made parallel cultures of cells that expressed 6His-Inn1, Strep-tag-Chs2-215-629 and untagged Cyk3. We then mixed them and prepared a common cell extract containing the three proteins. After consecutive purification of the Chs2 fragment and Inn1, we were able to show that these factors formed a ternary complex with Cyk3 (Fig 5C). Therefore, our findings so far show that Chs2, Inn1 and Cyk3 proteins all bind directly to each other. These findings also indicate that both the transglutaminase-like domain of Cyk3 and the C2 domain of Inn1 regulate chitin synthase activity at the site of division.
We have previously found that Cyk3 is essential in cells expressing the fusion C2-Hof1 [20] despite the presence of wild-type Inn1 in these cells. We proposed that understanding why Cyk3 becomes essential in C2-HOF1 cells could reveal the molecular details of how Inn1 and Cyk3 regulate primary septum deposition. To determine whether the problem associated with those cells was related to the function of chitin synthase Chs2, we constructed a diploid strain that lacked one copy of CYK3 and harboured the fusion C2-HOF1 and the hypermorphic allele of CHS2 (Fig 6A). We found that hypermorphic Chs2 (CHS2-V377I) suppressed the cytokinesis defect caused by the lack of the Cyk3 protein in C2-HOF1 cells (Fig 6A, compare double and triple mutant), which confirms that C2-HOF1 cyk3Δ cells fail cell division because of the defects associated with primary septum formation, despite the presence of wild-type Inn1 (Fig 6A).
One possible explanation for why Cyk3 becomes essential in C2-HOF1 cells might be that wild-type Inn1 is unable to localise at the cleavage site, such that these cells would have C2 function (via C2-Hof1 fusion) but they would lack Inn1 C-terminus function. This hypothesis would argue that the Inn1 C-terminus and Cyk3 could share a function and cells would cope with the absence of one of them, but not with the lack of both at the same time. To determine whether wild-type Inn1 is able to localise in C2-HOF1, we synchronised INN1-GFP and C2-HOF1 INN1-GFP cells at 24°C in the G1 phase and then released cells to study Inn1-GFP localisation at the cleavage site. We could observe no defect in Inn1 localisation (S7A Fig).
Since C2-HOF1 cells have two C2 domains (C2-Hof1 and full-length Inn1), second option could be that cells containing two active C2 domains might require the presence of Cyk3, presumably to regulate Chs2 function. We generated yeast strains harbouring the C2-HOF1 fusion and the ‘auxin inducible degron’ (‘aid’) cassette on CYK3 to conditionally inactivate Cyk3 protein [50]. We reproduced previously described synthetic lethality using tetrad analysis (Fig 6B(i); see 3 and 4). In addition, to further check this second possibility, we made strains in which cells carried a degron version of Inn1 in order to be able to deplete Inn1. In parallel these cells expressed an extra copy of INN1, which had been mutated to inactivate C2 function (C2-K31A), although the Inn1 C-terminus remained fully functional (C2-HOF1 td-inn1-aid leu2::K31A) (Fig 6B(ii)). After Inn1 inactivation, these cells grew as the C2 function was carried by the fusion C2-HOF1, despite the expression of mutated Inn1 (Inn1-K31A) (Fig 6B(ii); see 5 and 6). To determine whether Cyk3 was essential for cells expressing C2-Hof1 and Inn1-K31A, we inactivated Cyk3 and assayed cell growth after three days to show that cells died (Fig 6B(ii); see 6 and 7). Thus, we noted that the reason why C2-HOF1 cyk3Δ cells cannot grow is not due to the presence of extra C2 activity, but it is down to the lack of Cyk3 when cells express a functional Inn1 C-terminus (Fig 6B; see 4 and 7). Furthermore, tetrad analysis of diploid C2-K31A-HOF1 cyk3Δ CHS2-V377I (containing wild-type levels of Inn1) revealed that C2-K31A-HOF1 cyk3Δ cells are unable to form a colony (Fig 6C). Intriguingly, in those cells, despite the presence of non-functional C2 fused to HOF1, C2 function can be performed by wild-type copy of Inn1 (Fig 6C, see C2-K31A-HOF1 cells). It thus appears that it is the presence of Inn1 C-terminus and the lack of Cyk3 function that are responsible for the death of C2-K31A-HOF1 cyk3Δ cells. This defect can be rescued by the hypermorphic allele of CHS2 (Fig 6C, see C2-K31A-HOF1 cyk3Δ CHS2-V377I cells). To determine whether the lack of Inn1 C-terminus and the absence of Cyk3 can be fully bypassed by hypermorphic CHS2, we generated a diploid carrying the fusion C2-HOF1, deletions of both CYK3 and INN1, together with hypermorphic allele of CHS2 (Fig 6D). We found that the lack of Inn1 C-terminus function in C2-HOF1 cyk3Δ inn1Δ cells (Inn1 C2 function is carried by the fusion C2-HOF1) is completely rescued by increasing the chitin synthase activity associated to Chs2 (Fig 6D; the result was confirmed using a growth assay in S7B Fig). Finally, we confirmed the same observations inactivating specifically the transglutaminase-like domain of Cyk3 (cyk3-2A) in C2-HOF1 inn1Δ CHS2-V377I cells (Fig 6E). Our data would indicate that Inn1 C-terminus and Cyk3 are involved in the regulation of chitin synthase activity associated to Chs2.
Therefore, the third possibility is that, despite the presence of C2 function (via C2-Hof1), the Inn1 C-terminus could be regulating chitin synthase Chs2 activity, in such a way that Cyk3 would be required as well. Interestingly, the C-terminus of Inn1 localises at the site of division in a manner similar to full-length Inn1 [19], which shows that the Inn1 C-terminus can still interact with components of the actomyosin ring. Using the yeast two-hybrid assay we found that the Inn1 C-terminus interacted with Chs2 (Chs2-1-629). Interestingly, the Inn1 C-terminus binds to the N-terminal tail of Chs2 (Chs2-1-215), which has been shown to be regulated by CDK activity, whereas a fragment of Chs2 that contains only its catalytic domain (Chs2-215-629) was unable to interact with the Inn1 C-terminus (Fig 7A). In order to test whether the Inn1 C-terminus could bind Chs2 in vivo, we generated a yeast strain in which the TAP epitope was fused to the Inn1 C-terminus. We then cultured Inn1 C-terminus-TAP and control cells (as detailed in Fig 2A) and found that Chs2 interacted with the Inn1 C-terminus (Fig 7B). To study whether the Inn1 C-terminus could regulate chitin synthase activity associated with Chs2, chs3Δ cells were transformed to generate strains that overexpressed Chs2 or Chs2 together with the Inn1 C-terminus in order to perform an in vitro chitin assay (Fig 7Ci). We found that the Inn1 C-terminus had an inhibitory effect on Chs2 activity (Fig 7Cii). We have shown that C2 overexpression had a direct positive impact on Chs2 chitin synthase activity (Fig 2C), whereas the remainder of the protein, that is the C-terminus, displayed an adverse effect on Chs2 function (Fig 7C). Subsequently, we aimed to find out how full-length Inn1 might regulate the chitin synthase activity of Chs2. Cells overexpressing Chs2, or Chs2 at the same time as full-length Inn1 (Fig 7Di), were used to assay chitin synthase activity. We found that full-length Inn1 negatively regulated Chs2 enzymatic activity (Fig 7Dii). Thus, our findings indicate that the Inn1 C-terminus blocks the ability of the C2 domain to induce chitin synthase activity associated with Chs2.
To determine the role of Inn1 and Cyk3 during cytokinesis, we aimed to study the defects associated with C2-HOF1 cells in which Cyk3 was depleted. First, to investigate whether these cells had a defect in actomyosin ring formation or contraction we followed the presence of Myo1 protein at the site of division. To tightly control Cyk3 inactivation, we included a ‘heat-inducible degron’ cassette at the N-terminus of Cyk3-aid and created a double degron td-cyk3-aid as previously reported (S8A Fig) (‘td’ indicates the temperature sensitive degron) [51] [52] [53] [50] [20]. We grew asynchronous cultures of C2-HOF1 td-cyk3-aid MYO1-GFP and control cells at 24°C before synchronising cells in G1 phase with mating pheromone (Fig 8A). After the induction of both Ubr1 E3 ligase and Tir1 F-box protein, together with the addition of auxins to rapidly deplete Td-Cyk3-aid protein, cells were released at 24°C from G1 block. We observed that both mutant and control cells progressed up to anaphase in a similar manner. Unlike control cells, C2-HOF1 td-cyk3-aid MYO1-GFP accumulated as binucleate cells, which would reflect a failure in cell division (Fig 8Ai). Localisation of Myo1 at the site of division was observed with similar kinetics in both strains, which would indicate that mutant cells were able to assemble and contract the actomyosin ring (Fig 8Aii). To confirm the kinetics of ring contraction, time-lapse video microscopy was used (Fig 8B). C2-HOF1 td-cyk3-aid MYO1-GFP and control cells were grown in a similar way as for Fig 8A. After the cells had budded and completed S-phase, nocodazole was added to synchronise the cells, this time in G2-M-phase. Cells were washed into fresh medium and subsequently placed in the time-lapse slide to examine the localisation of Myo1 every two minutes as cells completed mitosis at 24°C. To ensure that both strains were treated in an identical fashion, the cultures were mixed before the cells were transferred to the time-lapse slide (the control cells expressed Spc42-eQFP and thus could be distinguished from C2-HOF1 td-cyk3-aid cells) (see Materials and Methods for details). Twenty-two movies each were examined for control and C2-HOF1 td-cyk3-aid MYO1-GFP cells, and contraction of the actomyosin ring was observed with similar kinetics (Fig 8B). The average period from the initiation of contraction to the final disappearance of the ring was similar in control and C2-HOF1 td-cyk3-aid MYO1-GFP cells (a mean value of 5.54 min in control cells compared with 5 min in the mutant strain). C2-HOF1 td-cyk3-aid MYO1-GFP cells never showed a contracted ‘spot’ as control cells. In mutant cells, the actomyosin ring disassembled before reaching the final contraction stage (‘spot’), which explains the slightly shorter contraction period in mutant cells (Fig 8B). Taken together, these experiments demonstrate that C2-HOF1 cells in which Cyk3 has been inactivated are able to form an actomyosin ring and subsequently to contract and disassemble.
To examine whether Inn1 localisation was altered by the lack of Cyk3 in C2-HOF1 strain, we cultured control and C2-HOF1 td-cyk3-aid cells, both of which expressed Inn1-GFP. Cells were treated in the same way as for Fig 8A. We found a higher percentage of cells with Inn1-GFP accumulation in C2-HOF1 td-cyk3-aid cells (Fig 9A). To confirm that the lack of Cyk3 function in C2-HOF1 cells promoted Inn1 accumulation, we transformed C2-HOF1 td-cyk3-aid with either CYK3 or cyk3-2A alleles and cultured them in an identical way as above (Fig 9A). We showed that inactivation of the transglutaminase-like domain of Cyk3 (cyk3-2A) induced Inn1-GFP accumulation (Figs 9B and S8B). Since actomyosin ring contraction seemed to have no delay in C2-HOF1 td-cyk3-aid cells (Fig 8), which would explain the accumulation of Inn1-GFP, we aimed to determine whether Inn1-GFP localisation is slightly advanced in mutant cells. We grew control and C2-HOF1 td-cyk3-aid cells in the same fashion as described previously (Fig 8B). After inactivation of Td-Cyk3-aid protein, cells were synchronised in G2-M-phase with high mitotic CDK by addition of nocodazole to the culture medium. Inn1 protein is unable to be localised at the site of division before cells down-regulate CDK activity at the end of mitosis [31] [41] [54] (Fig 9C). However, we found that 33% of C2-HOF1 td-cyk3-aid cells were able to localise Inn1-GFP with high mitotic CDK, which indicates that inactivation of Cyk3 prompts earlier Inn1 localisation.
So far our findings indicate that Inn1 and Cyk3 formed a ternary complex with chitin synthase Chs2 (Fig 5C). In addition, our biochemical and genetic analysis show that Inn1 and Cyk3 control chitin synthase activity associated to Chs2 (Figs 6 and 7). Therefore, we aim to determine whether C2-HOF1 td-cyk3-aid cells have a defect in primary septum formation in vivo. We cultured C2-HOF1 td-cyk3-aid and control cells under the same conditions as indicated above for Fig 8A, but in the presence of calcofluor upon release from G1 block to stain the primary septa. We found that cells expressing C2-Hof1 together with wild-type Inn1 in the absence of Cyk3 were unable to lay down a primary septum, which would suggest that Chs2 function was impaired (Fig 9D). We were unable to determine whether Chs2 localisation occurred in C2-HOF1 td-cyk3-aid cells, as triple mutant cells are dead or extremely sick (C2-HOF1 CHS2-GFP td-cyk3-aid) (S8C Fig). However, we found that the delivery of Chs2 to the site of division seems to be similar in control and Cyk3-depleted cells, since both type of cells showed similar dynamics of the localisation of Chs2-GFP (S8D Fig) and the formation of primary septum (S8E Fig). Therefore, our data indicate that the lack of Cyk3 does not prevent Chs2 localisation and primary septum deposition.
Overall these data show that Inn1 regulates Chs2 activity at the site of division, where the C2 domain induces Chs2 function, whereas the C-terminus of Inn1 seems to have an inhibitory effect on Chs2. Our results indicate that Cyk3 counteracts this inhibitory role since Cyk3 becomes essential under conditions in which the Inn1 C-terminus plays a more relevant role, such as in C2-HOF1 cells. In addition, Cyk3 seems not to have a role in the delivery of Chs2 vesicles to the cleavage site.
Chs2 protein interacts with actomyosin ring components to build the IPCs at the end of mitosis. To understand the importance of these interactions for Chs2 localisation and maintenance at the site of division we studied the fluorescence signal associated with Chs2-GFP in controls cells and in cells in which a particular actomyosin ring component had been previously inactivated. Cultures of CHS2-GFP and iqg1-td CHS2-GFP cells were grown at 24°C and cells were synchronised in G1 phase of the cell cycle with mating pheromone, before rapidly inactivating Iqg1 at 37°C. Upon release from G1 arrest at 37°C, iqg1-td cells completed mitosis but were unable to divide unlike the control cells (Fig 10Ai). Importantly, medial rings or contracted dots of Chs2 were not observed in the absence of Iqg1 (Fig 10Aii and 10Aiii). This shows that Iqg1 is essential for the localisation of Chs2. Additionally, it has been described that Iqg1 interacts with Hof1 and Inn1, and we have previously reported that Inn1-Iqg1 interaction is required for the Inn1 protein to be localised at the division site [19]. Our next step was therefore to determine whether Iqg1 is important for Hof1 to interact with the actomyosin ring. We grew HOF1-GFP and iqg1-td HOF1-GFP cells as detailed in Fig 10A and we found that the absence of Iqg1 caused a defect in Hof1 localisation (Fig 10B). Taken together, these experiments indicate that the Iqg1 protein is crucial for building functional IPCs at the end of mitosis in budding yeast.
To determine whether Chs2 localisation requires the presence of Hof1 or Inn1 at the site of division, we grew CHS2-GFP and hof1-td CHS2-GFP cells asynchronously at 24°C and cultured them in the same manner as described for Fig 10A. After rapid depletion of Hof1 in cells we followed Chs2-GFP localisation and found that Chs2-associated signal at the cleavage site was compromised when Hof1 protein was inactivated (S9A Fig). Subsequently, we investigated whether Chs2 localisation depends on Inn1 protein. We observed that Chs2 protein was still able to localise, although Chs2 dynamics at the site of division seemed to be affected, since we detected less Chs2 at the division site (S9B Fig). Finally, we investigated whether the lack of Hof1 and Inn1 at the same time would affect Chs2 recruitment. CHS2-GFP and hof1-td inn1-td CHS2-GFP cells were grown and, after rapid inactivation of Hof1 and Inn1, we found that Chs2 localisation at the cleavage site was entirely dependent on the presence of both Hof1 and Inn1 (Fig 10C), which is the same result as showed above for iqg1-td cells. Our findings would suggest that the dynamics of the chitin synthase Chs2 at the site of division in budding yeast requires the interaction with either Iqg1 protein or Hof1 and Inn1 together.
Our data indicate that budding yeast cells assemble at the end of mitosis protein complexes that we have named ingression progression complexes (IPCs) to coordinate actomyosin ring contraction, plasma membrane ingression and primary septum formation. The IPCs include Myo1, Iqg1, Hof1, Inn1, Cyk3 and Chs2. We propose that the IPCs indeed form the central machinery with which cells are able to coordinate cytokinetic events, which provides a mechanistic explanation for the tight coordination between them [55] [17] [8].
Our data support a model whereby IPCs are assembled in a sequential and highly regulated fashion to control first the localisation and then the activation of the chitin synthase Chs2 at the end of mitosis, which plays a key role in the tight coordination of actomyosin ring contraction, plasma membrane ingression and primary septum formation (Fig 11) [8] [37] [38] [39] [14]. Myo1 and Iqg1 serve as initial building blocks with which the other IPC components Hof1, Cyk3, Inn1 and Chs2 then interact [19] [31] [32] [33] [34] [30] [29] [27]. Specific inactivation of Myo1 or Iqg1 prevents the localisation of the other components of IPCs, which highlights their scaffolding role [19]. Moreover, IPC members commonly displayed Myo1-dependent immobility during cytokinesis, supporting further that Myo1 plays a scaffolding role in the assembly of IPCs [24].
Our findings indicate that Hof1 would facilitate Chs2 localisation at the site of division whereas Inn1 and Cyk3 are essential for the activation of chitin synthase activity. In addition to its role in promoting the formation of the primary septum, Hof1 plays an important role in the assembly of the actomyosin ring in S. cerevisiae, together with the protein Rvs167 [56] [55] [28] [29] [30]. It appears that the basic principles of action of budding yeast Hof1 and its fission yeast orthologue Cdc15 are likely to be similar, with both proteins contributing to assembly of the actomyosin ring as well as to the stability of the contracting ring and/or septum formation [57] [58] [59] [29] [30] [11]. In budding yeast it has been described that Hof1 interacts directly with Chs2 and stabilises the chitin synthase at the cleavage site [29]. Accordingly, we found that Chs2 localisation at the bud neck is clearly compromised in Hof1-depleted cells. In addition, we observed that increased expression of Chs2 rescues defects associated with the lack of Hof1 (S9C Fig). However, a hypermorphic version of Chs2 was unable to supress the cell division defect produced by Hof1 inactivation (S9D Fig), unlike what occurs with Inn1 or Cyk3-depleted cells [20]. Our findings indicate that Hof1 assists in the incorporation of Chs2 at the cleavage site and not in its activation. Our data is consistent with recent reports that showed how fission yeast cells lacking Cdc15 fail to accumulate at the cleavage site the protein that plays an analogous role to Chs2 [11], namely the transmembrane glycosyltransferase Bgs1 (beta(1,3)-glucan synthase), which lays down the primary septum during cytokinesis in fission yeast cells [9] [60].
Chs2 appears to be delivered to the plasma membrane in an inactive form and is then activated in situ by a mechanism that has not previously been understood [31] [19] [20] [32] [40]. Our data indicate that the Inn1 and Cyk3 proteins are indeed directly responsible for such activation (Fig 11). The interaction of Chs2, Inn1 and Cyk3 would require the inactivation of mitotic forms of Cyclin Dependent Kinase (CDK) and the dephosphorylation of CDK targets such as Chs2 and Inn1 by the Cdc14 phosphatase [38] [39] [33] [54] [61] [62].
Our model proposes that Inn1 binds to Chs2 at the end of mitosis, but the C-terminus of Inn1 keeps Chs2 chitin synthase inactive. It has been reported that a version of Chs2 that lacks its N-terminal domain show high levels of chitin synthase activity (Martinez-Rucobo et al 2009), which suggests that the N-terminal tail of Chs2 (Chs2-1-215) negatively regulates its own activity. Interestingly, our data indicate that the C-terminus of Inn1 binds precisely to the N-terminal tail of Chs2 (Fig 7A) and, consistently with these data, we described how the C-terminus of Inn1 blocks chitin activity associated to Chs2. In addition, it seems that the N-terminal tail of Chs2 could be regulating as well the interaction between Chs2 and Cyk3 (Fig 5A). Cyk3 becomes incorporated to the IPCs and precisely releases chitin synthase activity of Chs2 from such a block and consequently the C2 domain of Inn1 acts in conjunction with Cyk3 to activate the catalytic domain of Chs2 (Fig 11). In addition to its catalytic activity, Chs2 has attractive features to serve as an anchor between the actomyosin ring and the plasma membrane, since it is the only component of the IPCs that has a transmembrane domain embedded in the plasma membrane. It appears that, in budding yeast, having such a physical link with the plasma membrane is not sufficient for ingression, since cells need active glycosyltransferase for extracellular matrix remodelling and its coordination with actomyosin ring contraction during cytokinesis [19] [20].
Interestingly, each component of budding yeast IPCs has an orthologue in fission yeast cells with a role during cytokinesis, although the molecular mechanism by which they regulate cell division is not yet understood in all cases [60] [22] [63] [58] [49]. Fission yeast cells have orthologues of Inn1 and Cyk3, namely Fic1 and Cyk3, which share the same structure as their budding yeast counterparts and, in addition, they have been described to play a role during cytokinesis [58] [49]. The Fic1 protein interacts with Cdc15 and adds structural integrity to the actomyosin ring and prevents it from collapsing during cell division [58], whereas fission yeast Cyk3 has been suggested to play a role in coupling actomyosin ring contraction and primary septum formation, although the molecular mechanism remains unclear [49]. The presence of chitin in S.pombe septum is uncertain, but instead it has been proposed that S.pombe Chs2 would play a structural role and would be required for proper actomyosin ring contraction and stability [64]. Intriguingly, in fission yeast the glycosyltransferase Bgs1 could play the same role as Chs2 in budding yeast, although the polysaccharide that Bgs1 produces is different (Chs2 synthesises chitin, which is a polymer of N-acetylglucosamine; Bgs1 synthesises glucan, which is a polymer of D-Glucose). Bgs1 is an integral membrane protein with its catalytic domain located at the cytoplasmic side of the membrane like Chs2 [9] [60]. Interestingly, the lack of Bgs1 promotes actomyosin ring sliding along the plasma membrane, which supports the idea that Bgs1 could function as an anchor between the actomyosin ring and the plasma membrane [11] [65].
How human cells perform the coordination of actomysin ring contraction, plasma membrane ingression and ECM remodelling remains largely unknown. Taking into account the conservation of the basic cytokinetic mechanisms [22] [60] [3], it will be interesting to determine whether glycosyltransferases also play a role during cytokinesis in higher eukaryotes.
The strains used in this study are listed in S1 Table. Yeast cells were grown in rich medium (1% Yeast Extract, 2% peptone, 0.1 mg per ml adenine) supplemented with 2% glucose (YPD), 2% Raffinose (YPRaff) or 2% Galactose (YPGal) as the carbon source with the exception of cells for time-lapse video microscopy, for which we used Synthetic Complete medium at the end of the experiment. We arrested cells in the G1 phase of the cell cycle by the addition of alpha factor mating pheromone to the medium at a final concentration of 7.5 μg per ml. We arrested cells in the G2-M phase of the cell cycle by the addition of nocodazole to the medium at a final concentration of 5 μg per ml. To degrade proteins fused to the ‘heat-inducible degron’ and ‘auxin-inducible degron’ we followed procedure described previously [66] [50]. In experiments with temperature sensitive degron strains and for strains expressing fused proteins (C2-HOF1, C2-CHS2 and TG-HOF1), 0.1mM CuSO4 was included in the growth medium, as all of them were expressed from the CUP1 promoter. To stain primary septa of living cells, calcofluor was added when specified 30 minutes after release from G1 block to a final concentration of 0.05 mg per ml and culture was incubated further for at least 60 minutes.
Two-hybrid analysis was performed using the vectors pGADT7 and pGBKT7 (Clontech). Cells were grown for two or three days at 30°C on Synthetic Complete medium lacking leucine and tryptophan (non-selective) or lacking leucine, tryptophan and histidine (selective).
The plasmids used in this study to express recombinant proteins in E.coli are based on the ‘pET’ series (Novagen) and are listed in S2 Table. To isolate recombinant protein complexes from extracts of E.coli cells, we followed the scheme illustrated in S1A Fig and as it was described previously [30]. The various protein fragments were expressed individually as ‘Streptag’ or ‘6His-tag’ fusions. Cells containing each of the fusions were grown at 37°C, after which the expression of the recombinant protein fragments was induced with 1 mM IPTG. Subsequently, the cultures were mixed, so that each cell extract would contain two recombinant proteins (Figs 1B, 1C, 1D, 2A(ii), 2A(iii) and 5B) or three recombinant proteins (Fig 5C). In the case of the controls, a culture with an empty vector was mixed with the corresponding cultures expressing recombinant proteins. The Streptag-fusions were then isolated from the cell extracts in 1 ml of Strep-Tactin Superflow (2-1206- 025, IBA GmbH), before elution with 2.5 mM d-Desthiobiotin (D1411, Sigma). The eluted material was then diluted and incubated with 1 ml of Ni-NTA Agarose (30230, Qiagen), and bound protein complexes were eluted with sequential 0.5 ml aliquots of buffer containing 250 mM imidazole. Following the addition of 3X Laemmli buffer to the eluted samples, 20 μl of each purified sample was resolved by SDS-PAGE.
To monitor the association of proteins in yeast cell extracts we followed methods previously described [67] [46] with slight modification as cell extracts were spun down at 20 000 x g. We have isolated tagged proteins by immunoprecipitation with magnetic Dynabeads M-270 Epoxy (Invitrogen) coupled to rabbit anti-sheep IgG (Sigma S-1265), 9E10 anti-MYC monoclonal antibody (Cancer Research Technology) or 12CA5 anti-HA monoclonal antibody (Cancer Research Technology).
We detected the indicated proteins by immunoblotting with previously described polyclonal antibodies to Inn1 [20] or by using polyclonal anti-FLAG antibody (Sigma F-7425), or monoclonal 9E10 (anti-MYC), or 12CA5 (anti-HA). To detect Chs2 (rabbit polyclonal) and Cyk3 (sheep polyclonal), we raised polyclonal antibodies to 25 kDa portions of each protein (S1C and S1D Fig), expressed as His-tagged recombinant proteins in E.coli and purified in a denatured form. The TAP tag was detected using the rabbit peroxidase anti-peroxidase complex (Sigma P-2026). For mass spectrometry analysis of protein content, the digested peptides were analysed by nano LC/MS/MS with an ‘Orbitrap Velos’ (ThermoFisher) and the data were processed as described previously [68] [69].
Cell membrane isolation and chitin synthase activity assays were performed as described previously [20], clearly detecting chitin synthase activity associated to Chs2 (S10A Fig) In isolated cell membranes Chs2 is nearly inactive, unless protease treatment is used to bypass inhibition [14]. To clearly identify whether full-length Inn1, Inn1 C2 or Inn1 C-terminus had an effect on chitin synthase associated with Chs2, we plotted the percentage of active chitin synthase calculated as percentage of chitin synthase activity (without trypsin) compared to the maximum chitin activity reached by the same sample (with trypsin): (chitin synthase activity without trypsin / chitin synthase activity with trypsin) x100. Error bars represent SEM values calculated for each of the experiments.
Samples used to measure the DNA content were fixed with 70% ethanol. Subsequently samples were processed and stained with propidium iodide after RNA digestion, as described previously [53]. The proportion of binucleate cells was determined by observing the same samples under the microscope as those employed for flow cytometry [70]. We examined 100 cells for each sample.
Pictures of cells and colonies on agar plates were taken after 24 hours (YPD medium) or 30 hours (YPGal medium) with a Nikon CoolPix 995 camera attached to a Nikon Eclipse E400 microscope. We used calcofluor (Fluorescent Brightener 28; Sigma; F3543-1G) to stain the primary septa of live cells. We tested that there was a direct correlation between calcofluor staining and chitin synthase activity associated to Chs2 in vivo (S10B Fig). Calcofluor was added 30 minutes after release from G1 block to a final concentration of 0.05 mg per ml and the culture was further incubated for at least 60 minutes. Calcofluor-stained cells were observed live. To quantify primary septum deposition, we examined 100 cells with primary septum for each sample, and measure the relative signal intensity of the primary septum using Image J software [71]. To observe GFP-tagged proteins, the cells were fixed with 8% formaldehyde for 10 minutes and subsequently washed twice with PBS [19]. We examined 100 cells for each sample. Phase contrast and fluorescence microscopy images of cells grown in liquid culture were obtained with a Nikon A1R Microscope and an Orca R2 camera (Hamamatsu) with objective lens Plan Apo TIRF 100x oil DIC 1.49NA, and LightLine single-band filter set FITC Semrock. The illumination source was the Nikon Intensilight C-HGFIE (ultrahigh Presure 130W Mercury lamp), and we used NIS elements software. We analysed eleven z-sections with a spacing of 0.375 μm to facilitate the examination of the whole cell for all experiments. In all cases, the exposure time, sensor gain, and digital adjustments were the same for the control and experimental samples.
Time-lapse video microscopy was performed using DeltaVision system with Olympus IX-71 microscope and CoolSNAP HQ2 Monochrome camera. The objective lens was Plapon 60X0 1.42 NA. The illumination source was the 300W xenon system with liquid light guide, and we used Softworx Resolve 3D acquisition software. Cells were grew in an IBIDI cells in focus 15 micro-slide (8 well 80827 glass bottom). The base of the time-lapse chamber is formed by a glass coverslip that we coated with a 5 mg per ml solution of the lectin Concanavalin A (Sigma L7647), and then washed with water and dried for 30 minutes. We analysed 10 z-sections with a spacing of 0.4 μm.
The microscopy data were deconvolved, except for cells stained with calcofluor, using Huygens (SVI) according to the “Quick Maximum Likelihood Estimation” method and a measured point spread function. The deconvolved data set was viewed with Image J software [71].
|
10.1371/journal.pcbi.1002678 | Deciphering Interactions in Moving Animal Groups | Collective motion phenomena in large groups of social organisms have long fascinated the observer, especially in cases, such as bird flocks or fish schools, where large-scale highly coordinated actions emerge in the absence of obvious leaders. However, the mechanisms involved in this self-organized behavior are still poorly understood, because the individual-level interactions underlying them remain elusive. Here, we demonstrate the power of a bottom-up methodology to build models for animal group motion from data gathered at the individual scale. Using video tracks of fish shoal in a tank, we show how a careful, incremental analysis at the local scale allows for the determination of the stimulus/response function governing an individual's moving decisions. We find in particular that both positional and orientational effects are present, act upon the fish turning speed, and depend on the swimming speed, yielding a novel schooling model whose parameters are all estimated from data. Our approach also leads to identify a density-dependent effect that results in a behavioral change for the largest groups considered. This suggests that, in confined environment, the behavioral state of fish and their reaction patterns change with group size. We debate the applicability, beyond the particular case studied here, of this novel framework for deciphering interactions in moving animal groups.
| Swarms of insects, schools of fish and flocks of birds display an impressive variety of collective patterns that emerge from local interactions among group members. These puzzling phenomena raise a variety of questions about the behavioral rules that govern the coordination of individuals' motions and the emergence of large-scale patterns. While numerous models have been proposed, there is still a strong need for detailed experimental studies to foster the biological understanding of such collective motion. Here, we use data recorded on fish barred flagtails moving in groups of increasing sizes in a water tank to demonstrate the power of an incremental methodology for building a fish behavior model completely based on interactions with the physical environment and neighboring fish. In contrast to previous works, our model revealed an implicit balancing of neighbors position and orientation on the turning speed of fish, an unexpected transition between shoaling and schooling induced by a change in the swimming speed, and a group-size effect which results in a decrease of social interactions among fish as density increases. An important feature of this model lies in its ability to allow a large palette of adaptive patterns with a great economy of means.
| Collective motion occurs across a variety of scales in nature, offering a wealth of fascinating phenomena which have attracted a lot of attention [1]–[5]. The self-organized motion of social animals is particularly intriguing because the behavioral rules the individuals actually follow and from which these remarkable collective phenomena emerge often remain largely unknown due to the tremendous difficulties to collect quality field data and/or perform controlled experiments in the laboratory. This situation does not prevent a thriving modeling activity, thanks to the relative ease by which numerical simulations can be conducted. However, most models of moving animal groups are built from general considerations, educated guesses following qualitative observations, or ideas developed along purely theoretical lines of thought [6]–[9]. Even when authors strive to build a model from data, as in the recent paper by Lukeman et al. [10], this model building amounts to writing down a fairly complicated structure a priori, involving many implicit assumptions, and to fit collective data to determine effective parameters, yielding a best-fit model.
On the other hand, recent studies within the physics community of simple, minimal models for collective motion have revealed an emerging picture of universality classes [11]–[15]: Take, for instance, the Vicsek model, arguably one of the simplest models exhibiting collective motion. In this model, point particles move at constant speed and choose, at discrete time-steps, their new heading to be the average of that of their neighbors located within unit distance. Many of these behavioral restrictions can be relaxed without changing the emerging collective properties. Fluctuations of speed can be allowed, some short-range repulsion (conferring a finite size to the particles) can be added, even explicit alignment can be replaced by inelastic collisions, etc., all these changes will still produce the remarkable nonlinear high-density high-order bands emerging near onset of collective motion, and, deeper in the ordered moving phase, the anomalously strong number fluctuations which have become a landmark of the collective motion of polarly aligning self-propelled particles [16]–[20]. The Vicsek model, in this context, is one of the simplest members of a large universality class defined by all models sharing the same large-scale properties. This universality class can be embodied in the continuous field equations that physicists are now able to derive. With such a viewpoint, different models in this class merely differ in the numerical values of their parameters [21]–[23], very much like different fluids are commonly described by the Navier-Stokes equations and differ only in their viscosity and other constitutive parameters.
Significant features nevertheless may be altered when a qualitatively important feature is changed, such as the symmetry of the aligning interaction, or added, as when local attraction/repulsion between individuals is also considered [8], [24] In this latter case, for instance, no strong clustering and high density band appears when attraction is sufficiently strong, and finite groups may keep cohesion in open space as most natural groups do. These models yield a more complex phase diagram where collectively moving groups may assume gas-like, liquid-like or even moving crystal states as the two parameters controlling alignment and cohesion are varied.
So, it remains important to know how individuals make behavioral choices when interacting with others, not only from a social ethology and cognitive viewpoint, but also because i) different behavioral rules may make a difference in small enough groups and ii) the analysis of local-scale data that this requires may lead to discover features eventually found to give rise to different qualitative collective properties. A recent instance can be found in the results on the structure of starling flocks gathered by Ballerini et al. [25]: They have ignited an ongoing debate about the possibility that individuals might interact mostly with neighbors determined by topological rules and not by metric criteria as assumed in most models. While this message has intrinsic value for the study of decision-making processes in animal groups, it was also shown recently that such metric-free, topological interactions are relevant, in the sense that they give rise to collective properties that are qualitatively different from those of metric models [26]. Thus, in this case, an individual-level ingredient suggested by data, which had been only partially and theoretically considered before [6], [7], [27], defines new classes of collective properties. Given that animals are likely to possess more sophisticated behavior than, say, sub-cellular filaments displaced by molecular motors, one can expect more hidden features to play an important role at the collective level. This is a central finding of the recent work by Katz et al. where a careful analysis of groups of two and three fish revealed that the mechanisms at play are, at least in the golden shiners studied there, much more subtly intertwined that in existing fish models [28]. Indeed they concluded that alignment emerges from attraction and repulsion as opposed to being an explicit tendency among fish. Whether fish display some mechanisms of active alignment or only attraction/repulsion is likely to lead to different patterns as interactions accumulate over time. In short, extracting interaction rules from individual scale data is crucial not only for animal behavior studies, but also because heretofore overlooked features can be found decisive in governing the emergent collective properties of moving animal groups.
Here, we assess the power of a bottom-up methodology to build models for animal group motion from data gathered at the individual scale in groups of increasing sizes. We use data obtained by recording the motion of barred flagtails ( Kuhlia mugil) in a tank. In natural conditions, the barred flagtail form schools with a few thousands individuals along the reef margin of rocky shorelines, from just below the breaking surf to a depth of a few meters. However the size of these schools is much smaller than in species like the sardine or the Atlantic herring.
Our analysis is incremental: in a previous work we characterized the spontaneous behavior of a single fish, including wall-avoidance behavior [29]. Here, using pairs of fish, we first characterize the response function of one fish depending on the position and orientation of the other fish. Then we calibrate multiple fish interactions, using data in larger groups. At each step, the already-determined factors and parameters are kept unchanged and the new terms introduced in the stimulus-response function and the corresponding new parameters are determined from data with nonlinear regression routines (see Statistical Analysis in Materials and Methods). The resulting model is validated by comparing extensive simulations to the original data. Often, different functional forms are tested and we determine which one is most faithful to the data. When no significant difference is found, the simplest version is retained, following a principle of parsimony.
Experiments with 1 to 30 fish were performed in shallow circular swimming pools that let the fish form quasi 2-dimensional schools (see Fig. 1A and Video S1, S2, S3, S4). At the collective level, we observe a transition from schooling to shoaling behavior when the density of fish increases in the tank: the group polarization , which measures the degree of alignment, is high in groups of two and five fish, even if sometimes we do observe some breaks in the synchronization, while in larger groups, when , it remains low (Fig. 1B). Within each group size, we notice some variability, the most striking effect being an increase of the synchronization level with the individuals velocity in groups of two fish.
For every group size, fish move continuously and quickly synchronize their speed to a well defined, but replicate-dependent value (Fig. S1). The fish trajectories are smooth, differentiable and the instantaneous speed has a well-defined mean and root mean square fluctuations of about 10–20% which are found to be uncorrelated to , the angular velocity of the fish orientation (Fig. S2). On this basis, fish can be modeled as self-propelled particles moving in 2D space at constant speed and the only dynamical variable retained is . Moreover, since the recorded trajectories, be they extracted from a single fish or from small groups in the tank, are always irregular/stochastic, our model takes the form of coupled stochastic differential equations for the angular velocities of each fish. Note that if noise acts on rather than the fish position or heading, trajectories are smooth and differentiable, as observed.
We have shown elsewhere that single fish trajectories in barred flagtails are very well described by an Ornstein-Uhlenbeck process acting on the instantaneous curvature, or, equivalently, on [29]. When the fish is away from the tank wall, the distribution of is nearly Gaussian with zero mean and variance , where is the characteristic time of the (exponentially decaying) autocorrelation function of . To avoid collisions with the tank walls, we found that a single fish adjusts its current turning speed towards a (time-dependent) target value where is a parameter, is the distance to the point of impact on the wall should the fish continue moving straight ahead, and is the angle between the current heading of the fish and the normal to the point of impact (see Fig. 2A). In short, obeys the stochastic differential equation:(1)where is a Wiener process of variance reflecting the stochasticity of the behavioral response. Non-linear regression analysis of the above model against our experimental data yielded excellent agreement and accurate estimations of and . Note that in the present work we adopted a slightly different form for the wall avoidance term with regards to the exponentially decreasing one of Ref. [29], since it actually prevents fish from crossing the tank boundary, while both ansatz are similar as fish moves away from tank walls (Fig. S8A).
The stimulus/response function of a single fish in the tank is directly expressed by how varies with the relative position of the fish and the wall. We now assume that this framework holds when two fish and are present in the tank by defining how, for fish , its turning speed is modulated by the combined stimuli due to the wall and to fish . Almost all existing fish behavior models, on the basis of common sense, intuition, and sometimes experimental evidence [30]–[37], offer a combination of three basic ingredients: short distance repulsion (to avoid collisions), alignment for intermediate distances, and attraction up to some maximal range. Here, we dispose of repulsion not only because we want to allow for the rare experimentally observed over- and under-passings events, but mostly because we do not need to incorporate it explicitly to avoid collisions (see below and Video S1, S2, S3, S4). In contrast with most existing “zonal” models, and because there is little cognitive/physiological evidence for a sudden switch between alignment and attraction, we want to allow for continuous, distance-dependent weighting between alignment and attraction in agreement with the recent findings of Katz et al. [28]. These two factors a priori depend on the geometrical quantities defining the location of fish from the viewpoint of fish : their distance is involved, but also , the angular position of fish with respect to , the current heading of fish , as well as their relative heading difference (Fig. 2A). The main angular variable for explicit alignment is, as usual, , whereas for attraction it is ; both may also depend on . The stimulus/response function of fish thus combines a priori wall avoidance, alignment and attraction in some unknown function with parameters and (reaction to the wall), , and : .
Next, in the spirit of an expansion around the no-interaction case, we write the expression for above as the sum of three terms:(2)where the “main” variables have been placed first for each term. The wall avoidance term depends explicitly on to reflect a possible screening of the wall by the other fish. We have tested the influence of this by introducing a dependence in the wall avoidance term determined for the single-fish behavior. Essentially, was made smaller for . But this brought no significant improvement, so we keep as found previously.
On general grounds, one expects that the relative importance of the positional interaction (attraction) to the velocity interaction (alignment) increases with . Given that the fish are constrained in a rather small tank, a limited range of inter-distances is effectively explored. In the spirit, again, of a small-distance expansion, a satisfactory choice is given by a linear dependence of on , while is independent of . Of course, such a functional choice cannot be correct at large distances since then would take large unrealistic values, meaning that the fish would spend enormous amounts of energy turning toward a distant “neighbor” (see the Discussion for more comments on this point).
The attraction interaction must depend on , the relative angle with the other fish position: it is reasonable to assume that a fish is not attracted much towards a neighbor located behind, and of course this term must be zero when the other fish is right ahead, yielding . A simple, compatible, trigonometric function representing the leading term of a Fourier expansion is the sine function. We thus write where is a parameter controlling the weight of the positional information. Finally, we neglect the possible dependence on : the way a fish would turn toward the position of a neighbor does not depend on the orientation of that fish. This is especially natural when this interaction dominates, i.e. when the neighbor is far away. Moreover knowing the other fish orientation is a cognitively expensive and/or time consuming process at larger distances.
The alignment interaction is mostly characterized by its functional dependence on . The main constraint here is that (the two fish are then already aligned). Here again, the simplest choice is as in most models [8]–[10]. Including higher harmonics (e.g. ) would allow to account for the few observed nematic alignment events where a fish remains anti-aligned with its neighbors. However, incorporating this term did not improve the faithfulness of the model to our dataset, so we keep only the leading sine function. In principle, the strength of alignment can also depend on : less attention may be paid to “back neighbors”. We have tested simple and reasonable choices for the dependence of on , e.g. , but this did not lead to significant improvement so we kept no angular position dependence in the alignment interaction. We thus write, finally: where is a parameter controlling the weight of the orientational information.
To summarize the case of two fish and , the stimulus/response function in the general evolution equation (1) is thus finally written:(3)
Using nonlinear regression analysis, the faithfulness to our data of the model consisting of Eqs. (1) and (3) was found very good for each of our two-fish recordings and the 5 parameters , , , and were estimated for each fish. We find clear dependences of the estimated parameters on , the average speed of each fish (see Fig. 3A). In particular, , , and are found proportional to , whereas and no significant -dependence appears for . Results regarding this last parameter are the least convincing, with a large dispersion of individual values. This is mostly due to the confinement of fish in the tank: the positional interaction never dominates alignment, preventing its accurate estimation. Nevertheless it is crucial to note here that without these positional interactions the model fails to match the data. Furthermore, we have tested a posteriori our ansatz by testing each contribution (either wall avoidance, neighbor position or neighbor orientation) after the other twos have been subtracted from the fish response according to Eq. (3). Results show an excellent agreement between our ansatz and the mean fish response (for more details see Fig. S8 B–D).
Note that these results mean also that the wall avoidance is actually governed by , the time it would take the fish to hit the wall, rather than the distance . Conversely, , the relaxation time of the angular velocity, is better expressed as the ratio between a characteristic length and the speed . These -dependences were then incorporated explicitly in the model:(4)with(5)where , and are now constants over all fish. Running again our nonlinear regressions using this form, and using data for all replicate, allows for a more accurate estimation of the parameters , , , and now the same for all fish. We find , , , and .
To validate this experimental finding, these parameter values were used in simulations of the model which were compared directly to the data. Good agreement is found not only for statistical quantifiers of the emergent synchronization between the two fish (see Fig. 3C), but in fact also for the dynamics: see for instance Video S1, S2, S5, S6 and the time series of polarization which show the same intermittent behavior (Fig. 3B). We emphasize that the model captures the experimental observation that the orientational order is lower when the swimming speed is lower, and is better in faster groups (Fig. 3B, C).
Can multiple-fish interactions be factorized into pairs? This is often taken for granted, following a typical physics approach where this assumption is routinely made. However, recent work has suggested that this is not valid when describing pedestrian interactions in a crowd [38]. Even more recently, Katz et al. argued that this is also the case for groups of three golden shiners [28] (but see [39] for the case of birds). Here, our data set is too small to allow for an in-depth analysis of group behavior at the level of detail that was accomplished above for two fish, mostly because many more variables are involved, but the quality of the pair approximation can be evaluated a posteriori. Assuming that multiple fish interactions are indeed essentially made of the sum of the pair interactions involved, Eq. (5) is extended to(6)where is the (current) neighborhood of fish which contains individuals. In our observations with fish, individuals mostly stayed together, suggesting that individuals remains aware of all others. Using all-to-all, equal-weight coupling, we found good agreement between data and simulations of Eqs. (4) and (6) (see Fig. S3). This justifies a posteriori the factorization in pairs and the use of two-fish parameters for groups, but also the overall normalization factor in Eq. (6), which indicates that, in the stimulus response of a fish, wall avoidance and the averaged influence of neighbors keep, on average, the same relative importance irrespective of the group size. The raw, “force-like” un-normalized superposition would yield too strong a coupling.
For the larger group sizes, all-to-all equal-weight coupling quickly becomes unrealistic, and one must determine the set of neighbors a fish interacts with. In principle, abundant data recorded in larger tanks would allow to discriminate between alternative choices, but our experimental recordings are too short for this. Nevertheless, many choices can be eliminated: the usual one, which consists in cutting off interactions at fixed distances (zonal models), is inconsistent with our continuous weighting of alignment and attraction with fish inter-distance. Based on an analysis of starling flocks, Ballerini et al. have argued that these birds actually pay attention to their 6–8 closest neighbors, irrespective of the density of the flock [25]. Coming back to our observations, this non-metric choice of neighbors can, however, lead to unrealistic situations when, for instance, a fish is leading a small group, since then this fish will only pay attention to those behind, even if individuals are located at intermediate distances ahead (but see Fig. S7). A simple, reasonable, non-metric solution is that of neighbors determined by the Voronoi tessellation around each individual: this allows for continuous weighting between alignment and attraction and avoids the caveat mentioned above in the case of a fixed number of closest neighbors. Moreover, given the rather small inter-distances observed, individuals beyond the first shell of Voronoi neighbors are largely screened out, so that our final choice was that of the first shell of Voronoi neighbors (see Fig. 2B). Using this, the validation of the model simulated with fish using the parameters is again quite satisfactory (see Fig. S3).
This is however not true anymore for larger groups which display too high a polarization when using the parameters (whereas distance predictions remains satisfactory, see Fig. S3). Our approach actually allows to further investigate this discrepancy. We estimate the parameters at the individual scale for each fish with our nonlinear least-square procedure using the Ito-integrated version of the Ornstein-Uhlenbeck process of Eqs (4) and (6) for each fish time series (see Statistical Analysis). Thanks to this parametric inversion strategy, we have been able to extract the parameter values for each replicate separately (Fig. 4A). The model predictions with these replicate-based parameters yield a near-perfect match with the data (Fig. 4B). The results confirm that, within the limits of statistical accuracy, the parameters and their v-dependence remain about the same up to N = 10, in agreement with the above findings ; but in larger groups there is a decreased tendency of fish to react to their neighbors, which both concerns the alignment and positional interactions (Fig. 4A).
Characterizing and modeling the interactions between individuals and their behavioral consequences is a crucial step to understand the emergence of complex collective animal behaviors. With the recent progress in tracking technologies, high precision datasets on moving animal groups are now available, thus opening the way to a fine-scale analysis of individual behavior [37], [40]–[42]. Here we adopted a bottom-up modeling strategy for deciphering interactions in fish shoaling together. This strategy is based on a step-by-step quantification of the spontaneous motion of a single fish and of the combined effects of local interactions with neighbors and obstacles on individuals motion. At each step, one model ingredient is considered and checked against experimental data. The required parameters are determined using a dedicated inversion procedure and the numerical values of these parameters are kept unchanged in the following steps, yielding, in the end, a model without any free parameter. Such an incremental procedure fosters the explicit enunciation of the rationale behind each functional choice, and differs from searching the best set of free parameters to fit large-group data [10], [43]. Proceeding step by step also puts stronger constraints on matching, since the incorporation of additional behavioral features at each step assumes the stability of the previously explored behaviors and of the corresponding model parameters. Using pairs of fish, we were able to show how positional and directional stimuli combine, and the crucial role of the swimming speed in the alignment interaction. At intermediate sizes, multiple fish interactions could be faithfully factorized into pair interactions albeit in a normalized form. However we found that at even larger group sizes our incremental modeling approach fails to accurately reproduce the collective dynamics.
We explored this point further, still considering the statistical behavior of each fish separately, but only using the data corresponding to the large-group experiments. We concluded that our model could still grasp the observed individual and collective features but with smaller positional and alignment coefficients. We believe that this decrease in reactivity to neighbors is a consequence of the high density already imposed by confinement effects. Indeed, our model predicts that large groups adopting the high neighbor reactivity found in smaller groups would remain polarized also in open space, keeping group cohesion with an average distance to neighbors of about two body lengths (Fig. S6). Since the largest groups we observed in the tank are already characterized by such a typical neighbors distance due to confinement effects, we argue that lower interaction strengths may simply indicate the fish vanishing need to actively react to neighbors position and heading in order to maintain a high density. This could be, for instance, a physiological consequence of the density per se: the physiological and behavioral consequences, for an individual, of living in dense groups, known as group effect, have been described in numerous species from insects to vertebrates [44], [45]. Our results investigation suggests that this sensitivity may be represented in a quite straightforward manner, preserving the model shape of Eqs. (4) and (6) and only modifying the interaction parameters. This conjecture, of course, could only be validated by experiments on large groups conducted in open space or larger tanks. While we believe in a positive answer, namely that without too strong a confinement, individuals would react to the perceived neighbors the same way regardless of the overall group size, we leave this question for future investigations on group effect in fish schools.
Our approach yielded a novel type of fish school model whose main features are its built-in balancing mechanism between positional and orientational information, a topological interaction neighborhood, and explicit dependencies on fish speed. Note that similar features were recently uncovered for another species thanks to a novel data analysis procedure [28]. The smooth transition from a dominant alignment reaction when a neighbor is close to attraction when it is far away is in line with a simple additive physiological integration of both information [46]. The linear dependence of the positional interaction strength on fish inter-distance obviously cannot hold for sparse groups, and will have to be modified by introducing a long-distance saturation when dealing with situations where confinement effects are weaker. Even if we claim that a Voronoi neighborhood was the best choice to account for our data thus extending the relevance of topological interactions, we also checked that our conclusions were robust against this choice, by testing a simple K-Nearest Neighbors network of interactions (which remains topological [25]). We computed the model predictions with the parameters estimated for groups of N = 2 fish, but considering only the K nearest neighbors for increasing values of K (K = 1 to 7, and 10). The results are reported in Fig. S7 ; the main impact of a lower level of connectivity is a decrease of polarization, but it does not lead to better predictions at the collective scale. Interestingly, the best predictions were found with a number of nearest neighbors that corresponds to the average number of neighbors belonging to the first shell in a Voronoi neighborhood (, Fig S7–B). This number of influential nearest neighbors is remarkably similar to the one found in starlings [25] and in contrast with recent results found by Herbert-Read et al. in mosquito fish [47]. Further dedicated experiments will be required to discriminate between alternative choices of the relevant neighborhood.
The speed dependence of the parameters, directly derived from our data, is in contrast with most previous fish school models. It leads to an increase of group polarization with swimming speed, a direct consequence of the predominance of alignment at high speed (see Video S7). In natural conditions, this mechanism could be involved in the transitions from shoaling at low speed often associated with feeding behavior to polarized schooling at high speed associated with searching for food. Such speed change could also be elicited by the detection of a threat and abrupt transitions can occur when fish suddenly increase their speed, for instance generating a flash expansion (see Video S8). The question of whether the propagation of such an excitation wave within large schools can generate an efficient collective evasion call for further experimental tests [48].
The reason why our approach was fruitful in spite of the limited amount of data available lies largely in the suitable properties of the behavior of the fish studied: the smooth fluctuations of tangential speed and their de-correlation from angular velocity variations were essential in limiting the number of variables at play but also allowed for a faithful account of single fish behavior by a simple Ornstein-Uhlenbeck process. Clearly it is likely that more complicated solutions will be needed for other species where tangential and angular accelerations are intimately coupled and/or the underlying stochastic process is not as transparent [28]. Nevertheless, we expect that, pending sufficient amounts of data, our approach could be successfully applied to more complex situations occurring in various biological systems at different scales of organization.
Our experiments were all carried out in full accordance with the ethical guidelines of our research institutions and comply with the European legislation for animal welfare. The welfare of fishes in the tanks was optimized with a continuous seawater flow, a suitable temperature, and oxygen content. The maximum density in the holding tank was lower than . During the experiments, low mortality occurred (five individuals). At the end of the experiment, the fish were released at their capture site.
The experiments were performed from April to June 2001 at the Sea Turtle Survey and Discovery Centre of Reunion Island. Barred flagtail Kuhlia mugil (Forster) were caught in March 2001 in the coastal area around Reunion Island. 80–100 fishes were conveyed to the marine station and housed in a holding tank of 4 m diameter and 1.2 m depth. Fishes were fed daily ad libitum with a mixture of aquaria flake-food and pieces of fish flesh. Fishes were considered acclimatized when all of them feed on the aquaria flake-food. This weaning period lasted 15 days. Experiments were performed in a circular tank similar to the holding tank. Opaque curtains were placed around and above the tank to obtain diffuse lighting and to reduce external disturbances from the environment. The tank was supplied with a continuous flow of seawater [49]. Since currents may influence fish behavior, the seawater inlet pipe was placed vertically and the water flow was stopped throughout the observation periods. A digital video camera (Sony model CDR-TRV 900E) was fixed at 5 meters above the tank and tilted at to observe the whole tank. The remotely operated video camera was fitted with a polarizing filter and a wide-angle lens. Groups of N = 1 to 30 fish were introduced in the experimental tank and acclimatized to their new environment for a period of 20 min. Their behavior was then recorded at 24 fps for 2 mins. Prior to each trial, the fish were deprived of food for 12 hours to standardize the hunger level and were transferred to the experimental tank. The relative shallowness of the water ensured quasi two-dimensional motion. Five replicates per group size using different individuals were performed. Eighty per cent of the trials were performed in the morning to avoid possible conditions of strong wind that may disturb the fish, and sunshine that may render light inside the tank unsuitable for video recording. A first data processing consisted in sampling 12 images per second out of the 24 images recorded by the video camera. A custom-made tracking software was then used to extract high-quality, smooth trajectories from the video recordings, with crossing ambiguities resolved by eyes (see Video S3, S4). In order to get even higher precision data, the head position and the orientation of each fish in groups of N = 2 were acquired with a manual tracking software (Video S1, S2).
Model parameters were estimated from each fish time series separately (typical series are shown on Fig. S4). In order to perform the estimation of the parameters , , , and in the stochastic differential equation (1), (3) and (5), we considered its discrete-time version using Ito integration over , assuming is small enough so that is constant [50]:(7)where i = 1,2 and is given by Eq. (3) or (5). Estimates for the parameters were obtained using a standard non-linear least squares procedure (we employed the nls package of the statistical environment R [51]) either separately for each fish using Eq. (3) or for all fish together using Eq. (5). Residuals given by were checked to be Gaussian-distributed (see Fig. S5) and their variance yielded .
The model was simulated within a virtual tank, using the estimates of behavioral parameters extracted by statistical analysis from time-series in groups of fish. The fish heading (direction of motion) and position were updated by Euler integration, following:(8)where .
For each value, numerical simulations were performed over 120 seconds (a time corresponding to the duration of individual experiments with real fish) with a time step . A transient time of was discarded before measuring statistical averages. We computed the mean value and the variance over time of the global polarization(9)and of the neighbor inter-distance(10)
This yielded an estimation of the expected measures distribution under model hypothesis and over the typical observation time of experiments. We then computed the mean and confidence interval of such distributions, to obtain the expected mean and variance (with their confidence intervals) of alignment and of neighbor inter-distance. This provided the check of the model against experimental data. The above procedure was repeated varying the mean speed over the range covered by the experimental data, with the results plotted in Fig. 3C. The same procedure was adopted to make predictions for higher group sizes, using the stimulus/response function as determined by equation (5) with interacting neighbors defined by first neighbors in a Voronoi tessellation (For a set of points, Voronoi tessellation divides the space in different cells, each the locus of space closer to its center than to any other points in : at each time step space is divided in Voronoi cells centered around the fish position, with Voronoi neighbors being the fish lying in neighboring cells (Fig. 2B). For each experimental replicate, the same measures were repeated with the parameters extracted from the replicate, and the corresponding initial conditions (Fig. 4B).
By construction, our method does not “learn the parameters to make the model fit”, contrasting with a more usual procedure which consists in stating an a priori model and searching a best set of free parameters that optimizes its collective patterns towards the observed collective properties (namely, make the model fit at the collective scale). In such cases, it is known that several models can adjust the data at the collective scale (because the search for best match is unconstrained and can be performed for each model, so that the collective level underdetermines the individual level).
In the present study, once the model has been formulated, that is, once we identified in the experiments with pairs of fish the nature of stimuli (the orientation and relative position of neighboring fish, and how they combine to determine the response of a focal fish), we estimated the values of 5 parameters at the individual scale. So for each fish, we measured its behavioral response (i.e. the change of its turning speed) for each configuration of stimuli encountered in its path.
Only then, we tested whether these parameters measured at the individual level can explain the observations at the collective scale with no free parameters. For each group independently, we thus checked that the model allows a quantitative matching concurrently at individual and collective scales. This confirmed that our model calibrated with the parameters estimated from the third derivative of the fish position (i.e. the change in the turning speed) was able to reproduce quantitatively the statistics resulting from the time integration of the coupling between fish (polarization, inter-distance). Moreover the same procedure applied separately on each group size revealed, on the one hand, the dependences of the estimated parameters on the swimming speed (using groups of N = 2 fish), and on the other hand, the modulation of interactions' strength with group size (in the largest groups).
|
10.1371/journal.pbio.1001747 | Exocytosis of ATP From Astrocytes Modulates Phasic and Tonic Inhibition in the Neocortex | Communication between neuronal and glial cells is important for many brain functions. Astrocytes can modulate synaptic strength via Ca2+-stimulated release of various gliotransmitters, including glutamate and ATP. A physiological role of ATP release from astrocytes was suggested by its contribution to glial Ca2+-waves and purinergic modulation of neuronal activity and sleep homeostasis. The mechanisms underlying release of gliotransmitters remain uncertain, and exocytosis is the most intriguing and debated pathway. We investigated release of ATP from acutely dissociated cortical astrocytes using “sniff-cell” approach and demonstrated that release is vesicular in nature and can be triggered by elevation of intracellular Ca2+ via metabotropic and ionotropic receptors or direct UV-uncaging. The exocytosis of ATP from neocortical astrocytes occurred in the millisecond time scale contrasting with much slower nonvesicular release of gliotransmitters via Best1 and TREK-1 channels, reported recently in hippocampus. Furthermore, we discovered that elevation of cytosolic Ca2+ in cortical astrocytes triggered the release of ATP that directly activated quantal purinergic currents in the pyramidal neurons. The glia-driven burst of purinergic currents in neurons was followed by significant attenuation of both synaptic and tonic inhibition. The Ca2+-entry through the neuronal P2X purinoreceptors led to phosphorylation-dependent down-regulation of GABAA receptors. The negative purinergic modulation of postsynaptic GABA receptors was accompanied by small presynaptic enhancement of GABA release. Glia-driven purinergic modulation of inhibitory transmission was not observed in neurons when astrocytes expressed dn-SNARE to impair exocytosis. The astrocyte-driven purinergic currents and glia-driven modulation of GABA receptors were significantly reduced in the P2X4 KO mice. Our data provide a key evidence to support the physiological importance of exocytosis of ATP from astrocytes in the neocortex.
| Brain function depends on the interaction between two major types of cells: neurons transmitting electrical signals and glial cells, which control cerebral circulation and neuronal homeostasis. There is a growing evidence of the participation of astrocytes in regulating neuronal excitability and synaptic plasticity via the release of “gliotransmitters,” which include glutamate and ATP. The importance of ATP release from astrocytes was suggested by studies that demonstrated its contribution to neuronal activity and sleep homeostasis via modulation of known “purinergic” receptors. But the mechanisms underlying gliotransmitter release and the physiological significance of direct glia-to-neuron communication remain unknown and intensively debated. Here, we investigate the release of ATP from astrocytes of brain neocortex and demonstrate that astrocytes can release ATP by Ca2+-dependent exocytosis, most likely from synaptic-like microvesicles. We also find that vesicular release of ATP from astrocytes can directly activate excitatory signaling in the neighboring neurons, operating through purinergic P2X receptors. We saw that activation of these P2X receptors by astrocyte-driven ATP down-regulated the inhibitory synaptic signaling in the neocortical neurons. Our results imply that exocytosis of gliotransmitters is important for the communication between astrocytes and neurons in the neocortex.
| ATP acts as neurotransmitter mediating excitatory synaptic transmission and synaptic plasticity in the central nervous system [1],[2]. There is growing evidence that ATP can also play an important role in signal transfer between neuronal and glial circuits and within glial networks [3]–[6]. ATP can regulate growth and development of neural cells and contribute to various pathological processes [7]–[9]. Action of ATP is mediated by ionotropic P2X and metabotropic P2Y receptors abundantly expressed in many types of neurons and glial cells [1],[2]. By virtue of the high Ca2+-permeability of P2X receptors and the ability of P2Y receptors to stimulate IP3-dependent Ca2+ release from endoplasmic reticulum, purinergic receptors can transmit robust Ca2+-signals and thereby modulate activity and trafficking of excitatory and inhibitory receptors [10],[11]. In addition to direct actions mediated by P2 purinoreceptors, ATP can initiate secondary neuromodulation via P1 adenosine receptors after rapid degradation by ecto-nucleotidases to adenosine [2],[12]. Different mechanisms of ATP release have been identified, such as vesicular release from nerve terminals [13],[14] and several nonvesicular pathways, including concentration gradient-driven diffusion through gap-junction hemichannels, anion channels, and dilated P2X7 receptors [4],[7],[15],[16].
A physiological role of ATP release from astrocytes has been suggested by the participation of ATP in the propagation of glial Ca2+-waves [17]–[19] and significant contribution of ATP and adenosine to the astroglia-driven modulation of neuronal activity and sleep homeostasis [4],[12],[20]. There is growing evidence, albeit obtained mostly in cell cultures, that the release of gliotransmitters may share common mechanisms of vesicular neurotransmitter release such as a dependence on the proton gradient and vesicular transporters, SNARE proteins, andintracellular Ca2+ elevation [9],[21],[22]. Importantly, astroglial-driven release of ATP and modulation of synaptic plasticity in the hippocampus were suppressed in transgenic mice expressing a dominant-negative SNARE (dn-SNARE) domain selectively in astrocytes [12].
However, the mechanism of gliotransmitter release from astrocytes has been disputed [16]. Ca2+-dependent exocytosis of glutamate and ATP, mainly from cultured hippocampal astrocytes, has been reported [21]–[26]. By contrast, alteration of astroglial InsP3-mediated Ca2+-signaling did not have a significant effect on glutamatergic synaptic transmission in the hippocampal slices [27], fuelling the debate on the role for glial exocytosis in more intact tissue [16]. Still, more recent in situ and in vivo data demonstrated an effect of astroglial InsP3-mediated Ca2+-signaling on cholinergic modulation of synaptic plasticity in hippocampus and neocortex [28]–[30]. At the same time, two recent studies reported the possibility of nonvesicular release of astroglial glutamate through the TREK-1 and best1 channels [31] and the lack of immunostaining for vesicular glutamate transporters in brain astrocytes [32], contrasting with the bulk of evidence for glial exocytosis obtained by variety of different techniques [21]–[26],[33]. Thus, physiological relevance of Ca2+-dependent exocytosis of gliotransmitters remains controversial.
In this study, to avoid possible artifacts of cell culture, we investigate release of ATP from acutely isolated cortical astrocytes [34] and astrocytes in the neocortical slices. We provide several lines of evidence for (1) existence of functional vesicular mechanism of Ca2+-dependent gliotransmitter release in neocortical astrocytes, (2) quantal P2X receptor-mediated currents directly activated in neocortical neurons by release of ATP from astrocytes, and (3) glia-driven purinergic modulation of GABAergic transmission that is impaired by astrocytic expression of dn-SNARE or deletion of P2X4 receptors.
As a first step in demonstrating the presence of quantal Ca2+-dependent release of ATP from astrocytes, we used a sniffer cell approach where ATP release from astrocyte was detected by HEK293 expressing P2X2 receptors (see Materials and Methods for details). We compared release of ATP from astroglia of somatosensory cortex of wild-type mice and from mice conditionally expressing the SNARE domain of VAMP2 selectively in astrocytes (dn-SNARE mice) [12],[20].
Acutely isolated cortical astrocytes were separately loaded with Ca2+-indicator Fluo-4 and photosensitive Ca2+-chelator NP-EGTA [35] and then distributed into a recording chamber containing preplated HEK293-P2X2 cells (Figure 1). Whole-cell voltage-clamp recordings were made from a HEK293-P2X2 cell lying in the immediate vicinity to astrocytes (Figure 1A). Identification of astrocytes was confirmed by their functional characterization at the end of experiment, including low input resistance, lack of voltage-gated Na+-conductance, large K+-conductance, large conductance mediated by glutamate transporters, and NMDA receptors lacking Mg2+-block (Table S1). Thus, influence of nonastrocytic cells in the experiments reported below can be ruled out.
Brief flash of UV-light caused the uncoupling of Ca2+ from NP-EGTA (monitored by Fluo-4 Ca2+-indicator). This was accompanied by an asynchronous burst of phasic currents in the adjacent ATP-sensitive sniffer cell in 17 of 20 experiments (Figure 1B). P2 receptor antagonist PPADS (10 µM) prevented the detection of UV-evoked phasic currents (n = 7 of 7; Figure 1B), confirming that they were mediated by ATP acting via P2X receptor.
Some phasic currents were observed in the sniffer cells having astrocytes laying on their surface even before triggering the Ca2+-rise in the astrocyte, but at much lower frequency (Figure 1B). These baseline events were not detected in the absence of astrocytes in any of 10 native HEK293-P2X2 cells tested. Application of PPADS (10 µM) reduced the frequency of baseline and UV-evoked events correspondingly by 94%±3% and 97%±3% (Figure 1C), confirming the purinergic nature of all phasic currents. An exocytotic mechanism of ATP release was suggested by the activity-dependent staining of astrocytes with vesicular marker FM1-43 (Figure 1C).
To further test the role of an elevation of cytosolic Ca2+ concentration as a trigger for the release of ATP from astrocytes, we applied agonists of astroglial metabotropic PAR-1 receptors (see also Figure S1) [35],[36] and ionotropic NMDA receptors [6],[24]. The PAR-1 receptors were chosen as an ample method of astrocyte activation because of the ability to activate IP3 pathway and their predominant expression in glial but not in neuronal cells [35],[36]. Similar to UV-uncaging, activation of cytosolic Ca2+-transients in the astrocyte either by application of the PAR-1 agonist TFLLR (10 µM; n = 10) or by application of NMDA (20 µM; n = 7) elicited bursts of spontaneous currents in the adjacent ATP-sensitive sniffer cells (Figure 2A). TFLLR and NMDA did not evoke any activity in the HEK293-P2X2 cells when applied in the absence of astrocytes (n = 5 agonist, unpublished data). The phasic currents in the sniffer cells had amplitudes of 11.2±2.4 pA (n = 34) and rise and decay time of 1.6±0.5 ms and 15.2±3.9 ms correspondingly (Figure 2D), thus resembling parameters of purinergic synaptic currents [13],[14].
As a test of a vesicular mechanism of ATP release, we isolated astrocytes from cortical slices pretreated with blocker of vacuolar-type H-ATPase bafilomycin A1 (1 µM) for 2 h. Treatment with bafilomycin caused a decrease in the amplitudes and frequency of phasic currents initiated by Ca2+-elevation in the astrocytes (Figure 2B). The mean amplitude of phasic purinergic currents activated in the sniffer cell by stimulation of bafilomycin-treated astrocytes was only 3.1±0.4 pA (n = 19). The overall charge transferred by phasic currents activated after stimulation of bafilomycin-treated astrocytes by UV, TFLLR, and NMDA was 9.7%±4.2% (n = 7), 11.8%±5.6% (n = 5), and 12.6%±5.8% (n = 7) of the corresponding control values (Figure 2E).
In support of exocytotic mechanism of ATP release, astrocytes obtained from dn-SNARE mice elicited a diminished burst of purinergic currents in sniffer cells regardless of the method used to elevate cytosolic Ca2+-level (Figure 2C,E). We also observed the SNARE-complex-dependent release of ATP from the isolated hippocampal astrocytes (Figure S2).
The amplitude histograms of the phasic P2X-mediated currents activated by elevation of astrocytic Ca2+ exhibited prominent second peak (Figure 2D; see also Figure S3) at amplitude twice that of the primary peak. Fitting of amplitude distribution with simple multiquantal binomial model (shown in Figure 2D as dotted line) gave a quantal size of 7.9±0.13 pA and release probability of 0.28±0.04. Similarly, fitting of the distributions of P2X-currents evoked in sniffer cell after application of NMDA and TFLLR gave quantal size of 8.14±0.15 pA and 8.05±0.11 pA, respectively.
It should be noted that astrocyte-driven purinergic currents observed in our experiments had much faster kinetics (10–25 ms) than nonvesicular release of gliotransmitters from hippocampal astrocytes, which was mediated by TREK-1 potassium channels and best1 chloride channels [31]. To verify the lack of contribution of nonvesicular mechanisms to the quantal purinergic phasic currents in the sniffer cells, we activated astrocytes by TFLLR in presence of TREK-1 channels inhibitor fluoxetine [37] and large conductance chloride channels inhibitors DIDS and NPPB [38]. Application of fluoxetine (100 µM), DIDS (300 µM), and NPPB (100 µM) did not have marked effect on the astrocyte-evoked phasic currents in the HEK293 cells in any of five experiments for each inhibitor (Figure S4).
Combined, the above results strongly suggest that activity-dependent release of ATP from cortical astrocytes occurs mainly via quantal exocytotic mechanism, dependent on SNARE protein complex.
Vesicular mechanism of ATP release from neocortical astrocytes was also supported by immunostaining of living isolated astrocytes with antibodies to vesicular nucleotide transporter VNUT1 and various vesicular, neuronal, and glial marker proteins (Figures S5 and S6). Although immunostaining of live cells has certain limitations (see Text S1) and should be interpreted with great caution, our data suggest the good co-localization of VNUT1 and synaptic vesicle (SV) markers (Figure S5A), which is in agreement with previous reports of vesicular location of VNUT1 [26] and presence of synaptic-like vesicles in astrocytes [22],[24]. We observed weaker correlation between VNUT1 and lysosomal markers cathepsin D and LAMP3 (Figure S5B,C), which goes in line with previous observation of astroglial ATP release by the lysosome exocytosis [23].
However, lysosomal exocytosis from astrocytes exhibited much slower kinetics [23] than the purinergic currents measured in the sniffer cells (Figures 1 and 2). This argues against a major contribution of this mechanism to the present observations. As the kinetics of sniffer cell responses are more consistent with millisecond time-scale of SV exocytosis from astrocytes [24], we suggest that the astrocyte-driven purinergic currents observed in our experiments could be triggered by exocytosis of ATP from synaptic-like vesicles. We also observed an immunoreactivity for vesicular glutamate transporter (VGLUT1) in the fraction of cortical astrocytes (Figure S5D,E; Figure S6A), which goes in line with data reported previously for hippocampal and cortical astrocytes [22],[33]. Of course, further investigation of exocytosis of glutamate from neocortical astrocytes is required, which is beyond the scope of this article.
Previously we have shown that cortical pyramidal neurons express ionotropic P2X purinoreceptor, which can be activated by synaptic release of ATP [13]. Hence, it might be plausible to detect the glia-driven contribution to purinergic current in neurons. We recorded whole-cell currents in the pyramidal neurons of neocortical layer 2/3 of brain slice at membrane potential of −80 mV in the presence of glutamate receptor antagonists CNQX (50 µM) and D-APV (30 µM) and irreversible blocker of GABA receptors picrotoxin (100 µM). Like our previous results [13], we observed residual nonglutamatergic excitatory spontaneous synaptic currents (Figure 3A,B); neither amplitude nor frequency of residual currents were affected by further increase in concentrations of glutamate and GABA receptors antagonists (n = 10 cells tested, unpublished data).
The amplitude of inward spontaneous excitatory currents (sEPSCs) was reduced by specific antagonists of P2X receptors PPADS (10 µM) and NF-279 (3 µM) correspondingly by 45%±1 3% (n = 7) and 56%±19% (n = 16); the sEPSC frequency was reduced by PPADS and NF-279 by 65%±22% and 69%±27%, respectively (Figures 3 and S7). At concentrations used, both PPADS and NF279 are selective for P2X receptors [39]–[41]. Based on these data as well as our previous work [11],[13], the spontaneous inward currents observed in cortical neurons in the presence of glutamatergic and GABAergic antagonists can be confidently attributed to the ATP receptors.
The partial inhibitory action of PPADS and NF-279 on nonglutamatergic sEPSCs could be explained by participation of homomeric P2X4 receptors, which are insensitive to these antagonists [1],[41],[42]. Since P2X4 subunit-containing receptors are abundantly expressed in the brain and could potentially contribute to neuronal purinergic signaling [1],[42]–[44], we used previously characterized P2X4 receptor knockout mice (P2X4 KO) [38]. In the P2X4 KO mice, application of 10 µM PPADS decreased the amplitude and frequency of nonglutamatergic sEPSCs (Figure 3C) by 74%±10% and 97%±6% correspondingly (n = 12); difference from the wild-type mice was statistically significant with p = 0.05 and 0.01. Taking into account that significant attenuation of sEPSCs can put them below the detection threshold, nonglutamatergic sEPSCs can be confidently attributed to neuronal P2X receptors.
We triggered the release of gliotransmitters from the astrocytes by rapid application of an agonist of PAR-1 receptor TFLLR to neocortical slices. As in hippocampus [35],[36], TFLLR (10 µM) triggered cytosolic Ca2+ rise predominantly in astrocytes (Figure S1). Application of TFLLR caused a dramatic increase in the frequency of ATP-mediated spontaneous currents (Figure 3A,D). TFLLR also elevated astrocytic Ca2+ in dn-SNARE mice, but the burst of purinergic spontaneous currents was not detected (Figure 3B,E). In the P2X4 knockout mice, the average increase in the purinergic sEPSCs frequency reached only 28%±15% (n = 12), which was significantly lower (p<0.005) than 72%±21% increase observed in wild-type mice (Figure 3C,F). Combined, these results demonstrate that activation of astrocytes can evoke synaptic-like purinergic currents in neurons. In addition to effects of knocking out and inhibiting P2X receptors, the purinergic nature of glia-driven spontaneous currents was corroborated by inhibitory action of ATP-hydrolyzing enzyme apyrase (Figure S8). The apyrase application significantly decreased the mean amplitude and frequency of sEPSCs and abolished the TFLLR-induced burst. The decrease in the sEPSCs frequency is most likely related to the reduction of their amplitude below threshold of detection.
In the wild-type mice, purinergic sEPSCs showed bimodal amplitude distributions (Figure 3G–I; see also Figure S9) with peaks at 3.1±0.9 pA and 5.7±1.6 pA (n = 14); decay time distributions had peaks at 9.1±0.9 ms and 15.3±1.8 ms (Figure 3G). TFLLR selectively increased the probability of detection of smaller and slower sEPSCs in all 14 neurons tested. In contrast, recordings made from cortical neurons of dn-SNARE mice did not show two peaks in the distributions of amplitude or decay time; the amplitude and decay time were not altered after TFLLR application (Figure 4H). In the P2X4 KO mice, the amplitude distribution of purinergic sEPSCs showed peaks as 2.5±0.7 pA and 4.2±1.1 pA; decay time distribution had peaks at 9.0±0.9 and 15.7±1.7 in control. Activation of astrocytes caused just a moderate increase in the proportion of smaller and slower sEPSCs in the P2X4 KO neurons (Figure 3I).
Elimination of the distinct population of smaller and slower currents by astrocytic dn-SNARE expression strongly suggests that this population of purinergic currents was elicited by exocytosis of ATP from astrocytes. The vesicular origin of slower purinergic sEPSCs was also supported by elimination of these events by treatment of the cortical slices with 1 µM bafilomycin A1 (Table S2). The slower and faster purinergic currents recorded in cortical neurons exhibited different quantal behavior (Figure S7). The slower purinergic currents evoked by application of TFLLR in the presence of TTX (Figure S3A) exhibited multiquantal amplitude distribution, whereas faster currents exhibited monoquantal distribution typical for miniature synaptic currents (Figure S7B,C).
Thus, detailed analysis of purinergic sEPSCs in the pyramidal cortical neurons revealed two distinct populations of events, which differ by their amplitude and kinetics. Based on their insensitivity to astrocytic dn-SNARE expression, larger and faster sEPSCs most likely have a neuronal origin. In contrast, sEPSCs of smaller amplitude and slower kinetics can be attributed to the vesicular release of ATP from astrocytes. In the following sections, we provide further experimental support of this notion.
We sought to obtain a parallel line of evidence for the vesicular ATP release from cortical astrocytes via an alternative approach: we measured ATP concentration in neocortical slice using microelectrode biosensors (see Materials and Methods), a technique that has been applied previously for evaluation of ATP release in several brain areas [7],[19]. Selective activation of astrocytic PAR-1 receptor by TFLLR induced a robust increase in the extracellular ATP concentration in the cortical tissues of wild-type mice; this increase was impaired in the dn-SNARE mice and was blocked by bafilomycin, confirming its astroglial origin and vesicular nature (Figure S10A). The increase in the “tonic” concentration of extracellular ATP after activation of astrocytes in the wild-type mice reached 1.1±0.4 µM (Figure S10B) and was inhibited by 84%±5% (n = 7) after incubation with bafilomycin. The TFLLR-evoked elevation of ATP concentration in the dn-SNARE mice was decreased by 56%±12% (n = 12) as compared to wild-type. These results support the significant contribution of vesicular mechanism to the activity-dependent release of ATP from cortical astrocytes. Taking into account that bafilomycin can inhibit only re-charging of ATP-storing vesicles and not all the cortical astrocytes express dn-SNARE protein, one could not expect the full inhibition of vesicular ATP release in these experiments. Thus, the incomplete inhibition of TFLLR-evoked ATP transients in the dn-SNARE mice and in the bafilomycin-treated neocortical slices of wild-type mice could hardly be attributed to a large contribution of nonvesicular release.
We have shown previously that stimulation of intracortical afferents is able to significantly elevate cytosolic Ca2+-level in the cortical astrocytes acting via ionotropic and metabotropic receptors to glutamate and ATP [5],[6],[34]. We asked whether an episode of high-frequency stimulation (HFS) could similarly trigger release of ATP from astrocytes in situ. As before, we monitored spontaneous purinergic currents in the pyramidal cortical neurons at a membrane potential of −80 mV in the presence of CNQX and picrotoxin (Figure 4). The short HFS train triggered more than a 2-fold elevation of the frequency of purinergic sEPSCs in the pyramidal neurons of wild-type mice (Figure 4A,C). Such an elevation did not occur in the dn-SNARE mice, where HFS train caused only a modest transient increase in the sEPSC frequency (Figure 4B,D).
The HFS-induced changes in the amplitude and kinetics of phasic purinergic currents had a complex pattern in the wild-type mice (Figure 4C,E). During the first 30 s after HFS train, the average amplitude of events increased to 12.4±4.2 pA (n = 6) as compared to 8.6±2.4 pA in the baseline conditions and their decay time was slightly larger (11.2±2.3 ms, n = 6) than in control (9.9±2.7 ms). The sEPSCs appeared in the neurons of wild-type mice 1–3 min after HFS train had lower amplitudes (6.8±1.6 pA, n = 6) and much larger decay times (13.4±3.6 ms) than in control. Analysis of the amplitude and decay time distributions (Figure 4E) revealed a significant increase in the number of smaller, slower sEPSCs after HFS train that formed the distinct fraction of purinergic spontaneous events. In addition to smaller and slower currents, the number of fast sEPSCs with fast decay times (9.2±2.5 ms) and large amplitudes (19.6±2.7 pA) was observed during the first 30 s immediately after stimulation (Figure 5E, purple lines). The amplitudes of these large currents corresponded to the double of unitary amplitude of fast purinergic currents; this explains a short-lived increase in the average amplitude immediately after HFS train.
The existence of two functionally distinct populations of purinergic events in the wild-type mice was corroborated by correlation between amplitude and decay time of sEPSCs (Figure 4G). The slower currents (decay time of 15.4±2.2 ms) had smaller amplitudes (5.5±1.3 pA, n = 6), but the faster currents (decay time of 9.2±1.3 ms) had higher amplitude (9.9±2.4 pA, n = 6). The amplitudes of slower currents closely agree with quantal amplitude of TFLLR-evoked slow purinergic currents recorded at −80 mV, whereas quantal amplitude of fast currents is close to the unitary size of TFLLR-insensitive fast purinergic sEPSCs (Figure S6B). The train of HFS significantly increased the number of slower spontaneous currents with smaller amplitude (Figure 4G).
In contrast to wild-type mice, the only effect produced by HFS train on purinergic sEPSCs in the dn-SNARE mice was the transient increase in the number of double-quantal fast currents (Figure 4D,F), which led to the brief increase in the average amplitude. The expression of dn-SNARE in astrocytes caused a selective loss of the smaller and slower sEPSCs (Figure 4H). These data strongly support the different origins of fast and slow purinergic sEPSCs, from neuronal terminals and astrocytes correspondingly.
To verify that slower purinergic sEPSCs originated from astrocytic ATP release directly, we tested the effect of diadenosine triphosphate (AP3A) and UTP, which have been shown presviously to strongly inhibit transport of ATP into astrocytic vesicles [26]. Since these substance are not specific VNUT antagonists and can have an action on purinergic receptors, we applied them intracellulary to minimize side effects. In order to increase the impact of single-cell perfusion, we chose neuron-astrocyte pairs lying in a close proximity (Figure 5A). A similar strategy has been previously used to test the effects of perfusion of Ca2+-chelators into astrocytes [45]. The feasibility of this approach is based on the high probability of synapses of a single neuron falling within functional island enwraped and controlled by single nearby astrocyte [46]. When astrocytes of wild-type mice were perfused only with fluorescent dye, two consecutive HFS episodes caused the burst of slow purinergic sEPSCs in the neighboring neurons (Figure 5B) of the magnitude similar to previous experiments with intact astrocytes (Figure 4). Prolonged intracellular perfusion of astrocyte with 1 mM of AP3A or UTP significantly attenuated the frequency of purinergic sEPSCs (Figure 5B–E) in 10 of the 12 experiments. The effect was more prominent after the second HFS episode. This was most likely related to the depletion of the releasable pool of ATP in astrocytes. Analysis of amplitude and decay time distributions showed that inhibitors of vesicular nucleotide transporters selectively affected the fraction of slower sEPSCs (Figure 5C), significantly decreasing their amplitude and frequency (Figure 5D). Although slower purinergic sEPSCs were not abolished completely, this might be explained by incomplete perfusion of distal astrocytic processes or release from other astrocytes. These results strongly suggest that smaller and slower purinergic sEPSCs originated directly from vesicular release of ATP from neighboring astrocytes.
In summary, our data provide compelling evidence that quantal release of astrocytic ATP activates a distinct population of purinergic currents in cortical pyramidal neurons.
ATP release from astrocytes can activate neuronal P2X and P2Y receptors. The following increase in cytosolic Ca2+-signals may trigger a variety of intracellular cascades implicated in the modulation of synaptic strength [1],[2],[11]. In particular, phosphorylation of postsynaptic GABAA receptors might provide an endogenous pathway for Ca2+-dependent regulation of synaptic strength [47],[48].
To test this hypothesis, we recorded inhibitory synaptic currents (IPSCs) in neocortical pyramidal neurons at a membrane potential of −40 mV (Figure 6) in the presence of glutamate receptor antagonists CNQX (50 µM) and D-APV (30 µM). Under these conditions, we observed inward purinergic currents simultaneously with outward Cl−-currents mediated by GABAA receptors (Figure 6). Similar to our previous results [13], outward IPSCs were completely inhibited by bicuculline in all 12 cells tested (unpublished data). The burst of ATP-mediated currents, induced in cortical neurons by activation of astrocytic Ca2+ signaling via PAR1 receptors (as shown in Figure 3A), was accompanied by significant decrease in the amplitude of GABA-mediated synaptic currents (Figure 6A,B). Amplitudes of evoked IPSCs and spontaneous mIPSCs recorded in the wild-type mice were reduced after application of 10 µM TFLLR by 44.3%±6.1% (n = 7) and 39.4%±6.3% (n = 12) correspondingly. In the dn-SNARE mice, the inactivation of GABA receptor-mediated synaptic currents was greatly diminished (Figure 6A,B). Application of TFLLR reduced the amplitude of evoked and spontaneous IPSCs in the cortical neurons of dn-SNARE mice just by 6.4%±8.7% (n = 8) and 4.3%±6.6% (n = 8), respectively. The difference in the action of TFLLR on the IPSCs in the wild-type and dn-SNARE was statistically significant with p<0.005 for both evoked and spontaneous currents. These results confirm the importance of astroglial exocytosis for the observed inactivation of GABA receptors. It also indicates the lack of unspecific action of PAR-1 agonist on inhibitory synaptic transmission.
The effect of TFLLR on inhibitory synaptic currents was mimicked by application of nonhydrolysable ATP analog ATP-γS (10 µM) and was considerably reduced by P2 receptor antagonist PPADS (10 µM). In the P2X4 KO mice, activation of astrocytes decreased the amplitude of evoked and spontaneous IPSCs (Figure 6A,B) only by 14.3%±8.2% and 15.7%±9.5% (n = 12); the difference between P2X4 KO and wild-type mice was significant with p<0.005. These data strongly support the participation of neuronal ATP receptors in the astrocyte-driven modulation of IPSCs.
We found out that exocytosis of gliotransmitters also caused long-term homeostatic modulation of inhibitory synaptic transmission. We observed a marked difference in the amplitude distribution of the baseline (control) amplitude of postsynaptic inhibitory currents in the wild-type and dn-SNARE mice (Figure 6C). The amplitude of mIPSCs in the cortical neurons before the application of TFLLR was much higher in the dn-SNARE than in the WT mice. The average baseline amplitude of mIPSCs was 23.5±8.3 pA in the dn-SNARE mice (n = 8) and 14.9±6.9 in the WT mice. Application of TFLLR caused the leftward shift of amplitude distribution of mIPSCs only in the WT mice; this shift was significantly reduced by PPADS (Figure 6C). The considerable difference in the baseline amplitude of miniature inhibitory currents in the WT and dn-SNARE mice provides the first evidence that vesicular release of gliotransmitters may be involved in the long-term homeostatic regulation of inhibitory transmission in the neocortex.
In order to elucidate the role of post- and presynaptic mechanisms in the modulation of inhibitory transmission, we evaluated the changes in the mean quantal content (Figure 6A), changes in the frequency of spontaneous mIPSCs (Figure 6B), and paired-pulse ratio (PPR) of evoked IPSCs (Figure 7). Neither the mean quantal content of IPSCs (Figure 6A) nor the mIPSCs frequency (Figure 6B) exhibited marked changes. However, we observed a moderate change in the PPR of IPSCs in the neocortical pyramidal neurons of wild-type mice (Figure 7A). The IPSCs evoked with a 50 ms interval in the control showed paired-pulse depression with mean PPR of 0.79±0.11 (n = 12), and application of TFLLR increased PPR by 0.16±0.07 (n = 8). The application of TFLLR did not cause a significant change in PPR in the dn-SNARE mice (Figure 7B,D), indicating the involvement of glial exocytosis in the mechanism. The effect of TFLLR was significantly reduced by PPADS (Figure 7A,D) and reproduced by application of ATP-γS both in the wild-type and dn-SNARE mice (Figure 7A,B,D). Hence, the presynaptic increase in the PPR of IPSCs was most likely mediated by ATP acting via P2 purinoreceptors. This result is consistent with previous reports of presynaptic facilitation of GABAergic synaptic transmission by ATP and P2 receptors [1],[2],[49].
There was no significant difference in the effects of TFLLR and ATP-γS between wild-type and P2X4 KO mice (Figure 7C,D). So the P2X4 receptors hardly make the large contribution in the increase of PPR, contrasting with their prominent role in the astrocyte-induced decrease of IPCS amplitude (Figure 6). One could speculate that presynaptic facilitation of IPSCs by glia-driven ATP can be mediated by other subtypes of P2X receptors and P2Y receptors, whose role in the presynaptic modulation in the brain was widely reported [1],[2].
More importantly, the above data clearly demonstrate that the large decrease in the amplitude of evoked and spontaneous IPSCs cannot be attributed to the presynaptic modulation of GABA release, and astrocyte-induced down-regulation of inhibitory transmission operates via postsynaptic mechanism.
The postsynaptic mechanism of IPSCs inactivation was corroborated by experiments where the Ca2+-dependent phosphorylation of GABA receptors was impaired by intracellular agents. First, we found that activation of P2X receptors in pyramidal neocortical neurons caused marked reduction of GABA-activated currents via Ca2+- and protein kinase C–dependent mechanism (Figure S11). Second, application of TFLLR to the cortical slices of wild-type mice did not cause marked reduction in the mIPSCs recorded in neurons perfused with intracellular Ca2+-chelator EGTA (10 µM) or protein kinase C antagonist GF109203X (30 nM). The mIPSC amplitude was reduced by just 2.5%±5.9% (n = 6) in the presence of intracellular EGTA and by 11.2%±5.5% (n = 4) in the presence of the protein kinase C antagonist (Figure S12). These results agree with previous reports on Ca2+- and PKC-dependent down-regulation of GABA receptors [47].
In addition to the fast IPSCs (“phasic inhibition”), central neurons also receive a diffusional inhibitory signal mediated by the extrasynaptic GABA receptors continuously activated by small concentrations of GABA present in the extrasynaptic space (“tonic inhibition”) [50],[51]. Extrasynaptic GABA receptors, responsible for tonic inhibition, have been reported to undergo Ca2+-dependent phosphorylation [47],[48]. Thus, one could expect the impact of astrocyte-driven ATP not only on phasic but also on tonic inhibitory signaling in the neocortical pyramidal neurons. To verify this hypothesis, we used a conventional experimental paradigm where tonic inhibition is assessed by the change in the whole-cell holding current under action of GABA receptor blockers [50], in our case 50 µM bicuculline. In the wild-type mice (Figure 8A, top), layer 2/3 neurons showed the tonic current of 39.9±8.3 pA (n = 20) at membrane potential of −80 mV. The tonic current in the pyramidal neurons dn-SNARE mice was almost two times higher than in the wild-type (Figure 8B, top), reaching 76.9±15.1 pA (n = 10). A significant difference in the amplitude of tonic current between wild-type and dn-SNARE mice suggests that exocytosis of gliotransmitters from cortical astrocytes can modulate the activity of extrasynaptic GABA receptors in the adjacent neurons. Consistent with this notion, activation of gliotransmitter release by TFLLR caused a marked upward shift in the holding current in the pyramidal neurons of wild-type mice (Figure 8A, middle), and the rest of holding current was efficiently diminished by bicuculline. Application of TFLLR did not have a notable effect on tonic current in the dn-SNARE mice (Figure 8B, middle). The amplitude of tonic current recorded under action of TFLLR in the wild-type, dn-SNARE, and P2X4 KO mice was correspondingly 12.6±6.8 pA (n = 14), 72.2±9.1 pA (n = 7), and 43.4±7.2 pA (n = 6). The relative decrease in the amplitude of tonic current caused by TFLLR was, respectively, 68%±14%, 6%±8%, and 25%±11%. The down-regulation of tonic current by PAR-1 agonist was considerably attenuated by antagonist of ATP receptors (Figure 8A, bottom). The effects of TFLLR and PPADS in the wild-type mice as well as a difference between mice strains were statistically significant as indicated in Figure 7C. It has been recently shown that Ca-dependent modulation of astrocytic GABA GAT3 transporters in the hippocampus can alter an extracellular GABA level elevating the tonic current and decreasing the mIPSCs due to GABA receptor desensitization [52]. One should note that this pathway did not contribute significantly to PAR-1 agonist-induced modulation of inhibitory currents in neocortical neurons, since we observed a decrease in both phasic (Figure 6) and tonic currents (Figure 7). This notion was corroborated by our finding that blocking of the astroglial GAT3 GABA transporter increased the tonic current and decreased the amplitude of mIPSCs both in the wild-type and dn-SNARE mice (Figure S13). These data also show that dn-SNARE expression does significantly alter the activity of GABA transporters in cortical astrocytes.
Taken together, our experiments in brain slices demonstrate that exocytosis of ATP from cortical astrocytes in situ can activate the ATP receptors in the adjacent pyramidal neurons, leading to down-regulation of synaptic and extrasynaptic GABA receptors.
Detection of ATP released from acutely isolated cortical astrocytes using sniffer cells demonstrated the SNARE protein-dependent exocytosis of ATP from cortical (Figures 1 and 2) and hippocampal (Figure S2) astrocytes. Our experiments in situ show that release of ATP from astrocytes can be sensed by neighboring neurons where glia-derived ATP can activate neuronal purinoreceptors. Quantal behavior of astrocyte-induced phasic purinergic currents (Figures 3, 4, and S6) and their elimination in the bafilomycin-treated cortical slices (Table S2) confirm that they originated from the vesicular release of ATP. Combined with previous observation of astroglial purinergic modulation of neuronal activity in the hippocampus, cortex, and basal forebrain [12],[20], these data strongly support the universal character of a vesicular mechanism of ATP release as a gliotransmitter.
In the sniffer cell experiment we used P2X2 receptors as a detector for ATP. They have a moderate affinity for ATP (EC50 around 10 µM) that allowed us to avoid saturation and resolve the quantal behavior of astrocyte-evoked purinergic currents with a mean quantal size of about 8 pA (Figure 2D). To activate such response, the concentration of ATP released from astrocytes should have reached at least the low micromolar range. Detailed calculations, performed using the approach applied for extrasynaptic release and diffusion of glutamate [16],[53],[54], show that such concentration can be reached after release of about 1,000 molecules from a single glial synaptic-like vesicle, and the size of the “active spot” due to diffusion of ATP can reach as far as 1–2 µM (Figures S14A and S15). This estimation is in line with our data obtained using ATP microelectrode biosensors (Figure S10) as well as with previous evidence that the extracellular ATP level in the brain can reach 1–100 µM depending on the physiological and pathological context [1],[7],[55].
Similar calculations, as above, argue against involvement of lysosomal release in generation of purinergic currents observed in our experiments. Indeed, the typical lysosome has a diameter of about 300 nm and can contain more than 1,000,000 molecules of gliotransmitter [16]—that is, 1,000 times more than in the synaptic-like vesicle. Even though lysosome undergoes kiss-and-run exocytosis releasing only 10% of its content [16], one could expect the peak of a sniffer cell response to reach at least 200–500 pA, with most receptors on its surface saturated with agonist. So the observed quantal size of sniffer cell response (about 8 pA) is in much better agreement with release of ATP from synaptic-like vesicles.
The low micromolar level of ATP concentration can be sufficient for activation of P2X purinoreceptors abundantly expressed in the central neurons [1],[42],[44]. The data of electron microscopy and single molecular imaging [39],[56],[57] show both synaptic (mostly at the periphery of synaptic density) and extrasynaptic location of P2X receptors. Interestingly, P2X2 and P2X4 receptors are generally located peri-synaptically (i.e., at the edges of postsynaptic density) [56], which makes them accessible by ATP released from both nerve terminals and extrasynaptic glial release sites. This is consistent with our observation of strong sensitivity of astrocyte-driven sIPSCs to deletion of P2X4 receptors and inhibition of P2X2 receptors (Figures 3, 4, and S7). The distance to the source of ATP would affect the kinetics and amplitude of input mediated by peri-synaptic P2X receptors. Due to diffusion and rapid conversion to ADP, the transient of ATP concentration reaching P2X receptor after vesicular release from the distal (astroglial) site will have less magnitude but will decay longer than ATP transient after release from a close intrasynaptic site (as illustrated in Figure S15). This would lead to smaller quantal amplitude and slower kinetics of purinergic current of the glial origin as compared to neuronally activated synaptic currents. So the diversity in location of the ATP source could be the most plausible explanation for the difference in the amplitude and kinetics of purinergic sEPSCs of neuronal and glial origin observed in the neocortical neurons (Figures 3 and 4). An alternative explanation might be that EPSCs of smaller amplitude and slow kinetics were generated at neuronal synapses at a much longer electrotonic distance. In this case, however, the attenuation factor would increase gradually with distance, leading to smooth single-peaked distributions of EPSC amplitude and decay time due to the presence of large “intermediate” EPSCs. This contrasts with our observation of two peaks in the amplitude and decay time histograms (Figures 3–5). These arguments are supported by results of computer simulation using the model of neurotransmitter spillover [53],[54] adapted for release of ATP from glial and synaptic sites (Figures S14 and S15). Importantly, the key experimental evidence of the astroglial origin of purinergic currents of slower kinetics and smaller quantal size has been provided by their selective inhibition by disruption of vesicular ATP transport in astrocytes (Figure 5).
Although the possibility of nonvesicular gliotransmitter release from astrocytes has been previously reported in several brain regions [4],[7],[15],[58], our results do not show a strong contribution of nonvesicular pathways to the release of ATP in the neocortex. On a contrary, both sniffer cell and biosensor data suggest that vesicular mechanism brings major contribution to activity-dependent ATP release. One should note that incomplete inhibition of ATP release in the biosensor experiments could be explained by incomplete inhibitory action of bafilomycin treatment and dn-SNARE expression on vesicular release rather than notable contribution of nonvesicular mechanisms. It is possible that vesicular and nonvesicular ATP release operates in the neocortex in the different time scale, similarly to spinal cord astrocytes [15], where the fast initial exocytosis of ATP can be followed by slow secondary release of ATP through the pannexin and connexin hemichannels [15]. So it is plausible that vesicular and nonvesicular mechanisms of ATP release coexists in the neocortex, but nonvesicular release was not activated in our experimental context.
It is traditionally considered that physiological astroglial Ca2+ signaling is driven mainly by InsP3-mediated release from intracellular stores [16]. Calcium signals arising from activation of metabotropic receptors are believed to be primarily responsible for control of exocytotic release of gliotransmitters [9],[16],[17]. Although the role of InsP3-induced Ca2+ signaling in astroglial physiology was questioned [27], more recent data strongly support the importance of this pathway for glia-derived modulation of synaptic transmission in the hippocampus [29],[45] and neocortex [28],[30]. We have shown previously that ionotropic NMDA and P2X receptors can mediate a significant fraction of the synaptically driven Ca2+ rise in cortical astrocytes [5],[6],[59]. Our present data suggest that astrocytic NMDA receptors can trigger the release of ATP, acting in parallel to the metabotropic receptors (Figure 2). The capability of ionotropic receptors to control of exocytosis of gliotransmitters (Figure 2) might account for a lesser than expected impact of modulation of astroglial InsP3 signaling on synaptic transmission and plasticity.
Our results shown in Figures 1–4 provide a strong evidence of ATP release by exocytosis and support previous observations of presence of vesicular ATP transporters and synaptic-like vesicles in astrocytes [22],[24],[26]. Observation of vesicular location of VNUT1 (Figures S5 and S6), even with inherent limitations of immunostaining procedure (see Text S1), agrees with previous reports [22],[24],[26] and with our functional data as well. Co-localization of VNUT1 and synaptic vesicle markers (Figure S5), rather small quantal size and fast kinetics of glia-driven purinergic currents (Figures 1–4), suggest that release of ATP from synaptic-like vesicles can make a significant contribution into purinergic gliotransmission in cortical astrocytes.
Until recently, the action of ATP as gliotransmitter was associated mainly with presynaptic adenosine receptors [4],[12],[20]. It was also shown that astrocyte-derived ATP could operate through P2X7 receptors to enhance presynaptic release of glutamate in hypothalamic neurons [10]. Although we observed a moderate presynaptic modulation of inhibitory transmission (Figure 7), our main finding is that glia-driven ATP can directly activate excitatory currents in neurons acting via postsynaptic P2X receptors (Figures 3–5). We showed that activation of neuronal purinoreceptors by astrocyte-driven ATP caused a dramatic change in the inhibitory transmission in neocortex inhibiting postsynaptic and extrasynaptic GABAA receptors (Figures 6 and 7). ATP receptors can act through the Ca2+-dependent phosphorylation by protein kinase C (Figures S11 and S12); the latter mechanism is intrinsic for GABA receptors [43],[60].
The physiological relevance of exocytosis of gliotransmitters was strongly supported by our finding of significant increase in the baseline phasic and tonic inhibitory signaling in the cortical neurons of dn-SNARE mice (Figures 6 and 7). This result suggests that release of gliotransmitters, presumably ATP, can be involved in long-term homeostatic regulation of neuronal GABA receptors by a mechanism that has yet to be investigated. Our evidence of down-regulation of inhibitory synaptic transmission in the neocortex goes in line with observations that enhanced Ca2+ signaling in cortical astrocytes contributed to neuronal excitotoxicity and epilepsy [18],[61].
Our results outline a novel mechanism of action of ATP as a gliotransmitter: in addition to catabolism of ATP to ADP and adenosine and modulation of synaptic transmission via presynaptic purinoreceptors [12], ATP can enhance neuronal excitability by down-regulating the phasic and tonic inhibition acting via postsynaptic P2 receptors. Our present (Figures 6–8) and previous [11] results show that modulation of postsynaptic receptors activated by Ca2+ influx via purinergic P2X receptors can provide an efficient mechanism of regulation of signaling within tripartite synapse. Data on a large contribution of P2X4 receptors to glia-driven modulation of neuronal signaling (Figures 6–8) go in line with a previous observation of a facilitatory role for astroglial release of ATP [12] and P2X4 receptors [43] in the long-term potentiation of synaptic transmission in the hippocampus.
Our findings give further insight into the role of P2X receptors in the CNS. There is growing consensus that, despite a clear evidence of participation of P2X receptors in the excitatory synaptic transmission in several brain areas, they bring notable contribution to slow neuromodulation rather than fast excitation [44]. The above results suggest that P2X receptors can also mediate glia-to-neuron signals, which can be activated in the millisecond time scale and have more long-lasting consequences for neuronal excitability.
Hippocampal astrocytes have been recently reported to decrease the amplitude of mIPSCs via Ca2+-dependent modulation of GABA transport [52] and increase the frequency of mIPSCs via P2Y receptors of inhibitory interneurons, presumably activated by slow release of ATP from astrocytes through connexin hemichannels [62].
Our data suggest that presynaptic ATP receptors can enhance GABA release in the neocortex (Figure 7). Down-regulation of both phasic and tonic postsynaptic GABA receptors by astrocyte-driven ATP (Figures 7 and 8) can act downstream of these pathways and significantly affect their impact on inhibitory transmission. Interplay between post- and presynaptic pathways of glial modulation could have diverse effects on neuronal excitability in different physiological contexts. Apparently, the down-regulation of inhibitory transmission provided by postsynaptic P2X receptors prevails in the neocortical pyramidal neurons, at least in our experimental conditions (Figures 6 and 7).
It becomes evident now that even release of one gliotransmitter, ATP (followed by formation of ADP and adenosine), can activate a variety of pre- and postsynaptic regulatory cascades that can affect synaptic efficacy in opposite ways. Combined with diversity of vesicular and nonvesicular pathways of ATP release and the possibility of release of other gliotransmitters, such as glutamate and D-serine [9],[33],[63], this may confer more complex behavior to tripartite synapse than previously assumed [16],[27].
In summary, our data suggest that Ca2+ elevation in cortical astrocytes, which might occur in response to signals from neurons and/or propagation of glial Ca2+ waves, can trigger exocytosis of ATP from synaptic-like vesicles and activate neuronal P2X receptors that are located at the periphery of synapse. Ca2+ signaling via neuronal ATP receptors can cause phosphorylation-dependent inhibition of postsynaptic GABA receptors acting downstream of astroglial modulation of presynaptic GABA release and GABA uptake. Our results strongly support physiological importance of exocytosis of gliotransmitters, in particular ATP, in communication between astrocytes and neurons and modulation of synaptic efficacy.
Experiments were performed on astrocytes and neurons in the somato-sensory cortex of dn-SNARE transgenic mice [12],[20] and their wild-type (WT) littermates; in some experiments, the P2X4 receptor knockout mice [43] were used. Genotypes of animals were verified by PCR from ear samples. Initial experiments to investigate release of ATP from astrocytes were also performed in transgenic mice expressing enhanced green fluorescent protein (EGFP) under the control of the glial fibrillary acidic protein (GFAP) promoter [34],[59],[64]. Data obtained in the experiments on GFAP–EGFP mice (n = 5–6 for each type of experiments) did not differ significantly from data obtained in the WT mice of the same age. For clarity, all data referred to here as wild-type are reported solely for wild-type littermates of dn-SNARE mice; usage of GFAP–EGFP mice was explicitly stated where appropriate.
Mice (8–12 wk and 9 mo old) were anaesthetized by halothane and then decapitated, in accordance with UK legislation. Brains were removed rapidly after decapitation and placed into ice-cold physiological saline containing (mM) NaCl 130, KCl 3, CaCl2 0.5, MgCl2 2.5, NaH2PO4 1, NaHCO3 25, glucose 15, pH of 7.4 gassed with 95% O2 to 5% CO2. Transverse slices (280–300 µm) were cut at 4°C and then placed in physiological saline containing (mM) NaCl 130, KCl 3, CaCl2 2.5, MgCl2 1, NaH2PO4 1, NaHCO3 22, glucose 15, pH of 7.4, and kept for 1–4 h prior to cell isolation and recording.
Astrocytes were acutely isolated using the modified “vibrating ball” technique [34],[59]. The glass ball (200 µm diameter) was moved slowly some 10–50 µm above the slice surface, while vibrating at 100 Hz (lateral displacements 20–30 µm). Isolation protocol was adjusted to provide a high yield of astroglial cells. This technique preserves the function of membrane proteins and therefore is devoid of many artifacts of enzymatic cell isolation and culturing procedures. In particular, vibro-dissociated astrocytes retain many morphological features (e.g., GFAP–EGFP fluorescence, size, proximal processes) and functional properties (e.g., high potassium conductance, glutamate transporters, Ca2+ signaling) while being completely isolated from neuronal somata and nerve terminals. The composition of external solution for all isolated cell experiments was (mM) 135 NaCl, 2.7 KCl, 2.5 CaCl2, 1 MgCl2, 10 HEPES, 1 NaH2PO4, 15 glucose, pH adjusted with NaOH to 7.3. Astrocytes were identified by their morphology under DIC observation, EGFP fluorescence (astrocytes from dn-SNARE and GFAP-EGFP mice), and functional properties as described previously (see also Table S1) [34],[59].
In the experiments with dn-SNARE mice, administration of Dox [12],[20] has been removed 4 wk prior to electrophysiological studies. According to our observations, about 70% of astrocytes in the layer 2/3 of somatosensory cortex in situ and freshly isolated cortical astrocytes exhibited EGFP fluorescence. The astrocytic identity of EGFP reporter and dn-SNARE-transgene-expressing cells has previously been confirmed for the dn-SNARE line [12],[20]. So one can expect that at least 60%–65% of astrocytes in the somatosensory cortex of dn-SNARE mice express the dn-SNARE domain of synaptobrevin II [20]. In the experiments with freshly isolated astrocytes, only the fluorescent cells have been selected to ensure the impairment of SNARE complex function. In the experiments in the somatosensory cortex of dn-SNARE mice in situ, electrophysiological recordings have been performed in the areas with higher density of EGFP-expressing astrocytes to maximize the putative impact of the loss of SNARE function on neighboring neurons. To increase the probability of a neuron lying within a functional island of synapses [46] controlled by dn-SNARE astrocytes, we recorded from the neurons located in the close proximity to at least two fluorescent cells, as illustrated in Figure S16.
After isolation from a brain slice, cortical astrocytes were incubated with 5 µM Ca2+ indicator Fluo4-AM (or Rhod-2 AM) and 10 µM of photoliable Ca2+ chelator o-nitrophenyl-EGTA-AM (NP-EGTA) for 30 min, re-suspended in a small volume (200–300 µl) of fresh extracellular medium, and placed over cultured HEK293 cells expressing GFP-tagged P2X2 receptors to ATP (HEK293-P2X2, gift from Prof. R. Evans, University of Leicester, UK). To evaluate release of ATP, the transmembrane currents were recorded in the HEK293-P2X2 cells that had an astrocyte lying on their surface; simultaneously, elevation in cytosolic Ca2+ concentration was induced in the astrocytes by UV uncaging.
It has to be noted that contrary to the experiments with astrocytes in culture [21], the spatial density of acutely isolated cells in our experiments was rather low, as shown previously [34]. So we could easily select HEK cells contacting astrocytes with no other cell lying in the immediate vicinity (e.g., as shown in Figures 1 and S2). We performed recordings from the HEK293 cell–astrocyte couples, which were separated from any other cells by at least 15–20 microns of free space to ensure that HEK293-P2X2 cells were activated by ATP released only from contacting astrocyte.
Flash photolysis and fluorescent imaging were performed with the aid of a IX51 inverted microscope and epifluorescent illumination via UPLSAPO60XW/NA1.2 objective (Olympus, Tokyo, Japan). To monitor the intracellular Ca2+ level, astrocytes were constantly illuminated at 480±10 nm using OptoLED light source (Cairn Research, Faversham, UK), and fluorescence was measured at 535±25 nm. Astrocytes from dn-SNARE and GFAP-EGFP mice were loaded with Ca2+ indicator Rhod-2 and illuminated at 530±10 nm; fluorescence was measured at 590±30 nm. The fluorescent images were recorded using Retiga 2000R enhanced CCD camera (QImaging, Canada); exposure time was 35 ms at 2X2 binning. Elevation in the intracellular Ca2+ level was evaluated by a ΔF/F0 ratio averaged over the whole cell image after background subtraction.
For uncaging of intracellular Ca2+, cells were illuminated by a brief pulse (200 ms) of UV light (365±10 nm) emitted by high-power LED NCSU033AT (Nichia, Tokyo, Japan), peak power 500 mW, and estimated power at an objective >150 mW. Illumination was delivered via OptoLED dual-port epifluorescence condenser (Cairn Research, UK). In addition to photolysis, Ca2+ rise was induced in the astrocytes by fast application of agonists of PAR-1 receptors (10 µM TFLLR) or NMDA receptors (20 µM NMDA); these agonists did not cause any response in HEK293-P2X2 cells when applied without astrocytes placed over HEK cells.
Whole-cell voltage clamp recordings from HEK293-P2X2 cells, neurons, and astrocytes were made with patch pipettes (4–5 MΩ for neurons and HEK293-P2X2 cells and 6–8 MΩ for astrocytes) filled with intracellular solution (in mM): 110 KCl, 10 NaCl, 10 HEPES, 5 MgATP, pH 7.35. Intracellular solution for recording in neurons and astrocytes contained 0.2 mM EGTA, and the solution for HEK293-P2X2 cells contained 10 mM EGTA and 1 mM CaCl2. In some experiments (simultaneous recording of GABA-mediated and ATP-mediated synaptic currents), KCl was replaced by KGluconate. Currents were monitored using an AxoPatch200B patch-clamp amplifier (Axon Instruments, USA) filtered at 2 kHz and digitized at 4 kHz. Experiments were controlled by the PCI-6229 data acquisition board (NI, USA) and WinFluor software (Strathclyde University, UK); data were analyzed by self-designed software. Liquid junction potentials were compensated with the patch-clamp amplifier. Series and input resistances were, respectively, 5–7 MΩ and 500–1100 MΩ in the HEK cells and neurons and 8–12 MΩ and 50–150 MΩ; both series and input resistance varied by less than 20% in the cells accepted for analysis. For activation of synaptic inputs, axons originating from layer IV–VI neurons were stimulated with a bipolar coaxial electrode (WPI, USA) placed in the layer V close to the layer IV border, approximately opposite the site of recording; the stimulus duration was 300 µs. For the recording of EPSCs and IPSCs, the stimulus magnitude was set 3–4 times higher than minimal stimulus adjusted to activate the single-axon response in the layer II pyramidal neurons as described in [13]. In order to trigger synaptically driven astroglial Ca2+ transients, the single episode of theta-burst stimulation (HFS) was delivered; the HFS episode consisted of 5 pulses of 100 Hz stimulation, repeated 2 times with 200 ms interval (total 10 pulses per episode).
To monitor the cytoplasmic-free Ca2+concentraton [Ca2+]i, cortical astrocytes were loaded by 40 min incubation with Fura-2AM. For fura-2 excitation, cells were alternately illuminated at wavelengths of 340±5 nm and 380±5 nm using the OptoScan monochromator (Cairn, Faversham, UK). Fluorescent images were recorded using Olympus BX51 microscope equipped with UMPLFL20x/NA0.95 objective and 2× intermediate magnification and Andor iXon885 EMCCD camera; exposure time was 35 ms at 2×2 binning; experiments were controlled by WinFluor software. The [Ca2+]i levels were expressed as F340/F380 ratio averaged over the whole-cell image.
To investigate the Ca2+ signaling activated by PAR-1 receptor agonist, cortical neurons and astrocytes of wild-type and dn-SNARE mice were loaded with 50 µM Fluo-4. The whole-cell voltage-clamp recordings were used to confirm the identification of neurons and astrocytes and verify the lack of changes in the basic electrophysiological properties of the dn-SNARE astrocytes. The Fluo-4 fluorescence signal was excited at 488±10 nm and measured at 530±20 nm; the fluorescent images were recorded and analyzed as described above.
In parts of the experiments (Figure 1, Figure 5, Figure S5, Figure S16), two-photon imaging of neurons and astrocytes was performed using Zeiss LSM-7MP multiphoton microscope coupled with the SpectraPhysics MaiTai pulsing laser; experiments were controlled by ZEN LSM software (Carl Zeiss, Germany). Images were further analyzed offline using ZEN LSM (Carl Zeiss) and ImageJ (NIH) software.
For investigation of vesicular dynamics (Figure 1C), acutely isolated cortical astrocytes were loaded with FM1-43 fluorescent dye (2.5 µM), and FM1-43 fluorescence was excited at 820 nm and observed at 560±20 nm.
For immunolabeling of vesicular transporters, secretory organelles, and astroglial markers, acutely isolated astrocytes were incubated with 0.1 µg/ml of antibodies following antibodies: rabbit polyclonal anti-VNUT1 (T-12), goat polyclonal anti-cathepsin D (N-19, Santa Cruz Biotechnology), mouse monoclonal anti-VGLUT1 (McKA1), mouse monoclonal anti-PSD95 (6G6-1C9), mouse monoclonal anti-GLT-1 (10B7, Abcam), mouse monoclonal anti-NeuN (A60), rabbit monoclonal anti-S100b (EP1576Y, Millipore), goat monoclonal anti-SV2A, mouse monoclonal anti-NG2, and rabbit polyclonal anti-GFAP (Sigma). Prior to cell loading, antibodies were conjugated to the green fluorescent dye DyLight488 (VNUT1, VGLUT1, PSD-95, NeuN, NG2, and GFAP) or red fluorescent dye DyLight594 (VNUT1, SV2A, cathepsin-D, LAMP3, GLT-1, S100β) using Lighting-Link antibody conjugation system (Innova Bioscience, Cambridge, UK) according to the manufacturer's protocol. Antibodies to VNUT1 (recognizing intraluminal epitopes), VGLUT1 (recognizing cytoplasmic epitope), and cathepsin-D (intraluminal epitope) were applied to living astrocytes directly; other antibodies were conjugated with BioPORTER protein delivery reagent (Genlantis, San Diego, CA) 10 min prior to incubation. Immediately after isolation from the neocortical slice, living astrocytes were pre-incubated with 2% of normal bovine serum (Sigma) in the extracellular saline for 30 min to block unspecific binding sites. After them, cells were gently washed two times with clean extracellular saline for 5 min and then incubated with antibodies for 60 min at room temperature. After incubation, cells were washed with laminar flow of extracellular solution in the microscope recording chamber for 30 min prior to image recording.
Fluorescence was excited at 820 nm, GFAP-EGFP and DyLight488 signal was observed at 520±10 nm, and DyLight594 signal was observed at 590±20 nm. The photomultiplier gain of the two-photon microscope was adjusted to avoid saturation in both channels but in the same time did not differ more than 10% between red and green channel. Colocalization analysis of images was carried out using ImageJ software and methods described in [65].
The concentration of ATP within cortical slices (Figure S10) was measured using microelectrode biosensors obtained from Sarissa Biomedical Ltd (Coventry, UK). A detailed description of the properties of ATP biosensors and recording procedure has been published previously in [7],[55],[58]. Briefly, biosensors consisted of ATP metabolizing enzymes immobilized within a matrix on thin (25–50 µM) Pt/Ir wire. This allowed insertion of the sensors into the cortical slice and minimized the influence of a layer of dead surface tissue. The concentration of ATP has been calculated from the difference in the signals of two sensors: a screened ATP sensor and screened null-sensor, possessing the matrix but no enzymes. This allowed us to compensate for release of any nonspecific electro-active interferents. Biosensors show a linear response to increasing concentration of ATP and have a rise time of less than 10 s [55]. Biosensors were calibrated with known concentrations (10 µM) of ATP before the slice was present in the perfusion chamber and after the slice had been removed. This allowed compensation of any reduction in sensitivity during the experiment. Biosensor signals were acquired at 1 kHz with a 1400 CED interface and analyzed using Spike 6.1 software (Cambridge Electronics Design, Cambridge, UK).
All data are presented as mean ± SD, and the statistical significance of difference between data groups was tested by two-population t test, unless indicated otherwise. The spontaneous transmembrane currents recorded in HEK293-P2X2 cells and neurons were analyzed off-line using methods described previously [13],[54]. Briefly, phasic transmembrane currents of an amplitude higher than 2 SD of baseline noise were selected for the initial detection of spontaneous events. Then every single spontaneous event was analyzed within the 140 ms time window, and its amplitude and kinetics were determined by fitting the model curve with single exponential rise and decay phases. As a rule, mean square error of fit amounted to 5%–20% of peak amplitude depending on the background noise. Whenever error of fit exceeded 25%, spontaneous currents were discarded from further analyses.
The amplitude distributions of spontaneous and evoked currents were analyzed with the aid of probability density functions and likelihood maximization techniques [54]; all histograms shown were calculated as probability density functions. The amplitude distributions were fitted with either multiquantal binomial model or bimodal function consisting of two Gaussians with variable peak location, width, and amplitude. The decay time distributions were fitted with bimodal functions. Parameters of models were fit using likelihood maximization routine. For each particular distribution, the fit with quantal or bimodal model was accepted only when it has a confidence level α≤0.05. To monitor and analyze the time course of changes in the amplitude and frequency of spontaneous currents, the amplitude and frequency were averaged over the 1 min time window.
For the basic analysis of the time course of quantal parameters of the evoked IPSCs, the mean quantal content was evaluated as reciprocal square of coefficient of variation [54], and the quantal size was calculated as ratio of mean amplitude to the mean quantal content.
All animal work has been carried out in accordance with UK legislation and "3R" strategy. Research has not involved non-human primates.
|
10.1371/journal.ppat.1003382 | Structural Basis for Rab1 De-AMPylation by the Legionella pneumophila Effector SidD | The covalent attachment of adenosine monophosphate (AMP) to proteins, a process called AMPylation (adenylylation), has recently emerged as a novel theme in microbial pathogenesis. Although several AMPylating enzymes have been characterized, the only known virulence protein with de-AMPylation activity is SidD from the human pathogen Legionella pneumophila. SidD de-AMPylates mammalian Rab1, a small GTPase involved in secretory vesicle transport, thereby targeting the host protein for inactivation. The molecular mechanisms underlying Rab1 recognition and de-AMPylation by SidD are unclear. Here, we report the crystal structure of the catalytic region of SidD at 1.6 Å resolution. The structure reveals a phosphatase-like fold with additional structural elements not present in generic PP2C-type phosphatases. The catalytic pocket contains a binuclear metal-binding site characteristic of hydrolytic metalloenzymes, with strong dependency on magnesium ions. Subsequent docking and molecular dynamics simulations between SidD and Rab1 revealed the interface contacts and the energetic contribution of key residues to the interaction. In conjunction with an extensive structure-based mutational analysis, we provide in vivo and in vitro evidence for a remarkable adaptation of SidD to its host cell target Rab1 which explains how this effector confers specificity to the reaction it catalyses.
| The covalent attachment of adenosine monophosphate (AMP) to proteins, a process called AMPylation (adenylylation), has recently emerged as a novel theme in microbial pathogenesis. While AMPylases from various pathogenic microorganisms have recently been characterized, the only virulence protein with de-AMPylation activity known to date is the Legionella pneumophila effector SidD which catalyzes AMP removal from the host GTPase Rab1. Thus, both AMPylation and de-AMPylation constitute a novel catalytic mechanism to precisely control the function and membrane dynamics of a host Rab GTPase. In spite of this pivotal role, the molecular mechanism of AMP removal and the structural determinants for Rab1 recognition by SidD have remained largely unexplored. Here, we present the crystal structure of the de-adenylylation domain of SidD and reveal the catalytic mechanism of Rab1 de-adenylylation. Surprisingly, the structure of SidD is not related to the other known enzyme with de-AMPylation activity, the Escherichia coli GS-ATase. Instead, the catalitic domain of SidD is remarkably similar to that of the metal-dependent protein phosphatases (PPMs), however with distinctive structural features to distinguish AMPylated Rab1 from similarly modified substrates. Importantly, we provide a model for the SidD-Rab1 complex which sheds light into the specific details of substrate recognition and catalysis by this virulence factor.
| Microbial pathogens have developed a diverse spectrum of mechanisms to manipulate the human host and cause disease. Many bacterial proteins post-translationally modify host factors in order to alter their function. The covalent attachment of adenosine monophsophate (AMP) to threonine or tyrosine side chains within proteins, a process known as AMPylation (adenylylation), was discovered more than 40 years ago in the Escherichia coli protein glutamine synthetase adenylyl transferase (GS-ATase) which regulates the enzyme glutamine synthetase through reversible AMPylation [1]. This post-translational modification recently re-emerged with the discovery of several virulence proteins from Gram-negative bacteria such as Vibrio parahaemolyticus, Histophilus somni, and Legionella pneumophila that AMPylate host proteins [2], [3], [4]. Surprisingly, each of these AMPylators was shown to target host cell GTPases of the Rho or Rab family. VopS from V. parahaemolyticus and IbpA from H. somni covalently modify Rho GTPases such as Cdc42 and Rac1 with AMP, thereby causing a collapse of the host cell actin cytoskeleton resulting in cell rounding [2], [3]. In contrast, SidM (DrrA) from L. penumophila AMPylates host cell Rab GTPases [4] thereby exploiting intracellular vesicle trafficking routes.
The finding that host cell GTPases are a preferred target of bacterial AMPylators can be attributed to the fundamental role these proteins play in all eukaryotic cells. Rab proteins regulate virtually all aspects of vesicle transport [5], [6]. They function as molecular switches that cycle between an inactive GDP-bound state with predominantly cytosolic distribution and an active GTP-bound form that is associated with organelle membranes [7], [8], [9]. Rab activation requires a guanine nucleotide exchange factor (GEF) which promotes replacement of GDP with GTP to enhance the recruitment of downstream ligands, whereas Rab inactivation requires GTPase-activating proteins (GAPs) that stimulate the hydrolysis of GTP to GDP. Inactive GDP-bound Rabs are subsequently extracted from the membrane by a GDP dissociation inhibitor (GDI) and maintained in the cytosol for the next recruitment cycle.
The opportunistic pathogen L. pneumophila, the causative agent of a severe pneumonia known as Legionnaires' disease, subverts membrane dynamics of the host cell by intercepting and modulating Rab1 [10], [11], [12], [13], the regulator of endoplasmic reticulum (ER) to Golgi vesicle transport. The organism infects human alveolar macrophages and multiplies within a specialized compartment called the Legionella-containing vacuole (LCV). To ensure intracellular survival, L. pneumophila uses a specialized translocation machine known as the Dot/Icm type IV secretion system (T4SS) which mediates the delivery of over 200 proteins, termed effectors, from its own cytosol into the host cytoplasm [14]. The effector SidM (DrrA) binds phosphatidylinositol 4-phosphate present in the LCV membrane [15] and exhibits GEF as well as GDF activity towards host cell Rab1 [16], [17], thereby accumulating active GTP-Rab1 on the LCV surface. SidM then AMPylates tyrosine-77 located in the switch II region of Rab1 (Y77Rab1) [4]. The bulky AMP moiety is believed to sterically interfere with the ability of Rab1 to interact with downstream ligands, most importantly GAPs such as the L. pneumophila Rab1GAP LepB, thereby making Rab1 insensitive to inactivation and maximizing its accumulation on the LCV. Notably, activated AMP-Rab1 is gradually removed from the compartment in a process that depends on the L. pneumophila effector protein SidD [18], [19]. SidD is delivered into the host cell later than the AMPylase SidM and catalyzes AMP removal from Rab1, a reaction referred to as de-AMPylation (or de-adenylylation). Once Rab1 has been de-AMPylated, it becomes accessible to binding and inactivation by LepB and subsequent GDI-mediated extraction from LCV membranes [18], [19]. The ability of L. pneumophila to regulate Rab1 membrane cycling through AMPylation and de-AMPylation provides a precedent for how reversible post-translational modification may be used by pathogens to precisely control the function of small Rab GTPases within host cells.
To our knowledge, L. pneumophila SidD and the N-terminal domain (AT-N) of the E. coli GS-ATase are the only known enzymes with de-AMPylation activity, yet the reactions they catalyze differ significantly: While AMP removal performed by AT-N is strictly dependent on the presence of orthophosphate and produces ADP [20], Rab1 de-AMPylation by SidD is phosphate-independent and generates AMP [18], indicative of two fundamentally different mechanisms of de-AMPylation. SidD lacks any obvious sequence homology with the AT-N or other known proteins, although fold recognition analysis of the N-terminal portion of SidD predicted limited resemblance with members of the metal-dependent protein phosphatase (PPM) family. The conserved aspartate residues at position 92 and 110, which are crucial for the activity of other phosphatases, also contribute structurally or chemically to SidD's catalysis [19]. Nonetheless, the molecular mechanism of AMP removal and the structural determinants for Rab1 recognition by SidD have remained largely unexplored.
In this study, we use a multidisciplinary approach to characterize the structural and molecular details that determine substrate recognition and catalysis by SidD. We discover a unique mechanism by which SidD identifies AMPylated Rab1 but not Rho GTPases and that it performs de-AMPylation but not the chemically related de-phosphocholination reaction.
The primary sequence of SidD consisting of 507 amino acids shows no homology to other proteins. Thus, it was unclear which part of SidD possessed de-AMPylation activity and if the protein potentially exhibited additional functions. Guided by secondary structure predictions we created N- or C-terminally truncated variants of SidD, purified them from E. coli, and tested their ability to catalyze removal of radiolabeled [α32P]AMP from Rab1 in vitro (Figure 1A, B). We found that none of the C-terminal fragments and only the longest N-terminal variant spanning amino acid 1 to 379 (SidD1–379) displayed catalytic activity comparable to the full length protein. We also noticed that several of the shorter variants (SidD1–321, SidD1–260, or SidD1–164) were produced either as insoluble or unstable proteins in E. coli (data not shown), suggesting that proper folding of these fragments was compromised by the truncations.
To reduce folding or stability problems that might occur during protein production in E. coli, we employed a mammalian cell-based assay to analyze the de-AMPylation activity of the SidD variants within their host environment. We previously described that production of SidM in transiently transfected COS1 cells causes Golgi fragmentation and subsequent cell rounding and that this phenomenon can be partially repressed by simultaneously producing SidD in the same cell [18], consistent with the fact that SidD's de-AMPylation activity antagonizes SidM's AMPylation activity. When analyzing GFP-tagged SidD variants in this rescue assay we found that none of the truncated proteins was capable of efficiently preventing SidM-induced COS1 cell rounding (Figure 1A), not even SidD1–379, the longest N-terminal fragment that exhibited full Rab1 de-AMPylation activity in vitro (Figure 1B). The failure to rescue cell rounding was not due to the absence or instability of the truncated SidD variants (Figure S1). Rather, we noticed a difference in the intracellular localization pattern of some SidD fragments compared to that of full length GFP-SidD which, as we reported earlier, colocalizes with marker proteins of the Golgi and trans-Golgi network [18]. Upon closer examination, we found that only SidD variants containing the C-terminal 185 residues (amino acid 322 to 507) displayed colocalization with the Golgi marker giantin similar to that of full length SidD (Figure 1C). None of the N-terminal fragments were enriched at the Golgi but instead showed a predominantly cytosolic distribution pattern. Thus, the C-terminal region spanning amino acid 322–507 possessed the ability to target SidD to the Golgi by interacting with a yet unknown factor on this compartment, and failure of GFP-SidD1–379 to properly localize to the correct target organelle may explain the inability of this catalytically active SidD fragment to rescue SidM-mediated cell rounding (Figure 1A). Based on these results we divided SidD into two functional regions: an N-terminal domain with de-AMPylation activity (aa 1–379) and a C-terminal targeting region (aa 322–507).
To further investigate the molecular basis for Rab1 recognition and de-AMPylation by SidD we initiated its structural characterization by X-ray crystallography. We identified a proteolytically resistant N-terminal domain (residues 37–350; SidD-NT) that fell within the domain borders of the largest catalytically active domain discovered above (Figure 1) and that crystallized readily. The structure of SidD-NT assumed an α/β fold formed by two stacked six-stranded antiparallel β-sheets flanked by α-helices (Figure 2A). Pairwise alignment using the DALI server [21] revealed a notable resemblance to metal-dependent protein phosphatases (PPMs), including human PP2Cα [22] and the bacterial PstP [23] which are considered the defining members of this family (Figure S2A). Despite the overall similarity to PPMs, SidD-NT exhibits several major structural differences (Figure 2A,B). First, SidD-NT contains two extra β strands at the N-terminus (β1 and β2) that contribute to extend the central β-sandwich as compared to the classical β-sandwich of the PstP bacterial phosphatase. A second difference resides within the region equivalent to the flap subdomain of PPMs. In most prokaryotic enzymes, the flap subdomain consists of a loop and two helical stretches connecting the last two strands of the central β-sandwich. The length and orientation of the flap region relative to the catalytic pocket is variable between different phosphatases and appears to regulate substrate binding and catalysis [24]. In the case of SidD-NT, the corresponding flap segment (residues 209–236) is completely repositioned by a large hinge bent tangential to the catalytic groove. Furthermore, the N-terminal section of the equivalent flap segment in SidD-NT contains a β strand (β11) that is part of a novel three-stranded antiparallel β-sheet adjacent to the active site. The other two strands (β14 and β15) of this extra β-sheet correspond to an insertion between α6 and β13. Finally, the third main structural difference corresponds to two additional insertions (residues 73–78 and 311–325) that contribute to a noticeable extension of helix α6 and the formation of a two-stranded β-sheet (β4 and β16). The extended α6 and the extra β-sheet form a stalk-like protrusion positioned on one side of the catalytic pocket. In summary, the crystallographic structure of SidD-NT assumes a PPM fold with some conformational rearrangements and the presence of additional subdomains, most of them being grouped around the negatively charged active site (Figure 2C, D).
The active site of SidD-NT is located in a negatively charged cleft between the central β-sheets and comprises a relatively well-preserved binuclear metal center. The first metal (M1) is coordinated by four water molecules and residue D110, whereas the second metal (M2) is hexa-coordinated with the classical octahedral geometry formed by four water molecules, D110, and the main chain carbonyl of G111 (Figure 3A, B). The M1 position is slightly shifted as compared to other PPMs which can be attributed to the incomplete coordination derived from the absence of a highly conserved aspartic acid residue (Figure 3C). In this regard, D192 could accomplish the M1 hexa-coordination but the extended distance would require a conformational closure of the catalytic site. Interestingly, the conserved aspartic acid residue that is missing in the catalytic site of SidD coordinates a third ion (M3) in most bacterial PPMs (Figure 3C). The absence of this aspartic acid residue in SidD precludes a similar M3 coordination and no additional metal binding site is observed. Thus, in contrast to other bacterial PPM phosphatases, SidD appears to lack the capacity of binding a third ion at the equivalent M3 position.
Most PP2C phosphatases require either magnesium (Mg) or manganese (Mn) ions for their activity, with distinct preferences [25]. Quantitative Mg2+ analysis by inductively coupled plasma-optical emission spectrometry (ICP-OES) revealed a stoichiometry of Mg2+ relative to SidD of 1.7 to 1 (data not shown) suggesting that the active site of SidD contains two Mg2+ ions. Furthermore, mutation of D110A in SidD, which directly coordinates both M1 and M2 in the crystallographic structure, resulted in a dramatic reduction in the amount of Mg2+ to nearly negligible values. In this regard, the result from the quantitative ICP-OES analysis for Mg2+ correlates well with the two ions observed in the catalytic pocket of the SidD-NT structure.
The presence of Mg2+ ions within the catalytic pocket of SidD implied an important role of metal ions for the enzyme's activity. Consistent with this, we found that pre-incubation of SidD with the metal chelator ethylenediaminetetraacetic acid (EDTA) efficiently interfered with Rab1 de-AMPylation catalyzed by SidD (Figure 4A). Furthermore, the activity of SidD was fully restored by complementing the reaction with MgCl2 but not by adding other divalent ions such as calcium (Ca2+) or copper (Cu2+) (Figure 4B). A partial recovery of the de-AMPylation activity of EDTA-treated SidD was achieved when the reaction was supplemented with Mn ions (MnCl2). Together, these results indicate a strong preference of SidD for Mg2+ over other divalent ions which is further supported by the observation that SidD regained its maximum de-AMPylation activity at concentrations of 0.8–1.0 mM Mg2+ (Figure 4C) which correspond well with the physiological level of free Mg2+ [26].
Another notable difference between the catalytic site of SidD and other PPMs is the absence of a highly conserved arginine residue equivalent to R33 in PP2C, R17 in MspP, R20 in PstP and R13 in tPphA (Figure S2B), thought to play an important role for binding and neutralizing the negative charge of the phosphate monoester group during the catalysis [27]. The absence of this arginine side chain in SidD might reflect the difference in electrostatics between monoesterase and diesterase reactions, whereby the greater negative charge on the monoester (such as phospho-Tyr) relative to the diester phosphate (such as AMP-Tyr) might explain the necessity of an arginine side chain for stabilization. The pH dependency of PP2Cα in the presence of Mg2+ revealed the existence of two ionizable groups with pKa values of 7.2 and 8.9 [28]. The lower pKa has been interpreted as the binuclear bridging water which, in the form of a hydroxide ion, could attack the phosphorus substrate in a SN2-like mechanism. Using endpoint assays, we examined the pH-dependency of SidD's activity and observed two optimal pH values at ∼7.25 and ∼9.0 which suggest a ionization dependent catalytic mechanism (Figure 4D). Although the identity and protonation state of the amino acid side chains directly involved in the catalytic activity remains to be determined, the lower apparent pKa value of SidD is comparable to that of PP2C [28], consistent with a similar binuclear bridging water acting as the reaction-initiating nucleophile. Indeed, D326 in the crystal structure of SidD, like D282 in PP2C, is appropriately positioned to accept the proton from the bridging water when the hydroxide ion is generated (Figure 3A). According to this interpretation, the de-AMPylation reaction performed by SidD involves a hydrolytic cleavage of the adenylyl-O-tyrosyl linkage, whereas the catalysis of the E. coli GS-ATase, the only other known de-AMPylase, utilizes a phosphorolysis mechanism in which a phosphate ion, not a hydroxyl ion, carries out the nucleophilic attack (Figure S3A, B).
The structure of the active site revealed the presence of two Mg2+ ions coordinated by D92, D110, D326, the main chain carboxyl of G111, and several water molecules (Figure 3B). In addition, the nearby residue D192 could potentially fulfill the coordination of M1. In order to confirm the role of these residues in metal ion coordination, we created SidD mutant proteins in which each of the four aspartate residues was replaced with either alanine or with a similarly charged glutamate (Figure 5A; Figure S4). When assayed for [α32P]AMP removal in vitro we found that even a conservative substitution of aspartate for glutamate attenuated de-AMPylation activity of the SidD mutants considerably (D92E, D192E) or severely (D110E, D326E) (Figure S4). Upon a more drastic substitution of aspartate for alanine no residual activity was detected in three out of four SidD mutants (D92A, D110A, D326A) (Figure 5A). The recombinant mutant proteins displayed no detectable change in stability or solubility (Figure S4), suggesting the absence of major structural disturbances. In fact, we determined the crystallographic structure of SidD(D110A) at 1.9 Å resolution and confirmed the absence of coordinated Mg2+ ions within the catalytic pocket of the mutant protein without noticing any significant effect on its overall fold (Figure S3).
Next, we validated our in vitro de-AMPylation results in two independent mammalian cell-based assays. First, we analyzed the SidD point mutants for their ability to prevent SidM-induced COS1 cell rounding and cytotoxicity (Figure 5B). As expected, wild type GFP-SidD, which showed full de-AMPylation activity in vitro, prevented SidM-induced cytotoxicity in COS1 cells. In contrast, SidD(D92A) and SidD(D110A) were not capable of reducing the percentage of rounded cells that simultaneously produced SidM (Figure 5B), consistent with their lack of de-AMPylation activity in vitro (Figure 5A). SidD(D92E) which possessed residual de-AMPylation activity in vitro prevented morphological changes in twice as many COS1 cells as GFP alone (20% vs. 10%, respectively). Notably, the failure of SidD mutants to efficiently rescue cell rounding was not due to their inability to target to the Golgi compartment (Figure S4). In a second in vivo approach, we determined the effect of aspartate substitutions on the ability of SidD to catalyze de-AMPylation and, thus, removal of Rab1 from LCVs during the infection process (Figure 5C). Consistent with earlier reports [18], [19], L. pneumophila mutants lacking sidD showed a significantly prolonged colocalization with host cell Rab1 four hours post infection compared to LCVs containing wild type bacteria (36% vs 11% Rab1-positive vacuoles), in agreement with the failure of a sidD deletion strain to de-AMPylate Rab1 and to initiate Rab1 inactivation and removal from the LCV membrane by Rab1GAPs and GDI, respectively. The Rab1 removal defect of an L. pneumophila ΔsidD mutant was fully complemented by plasmid-encoded SidD but not by the catalytically inactive protein SidD(D92A). Remarkably, complementation with plasmid-encoded SidD(D92E) fully rescued the phenotype of a ΔsidD mutant, a phenomenon most likely attributable to the residual activity of this enzyme (Figure S4) which may have been further amplified by its overproduction from the high-copy plasmid within L. pneumophila. Taken together, our mutational analysis confirmed that the four aspartate residues at position 92, 110, 192, and 326 are crucial for SidD function both in vivo and in vitro (Figure 5) most likely by properly positioning the two catalytically essential Mg2+ ions inside the active site.
Despite significant efforts we were unsuccessful in obtaining crystals of the complex between SidD and either AMPylated and non-AMPylated Rab1. Furthermore, any attempts to crystallize SidD or the catalytically inactive mutant SidD(D92A) in the presence of AMP analogues such adenosine, adenosine 5′- monophosphate, 5′-(4-Fluorosulfonylbenzoyl)adenosine hydrochloride, adenosine 5′-(α,β-methylene)diphosphate, S-(5′-Adenosyl)-L-homocysteine, or 5′-Tosyladenosine were unsuccessful. Thus, to explore the interaction between both proteins we performed an energy-based rigid-body docking experiment with unmodified Rab1. By using the crystal structure of cacodylate bound to the MspP phosphatase [29] as initial constraint, we found that the docking solution with the Tyr77 hydroxyl O atom closest to the Mg ions (5.4 Å) was able to accommodate the AMP moiety in the same crystallized conformation without steric clashes. We then applied molecular dynamics (MD) in order to refine the SidD-Rab1(AMP) docking model as well as to evaluate its stability. In this regard, the initial docking showed only small fluctuations along the MD simulation indicating a stable SidD-Rab1 interaction (Figure S5A). Similarly, the Mg2+-phosphate interaction at the catalytic site remained constant during the MD simulation (Figure S5B). These results further attested a good structural complementarity between SidD and AMPylated Rab1 with a buried surface of approximately 1,300 Å2 and unrestricted access to the catalytic pocket without the need of large conformational rearrangements (Figure 6A,B). Next, we used in silico alanine scanning on the interfacial residues of SidD to predict relevant hotspots for Rab1 recognition. Interestingly, the residues with higher contribution to the binding free energy are grouped asymmetrically around the catalytic pocket (Figure 6C, and Table S2 in Text S1). Indeed, the average structure from the last nanosecond of the MD shows that F112SidD and Y113SidD form extensive hydrophobic interactions with Y77Rab1 (Figure S5C). Another participating residue is K217SidD which is facing the phosphate group of AMP and may function as proton donor for the leaving phosphate. More peripherally, Y223SidD contributes to the hydrophobic burial of Y109Rab1. Other residues such as E168SidD and D221SidD form hydrogen bonds with R79Rab1 (Figure S5C). Finally, the docking model shows that the AMP moiety is accommodated in a groove adjacent to the catalytic pocket of SidD without being detached from Rab1 (Figure 6A). The adenine base of AMP rests against F74SidD and K88SidD and lacks additional specific interactions whereas the ribose hydroxyl groups interact with R323SidD.
To validate the binding hotspot found in SidD, we mutated several residues that contribute to the interaction with Rab1 and examined the effect on the ability of SidD to remove [α32P]AMP from Rab1 in vitro. In agreement with the interactions described above, the single residue substitutions F112A, Y113A, K217A, Y223A, and Y253E as well as the double exchange F74A/K88A strongly affected the ability of SidD to de-AMPylate Rab1 (Figure 6D) without compromising the overall protein stability or solubility (Figure S5). Only the R323A mutant, designed to disrupt the interaction with the ribose of AMP, had no apparent effect on SidD activity, which may reflect a redundant interaction as a consequence of nearby contacts. Notably, while replacement of Y113 with glutamine strongly reduced Rab1 de-AMPylation by SidD, substitution with the structurally similar phenylalanine had no apparent effect on activity (Figure S4), consistent with phenylalanine but not glutamine being capable of mediating π stacking interactions with Y77Rab1. We also examined additional mutations outside the binding hotspot such as I321S, D271A, or H87A and even the triple mutant T261A/E264A/R281A and, as expected, observed no obvious reduction in SidD activity in vitro (Figure 6C, D) or in vivo (H87A; Figure 5B) which further validated the SidD-Rab1 docking model. Overall, the experimental results are in remarkable agreement with the model complex, with the majority of the mutations at the binding hotspot severely attenuating or preventing SidD-mediated de-AMPylation of Rab1.
The structure of SidD displays a stalk like protrusion on one side of the active site cleft and a binding hotspot on the other side, features that may contribute to recognizing and properly orienting Rab1 in a way that the AMPylated Y77Rab1 is correctly positioned inside the catalytic pocket. We speculated that this topological design might allow SidD to distinguish AMPylated Rab1 from similarly modified substrates such as Rho GTPases. To validate our hypothesis, we generated [32P]AMP-labeled Cdc42 by incubating it with either V. parahaemolyticus VopS, which AMPylates Cdc42 at Y32, or with H. somni IbpA, which AMPylates the neighboring T35 (Figure 7A), and tested the ability of SidD to convert Cdc42 back into the unmodified form. In contrast to [32P]AMP-Y77Rab1 which was efficiently de-AMPylated by SidD, neither [32P]AMP-T35Cdc42 nor [32P]AMP-Y32Cdc42 showed any detectable decrease in the AMPylation level in the presence or absence of SidD (Figure 7B). Thus, SidD did not accept Cdc42 as substrate for de-AMPylation in vitro even if the AMP modification in Cdc42 was located on a tyrosine residue (Y32), as it is the case in AMPylated Rab1 (Y77). Similar results were obtained in an in vivo assay where SidD failed to prevent rounding of COS1 cells transiently producing GFP-tagged VopS (Figure S6) confirming that Rho GTPases AMPylated by VopS cannot be de-AMPylated by SidD.
Finally, we determined if SidD could remove posttranslational modifications other than AMP from Rab1. The L. pneumophila effector AnkX/LegA8 covalently attaches phosphocholine to serine-76 in Rab1 (S76Rab1), the residue located immediately adjacent to Y77Rab1, the target of AMPylation by SidM [30]. Like AMP, phosphocholine is connected to Rab1 via a phosphodiester bond. Its removal requires the L. pneumophila effector Lem3 [31], [32], [33] which we predict assumes a PP2C-like fold similar to SidD (data not shown). Given the similarity of the two removing enzymes and of the chemical bond they hydrolyze we explored whether SidD or Lem3 are capable of catalyzing the other enzyme's reaction. While AMP was efficiently removed by SidD and phosphocholine by Lem3, neither modification was affected by the presence of the opposite enzyme (Figure 7C). Together, these data favor the idea that SidD from L. pneumophila (and probably Lem3 as well) has evolved to exclusively recognize its host cell substrate and to remove only a particular post-translational modification from a specific side chain location.
To our knowledge, SidD from L. pneumophila is the first known microbial effector protein with de-AMPylation activity. Together with the AMPylase SidM it forms an enzymatic cascade that enables the pathogen to post-translationally modify host cell Rab1 in a transient rather than permanent manner. Despite limited sequence homology, the crystal structure of the de-AMPylation domain of SidD revealed a notable similarity to Serine/Threonine phosphatases of the PPM family. However, in addition to the conserved PPM core, SidD-NT exhibits additional structural elements like a repositioned flap domain as part of a new three-stranded antiparallel β-sheet and a stalk-like protrusion, both derived from sequence insertions located around the catalytic site, thus with potential regulatory functions (Figure 2A, B). The finding that SidD is a PPM phosphatase with de-AMPylation activity constitutes a clear example of how L. pneumophila has adapted a common enzymatic fold and mechanism to effectively hydrolyze an unusual substrate. In contrast, AT-N from the E. coli GS-ATase assumes a nucleotidyl transferase fold, indicating that de-AMPylases have developed more than once during microbial evolution. From a chemical perspective, a common feature of enzymes that hydrolyze phosphate monoesters and diesters is the presence of a binuclear metal center. PPM phosphatases share an invariant M1 and M2 whereas the presence of an additional M3 in bacterial homologs is associated with a small flap subdomain adjacent to the catalytic site. The role of this M3 is still unclear, although it has been proposed to modulate the flap orientation and, thus, substrate binding [34]. More recently, the M3 has been associated with the activation of a water molecule that might function as a proton donor for the leaving phosphate [35]. The crystal structure of SidD-NT shows the absence of an absolutely conserved aspartate in the catalytic site that in other PPMs coordinates M1 and M3. This absence not only produces a slight shift in the M1 position but also compromises the coordination of a third ion. Indeed, the quantitative ICP-OES analysis of SidD together with the ion-dependent enzymatic activity assay (Figure 4) support the presence of two Mg2+ ions that are essential for Rab1 de-AMPylation. Collectively, the absence of an M3, the M1 shifted position, the strict requirement of Mg2+ ions for catalysis, and the absence of an arginine side chain to interact with the phosphate group appear to be variations through which L. pneumophila SidD has been converted into an enzyme that de-AMPylates Rab1, capitalizing on the existing PPM active site.
The crystal structure of cacodylate bound to the MspP phosphatase shows a direct interaction with M1 and M2 by bidentate coordination which has been interpreted as a mimicking phospho-substrate intermediate during the catalysis [29]. By using this metal-phosphate coordination as initial constraint in docking AMPylated Rab1 into the catalytic pocket of SidD, we found a remarkable surface complementarity (Figure 6). Subsequent analysis of the docking model by molecular dynamic simulations showed that both the root mean square deviation (RMSD) of the complex with respect to the initial model as well as the distance between the two Mg2+ ions and the phosphate group of AMP experienced only small fluctuations during the simulation process (Figure S5). These observations not only confirm the stability of the docking prediction but, more importantly, evidence a good structural complementarity between SidD and AMPylated Rab1 without the need for large conformational rearrangements to access the catalytic pocket. It should be noted that although our docking model is energetically favorable, the AMP-Tyr side-chain could adopt alternative conformations relative to the crystallized AMPylated Rab1 and that the actual protein complex may experience additional structural rearrangements beyond what has been sampled in our simulations. We also analyzed the interface features of SidD that enable the initial recognition of AMPylated Rab1. Using computational alanine scanning, we identified a hotspot in which the binding energy is largely concentrated on a few amino acids near the catalytic pocket. Indeed, the majority of individual mutations introduced at the binding hotspot severely attenuated or prevented the catalytic activity of SidD in vitro (Figure 6D), which is in remarkable agreement with the qualitative description of the SidD-Rab1 interaction derived from our docking model.
Our structural, computational and mutational analysis revealed the existence of distinctive features in SidD such as the binding hotspot flanking the catalytic site or the stalk-like protrusion that appear to be absent from generic phosphatases. We hypothesized that through this topological design SidD can distinguish AMPylated Rab1 from similarly modified substrates. Accordingly, we demonstrated that AMPylated Rho GTPases were not recognized by SidD under any of the conditions where Rab1 was efficiently de-AMPylated (Figure 7B, Figure S6). Likewise, we found that phosphocholinated Rab1 did not serve as substrate for SidD (Figure 7C) even though its post-translational modification was comparable to AMPylated Tyr77Rab1 with respect to its location (Ser76Rab1) and chemical linkage (phosphodiester bond). The fact that the activity of the de-phosphocholinase Lem3 was similarly restricted from targeting AMPylated Rab1 (Figure 7) suggests that these L. pneumophila effectors have adapted their catalytic activity towards their correct host target thought the acquisition of specific topological determinants.
Based on our domain mapping and cellular localization studies (Figure 1) we predict the existence of a second functional region in SidD that assists in localizing the protein to membranes, more precisely the LCV membrane within L. pneumophila-infected cells or the Golgi compartment within transiently transfected cells. The exact mechanism of membrane targeting of SidD, however, remains unclear. Several L. pneumophila effectors have been shown to specifically interact with phospholipids such as PI(3)P or PI(4)P in order to associate either with the LCV membrane or with other host cell compartments [36]. Indeed, targeting of effectors to a specific cellular compartment constitutes an additional mechanism to enhance substrate specificity. Using protein-lipid overlay assays we were unable to detect binding of SidD to any of the most common phosphoinositides (data not shown), suggesting that membrane targeting of the C-terminal region is mediated by binding to another lipid or proteinaceous host factor. Several attempts to demonstrate a stable association between purified recombinant SidD and Rab1 by pulldown studies failed, indicating that the interaction between both proteins is of transient nature. Nonetheless, it is likely that even a weak interaction with Rab1 is sufficient to retain the majority of SidD molecules in close proximity to the LCV membrane after their translocation by the L. pneumophila T4SS. It is worth mentioning that the prenylation anchor of Rab1 and the C-terminal targeting region of SidD are located on the same side of the complex thus with the potential to simultaneously contact the LCV membrane during catalysis, further strengthening the likelihood of our modeled complex (Figure S7). Future studies should help to reveal the mechanistic details of SidD membrane targeting and substrate detection during host cell infection.
In summary, the study presented here provides an important first look at the structure and catalytic mechanism of SidD and reveals that this L. pneumophila effector differs in many aspects from the E. coli GS-ATase, the only other known de-AMPylase. The finding that SidD is a converted phosphatase equipped with structural elements designed to distinguish AMPylated Rab1 from similar host cell substrates demonstrates the versatility of the phosphatase fold and suggests that it may have served as blueprint for a variety of thus far uncharacterized de-modifying enzymes capable of targeting an array of different post-translational modifications.
Native SidD37–350 was concentrated to 8 mg/ml and used for initial crystal screening. All crystallization conditions were carried out in a sitting drop setup of 0.1 µL protein solution mixed with 0.1 µL of mother liquor. Visible crystals appeared in several comparable conditions after 5 days at 18°C. Further optimization yielded good quality crystals in 1.6–1.8 M NaCl, 0.1 M NaOAc pH 4.8, 20% glycerol using 2 µL sitting drops with equal protein/mother-liquor ratio. The D110A mutant crystallized under the same conditions as the native SidD37–350.
The structure of SidD37–350 was solved by single anomalous dispersion with isomorphous replacement (SIRAS) using a single gadolinium derivative [37]. Gd positions were determined using the SHELX software [38]. The initial electron density map was then calculated with experimental phases derived from the Gd positions with phenix.phaser [39]. A preliminary model was automatically traced by phenix.autobuild and completed by hand in Coot [40]. The model was improved through alternating cycles of manual rebuilding using Coot and refinement using phenix.refine. A final refinement cycle was performed with REFMAC5 [41], [42]. This model was subsequently used for molecular replacement with the high-resolution diffraction data using phenix.phaser. Additional model building and refinement was performed using Coot, phenix.refine and REFMAC5. The final models have good structural geometries with no residues in disallowed regions of the Ramachandran plot. Statistics on data collection and refinement are provided in Table S1 in Text S1. All the molecular representations were prepared with PyMOL (The PyMOL Molecular Graphics System, Schrödinger, LLC) and ChemDraw (PerkinElmer).
We built a structural model of Rab1/SidD complex by rigid-body docking, based on the FFT-based docking program Zdock2.1 [43] and the energy-based pyDock scoring scheme [44]. Details of this procedure are described in Supplemental Materials and Methods.
In order to refine the docking model of the Rab1(AMP)/SidD complex we performed molecular dynamics (MD) simulation in explicit solvent using the force field AMBER parm99 of the AMBER10 package [45], [46]. Details of this procedure are described in Supplemental Materials and Methods.
We performed in silico alanine scanning on the MD refinement of the selected docking model to identify the key residues responsible for the binding process. The MMPBSA.py script in AMBER12 [47] was used to carry out all binding energy calculation using the MM-GBSA method on 200 snapshots extracted from the last 2 ns of the MD trajectory of the selected docking model. Each SidD interface residue was mutated to alanine and then we estimated the binding free energy change (ΔΔG) as the difference between the binding ΔG of the wild type and the mutated complex (van der Waals and electrostatic energy by the MM force field, electrostatic contribution to the solvation free energy by GB method, and nonpolar contribution to the solvation free energy by an empirical model). The conformational entropy contribution to binding was not included here, given the difficulty of computing it for a large protein-protein complex, and the small effect when calculating relative system free energies.
Rab1a (25 µM) was phosphocholinated at room temperature for 4 h in the presence of His-AnkX (0.25 µM) in a buffer containing 20 mM HEPES pH 7.5, 100 mM NaCl, 1 mM CDP-choline, 1 mM MgCl2, and 1 mM ATP. The reaction mixture was then incubated with 60 µl of HisLink beads (Promega) to remove His-AnkX before purification by gel filtration on a HiLoad 16/60 Superdex 75 pg column (GE Healthcare) at 4°C. Fractions containing phosphocholinated Rab1a (Rab1-PC) in either PBS-MM were pooled, concentrated, and stored at −80°C.
For de-phosphocholination, Rab1-PC (10 µM) was incubated for 2 h at room temperature with increasing amounts of the purified Lem3 or SidD in PBS-MM. Immunoblot analysis was used as described above for the de-AMPylation assays using the anti-phosphocholine-specific antibody TEPC-15 (Sigma) to detect phosphocholinated Rab1a.
Immunofluorescence microscopy was performed as previously described [18].
The structural coordinates of SidD37–350 and SidD37–350(D110A) have been deposited in the Protein Data Bank (http://www.rcsb.org.pdb) with the accession codes 4IIK and 4IIP respectively.
|
10.1371/journal.ppat.1002235 | The General Transcriptional Repressor Tup1 Is Required for Dimorphism and Virulence in a Fungal Plant Pathogen | A critical step in the life cycle of many fungal pathogens is the transition between yeast-like growth and the formation of filamentous structures, a process known as dimorphism. This morphological shift, typically triggered by multiple environmental signals, is tightly controlled by complex genetic pathways to ensure successful pathogenic development. In animal pathogenic fungi, one of the best known regulators of dimorphism is the general transcriptional repressor, Tup1. However, the role of Tup1 in fungal dimorphism is completely unknown in plant pathogens. Here we show that Tup1 plays a key role in orchestrating the yeast to hypha transition in the maize pathogen Ustilago maydis. Deletion of the tup1 gene causes a drastic reduction in the mating and filamentation capacity of the fungus, in turn leading to a reduced virulence phenotype. In U. maydis, these processes are controlled by the a and b mating-type loci, whose expression depends on the Prf1 transcription factor. Interestingly, Δtup1 strains show a critical reduction in the expression of prf1 and that of Prf1 target genes at both loci. Moreover, we observed that Tup1 appears to regulate Prf1 activity by controlling the expression of the prf1 transcriptional activators, rop1 and hap2. Additionally, we describe a putative novel prf1 repressor, named Pac2, which seems to be an important target of Tup1 in the control of dimorphism and virulence. Furthermore, we show that Tup1 is required for full pathogenic development since tup1 deletion mutants are unable to complete the sexual cycle. Our findings establish Tup1 as a key factor coordinating dimorphism in the phytopathogen U. maydis and support a conserved role for Tup1 in the control of hypha-specific genes among animal and plant fungal pathogens.
| Fungal plant pathogens cause serious damage to crops with huge social and economic consequences. To cause disease, many such fungi need to change their morphology between a yeast-like, unicellular form and a filamentous state. This change, known as dimorphism, is tightly controlled by complex genetic pathways to ensure successful pathogenic development. In animal pathogens, one of the most important genes controlling dimorphism is Tup1. In plant pathogens, however, the role for this gene is completely unknown. In this work, we describe the role of Tup1 in the dimorphism and virulence of Ustilago maydis, the plant fungal pathogen that causes maize smut disease. We show that mutant U. maydis cells lacking Tup1 are unable to properly change between yeast-like and filamentous forms, thus compromising its virulence. We look at the underlying genetic pathways, and find that Tup1 regulates key genes known to regulate dimorphism. We also show that Tup1 is essential for the production of mature fungal spores, which normally allow the fungus to disperse and infect new plants. Our results show that Tup1 is a key element in the control of both infectious and dispersible fungal forms and supports an evolutionary-conserved role for this gene in the regulation of dimorphism among animal and plant pathogenic fungi.
| Dimorphism, the capacity of certain fungi to change their morphology between yeast-like growth and a filamentous state in response to environmental signals, is frequently associated with the virulence of both animal and plant pathogenic fungi [1]–[6]. This morphological conversion is controlled by several conserved signaling pathways, such as the cyclic AMP-protein kinase A pathway and a mitogen-activated protein (MAP) kinase cascade [4], [6]–[10]. Another well known transcriptional regulator controlling dimorphism is the general transcriptional repressor Tup1, which is conserved from fungi to mammals [11]–[16]. The mechanism of action for Tup1 has been best studied in the yeast Saccharomyces cerevisiae. In this fungus, Tup1p forms a transcriptional co-repressor complex with Ssn6p, a protein that contains tetratricopeptide repeat (TPR) motifs known to mediate protein-protein interactions [17]–[20]. Neither Tup1p nor Ssn6p have direct DNA binding activity and their role in transcription depends on their recruitment to promoters by specific DNA binding proteins [18], [21]. Tup1p repression mechanisms include the interaction with RNA polymerase II holoenzyme components and the alteration of chromatin structure through interaction with histones H3 and H4 and histone deacetylases [22]–[26]. Tup1p controls S. cerevisiae dimorphism in both haploid and diploid strains. Deletions of TUP1 result in reduced haploid invasive growth and reduced diploid pseudohyphal growth, which are considered the filamentous forms of this yeast [11].
Although the role of Tup1 in fungal dimorphism seems conserved, the way it controls this process frequently differs between fungi. The deletion of tup1 from the animal pathogens Candida albicans, Penicillium marneffei and Cryptococcus neoformans give clear examples of this variability. In C. albicans, the homozygous mutant for TUP1 shows a constitutive filamentation phenotype, in contrast to the situation described for S. cerevisiae, and reduced virulence [11]. In P. marneffei, however, tupA is required for the maintenance of its filamentous form, negatively regulating yeast morphogenesis instead of filament formation [12]. In the case of C. neoformans, TUP1 is required for the formation of dikaryotic hyphae due to a mating defect of TUP1 mutant strains, and for virulence [27], [28]. In addition, the molecular mechanisms and genetic pathways by which Tup1 acts in fungal dimorphism are poorly understood in most species [7], [12], [27]–[33]. This role of Tup1 in regulating the dimorphic transition is completely unknown in plant pathogenic fungi, which require different morphogenetic changes to successfully colonize their hosts and cause disease. The only data that might link Tup1 to a role in plant fungal dimorphism are a study into the role of sql1, a gene functionally homologous to S. cerevisiae SSN6, in U. maydis. Here overexpression of truncated forms of Sql1 was shown to induce morphological changes in this fungus [34].
The corn smut fungus Ustilago maydis is a well established model for studying dimorphism and virulence in plant pathogens [35]–[38]. Pathogenic development of this fungus initiates with the transition from yeast-like growth to the formation of polar filaments on the plant leaf surface. Control of this process relies on a tetrapolar mating system consisting of the biallelic a and the multiallelic b loci. Only strains differing in the allelic composition at both loci can successfully form and maintain the infectious filamentous form of the fungus [39]. Locus a encodes the pheromone-receptor system that allow cells from different mating types to detect each other, form conjugation tubes, and fuse [40], [41]. Locus b is then responsible for determining the fate of the resulting dikaryon. This locus encodes a pair of homeodomain transcription factors, bE and bW, that form a compatible heterodimer if proceeding from different alleles, triggering filamentation and pathogenicity [42], [43]. Upon dikaryon filament formation, the hypha tip differentiates to form a specialized structure for plant penetration, known as the appressorium [44], [45]. Once inside the plant, mycelium expansion takes place, leading to the formation of plant tumors. In these tumors, fungal nuclei fuse prior to the separation and rounding up of each hyphal section to form diploid spores. In favorable conditions spores germinate in a meiotic process that forms new haploid cells [46].
The highly conserved cAMP and MAP kinase pathways play a central role in the control of several of the morphological changes required during U. maydis pathogenic development [47]–[51]. Both of these pathways are activated following the recognition of pheromones by receptors of opposite mating types during the yeast to infective hyphae transition, resulting in the transcriptional and post-translational activation of the Prf1 transcription factor [47], [51]–[53]. Once activated, Prf1 promotes the expression of a and b loci genes (for review see [38]) (Figure 1). Thus, U. maydis integrates the inputs that activate both pathways through Prf1 to promote the b-dependent infectious form of the fungus. In the animal pathogen C. albicans, cAMP and MAP kinase pathways induce filamentous growth by promoting the activation of Efg1 and Cph1 transcriptional regulators, respectively, that extend down to hypha-specific target genes [2], [7], [54]–[56]. Control of filamentation in this fungus also requires the transcriptional repression of hypha-specific genes via Tup1, which acts through a third parallel pathway involving Rfg1 and Nrg1 transcriptional regulators [7], [29]–[33]. In U. maydis, as a plant pathogenic fungus, it is unknown whether or not Tup1 plays a role in dimorphism and virulence. Analyzing the function of Tup1 in this plant pathogen could help better understand how it acts within the genetic pathways controlling these processes in different biological contexts.
In this work, we explore the roles of Tup1 during the life cycle of the maize pathogen U. maydis. We demonstrate that tup1 is required for normal mating and filament formation in this fungus and that it controls these processes by transcriptional activation of the Prf1 transcription factor through at least two of its direct regulators. Additionally, we show that tup1 is essential for full pathogenic development, affecting tumor formation and spore production. Our results indicate that Tup1 represents a key factor for the regulation of the pathogenic filamentous and dispersible spore forms of the corn smut fungus U. maydis.
To identify Tup1 homologues in U. maydis we performed a blast search against the MIPS U. maydis database (MUMDB) proteome using Tup1p from the S. cerevisiae database (SGD) as the query sequence. A U. maydis protein sequence, um03280, with an e-value of 9.5e-81 and 66% similarity to S. cerevisiae Tup1p, was retrieved. This sequence, already annotated in MUMDB as Tup1, shows homology to Tup1 proteins from other fungi; including the animal pathogens C. albicans (67% similarity), C. neoformans (73%) and P. marneffei (75%) (all data in Table S1). A sequence alignment of Tup1 proteins from these organisms revealed a number of conserved domains, based on S. cerevisiae: (1) the tup_N domain, located in the N-terminal region, which is known to be required for Tup1p/Ssn6p complex formation; (2) seven WD40 domain repeats in the C-terminal region, that mediate protein-protein interactions and (3) a poorly conserved central region, which possesses histone binding activity in S. cerevisiae [24], [57], [58] (Figure 2, Table S2 and Figure S1).
To test if Tup1 has a role during the U. maydis life cycle, we generated deletion mutants for tup1 in both mating compatible strains, FB1 and FB2, replacing the tup1 open reading frame with the carboxin resistance cassette from pMF1-c [35]. Examination of cell growth and morphology did not reveal any statistically significant differences in either of the tup1 mutants (Figure S2).
Since the U. maydis life cycle is intrinsically linked to its host, we assayed the virulence of tup1 deletion strains. For this purpose, we infected seven day old maize seedlings with compatible mixtures of either wild-type or Δtup1 fungi, and scored tumor formation 14 and 21 days post-infection (dpi). We noticed a considerable reduction in the number of Δtup1 infected plants that developed tumors compared to wild-type infections. Moreover, the size of tumors developed by Δtup1 strains were also considerably reduced (Figure 3A, 3B, and Figure S3). In addition, we observed reduced plant mortality for tup1 mutant infections, with no dead plants observed at 14 dpi and only 11% mortality versus 57% for the wild-type strain 21 dpi. (Figure 3B and Figure S3).
To ascertain whether tup1 mutants are able to complete the sexual cycle we assayed infected plants for the presence of spores 21 dpi. Interestingly, while we found large numbers of spores in wild-type tumors, we could not find spores in tup1 mutant infected plants. Microscopy analysis of the Δtup1 induced tumors revealed that none of the fungal hyphae observed had progressed beyond the rounded cell formation stage that occurs just before spore maturation [46] (Figure 3C).
These results indicate that tup1 is required for full pathogenic development of U. maydis and support a conserved role for tup1 in the virulence of animal and plant fungal pathogens.
During U. maydis plant infection, multiple morphological changes of the fungus are required (for review see [38]). To ascertain which steps of the infectious process are responsible for the decreased amount and size of tumors generated by tup1 mutants, we first determined the extent to which they were able to successfully undergo mating and develop dikaryon filaments. To test this, we co-spotted compatible combinations of tup1 mutants and wild-type strains on PD-Charcoal plates, where the appearance of “fuzzy” white colonies indicates successful mating and the formation of dikaryon filaments. As shown in Figure 4A, crosses between tup1 mutants were unable to form white fuzzy colonies, indicating a recognition or fusion defect between compatible partners, or a post-fusion filamentation defect. Similarly, crosses between tup1 mutants and compatible wild-type strains also showed fuzzy colony formation defects. Filamentation was partially affected when FB1Δtup1 was crossed with wild-type FB2, showing an intermediate phenotype between wild-type and Δtup1 crosses. In contrast, the FB1 and FB2Δtup1 cross showed the same loss of fuzzy colony phenotype as the double mutant cross. In order to check whether the differences observed in FB1Δtup1 and FB2Δtup1 strains could lead to different rates of tumor formation, we performed a plant infection assay using FB1 vs FB2Δtup1 and FB1Δtup1 vs FB2 crosses. As shown in Figure S4A the infection rates of these two strains were similar and slightly different to the rates observed for the cross of both wild type strains.
In addition, we analyzed white fuzzy colony formation in a SG200 background, which is able to form the infective hypha without the necessity of mating with a compatible partner, because of the presence of an active bE1/bW2 heterodimer and a constitutively expressed mfa2 gene [59]. Significantly, SG200Δtup1 did not generate fuzzy colonies on charcoal plates, suggesting a post-fusion role for tup1 (Figure 4B). In order to quantify the phenotype, we performed a filamentation assay by co-spotting SG200CFP [60] and SG200YFPΔtup1 labeled strains on PD-charcoal plates. After fuzzy colony formation, colony samples were used for the quantification of filaments formed by each strain. As shown in Figure 4C, 80% of the filaments corresponded to the wild-type strain, while only 20% belonged to the mutant. Maize infection experiments with tup1 mutants in the SG200 background revealed similar virulence defects to what we had observed in FB1 and FB2 backgrounds (Figure S5A and S5B). Insertion of a single copy of tup1 under the control of the constitutive otef promoter in the ip locus [34] of SG200Δtup1, restored its filamentation and pathogenic capacity, indicating successful complementation (Figure 4B, Figure S5A and S5B). Moreover in the case of the FBD11 diploid strain, which also do not need to mate with a compatible partner to cause virulence, the heterozygous mutant FBD11Δtup1/tup1 and the homozygous FBD11Δtup1/Δtup1 were almost completely avirulent in leaf infection experiments (Figure S4B and S4C). Because of the reduced infection capacity of the FBD11 wild-type strain, we also performed flower infections (where we usually observe bigger tumors) with these strains to better reflect the differences between them. This experiment revealed big tumors in the wild-type strain, medium tumors in the heterozygous and small tumors in the homozygous mutant strains (Figure S4D and S4E).
These results point to a post-fusion filamentation defect as a plausible reason for the impaired pathogenicity of Δtup1 strains. However, it has been reported that mating or filamentation defects on PD-Charcoal plates are not always conserved on the plant leaf surface [61]. To check this, we co-infected 7 day old maize seedlings with the labeled strains, SG200CFP and SG200YFPΔtup1 and quantified filament formation on the leaf surface. As shown in Figure 4C (on plant columns), the filamentation defect seen on charcoal containing media was also apparent on the leaf surface, with only around 5% of the filaments formed corresponding to the mutant strain.
Finally, to check whether tup1 could also be implicated in other morphological changes required during the U. maydis infection process, we checked for appressoria formation and the presence of clamp-like cells during mycelium expansion in tup1 mutant strains. We observed that both of these structures were formed in the deletion mutants for tup1 (Figure 5A and 5B), although at lower frequency than the wild type, which is very likely a consequence of the filament formation defect showed by these mutants. The frequency of appressoria formation by SG200YFPΔtup1 was reduced to a similar extent as filament formation (Figure S5C), and mycelium expansion was reduced in Δtup1 infected plants at 2 dpi (Figure 5C). These results, together with the capacity, albeit reduced, of tup1 mutants to induce tumors in maize, suggest that those tup1 mutant cells that overcome the filamentation defect are then able to undergo the morphological changes required for plant penetration and expansion. Thus, the role of tup1 in the morphological changes that occur during U. maydis infection seems to be specific to the yeast-to-hypha transition.
As tup1 mutants are unable to form dikaryotic hyphae at wild-type levels, we wondered whether tup1 regulates genes downstream of the b locus, thus compromising the fungal dimorphic transition in tup1 mutants. To this end, we used the AB33 strain in which expression of a compatible bE1/bW2 heterodimer is under the control of the nar inducible promoter [62]. When this strain is grown in inducing conditions it forms a b-dependent filament. We found that deletion of tup1 in this background did not affect its filamentation capacity (Figure 6A; see Figure S6 for quantification). This result suggests that Tup1 is affecting processes upstream of the b locus or, alternatively, is acting on a parallel pathway regulating filamentation. To discern between these two possibilities, we extracted total RNA from SG200 and SG200Δtup1 fungi grown on charcoal-containing media for 48 hours and quantified the expression of bE and bW by Northern blot. We observed a strong decrease in both gene transcripts in SG200Δtup1 indicating that tup1 is required for the normal expression of b loci genes (Figure 6B lanes 5 and 6).
To test if constitutive b expression could rescue the filamentation and virulence phenotypes of tup1 mutants, we took advantage of the HA103 strain, which harbors a compatible bE1/bW2 heterodimer under the control of constitutive promoters [52]. Deletion of tup1 in HA103 did not produce the filamentation and virulence defects described for the SG200 background (Figure 6C, 6D and Figure S7), indicating that constitutive b expression partially rescues these phenotypes. To better understand the effect of b expression on the tup1 mutant virulence phenotype, we used the HA103 parental strain, CL13 [59], which carries compatible bE1 and bW2 genes under the control of their own promoter and lacks the constitutively-expressed mfa2 gene present in SG200. Deletion of tup1 from CL13 led to a 90% reduction in maize tumor formation (Figure 6D and Figure S7), revealing an even clearer b-genes dependent rescue of tup1 mutant phenotypes. Interestingly, the expression level of the b genes correlated with the phenotype of the wild-type and Δtup1 strains (Figure 6B). Moreover, when we focused on the CL13 and SG200 backgrounds, we observed that the SG200Δtup1 strain had a b expression level, filamentation and virulence capacity comparable to the wild-type CL13 strain (Figure 6, Figure S7 and Figure S8). Thus, the effect of deleting tup1 from SG200 seems to be equivalent to removing its constitutive expression of mfa2, which would suggest a putative role for the pheromone responsive pathways in tup1 mutant phenotypes.
In our earlier experiment we bypassed the requirement for cell fusion by using the SG200 strain to identify a post-fusion requirement for tup1 in U.maydis filamentation. However, this experiment does not exclude a role for tup1 in mating between compatible strains as well, especially since both a and b loci genes are in the same position of the genetic pathway that controls the dimorphic transition. Moreover, as commented above, the similarity between SG200Δtup1 and CL13 strains may reflect a role for tup1 in the transduction of the pheromone signal.
To test this possibility, we extracted total RNA from a FB1Δtup1 vs FB2Δtup1 cross grown on charcoal-containing media for 24 hours and compared mfa1 and bE1 expression with a wild-type strains cross by Northern blot. In the wild-type cross, as a result of the recognition of pheromones by receptors of opposite mating types, activation of pheromone responsive pathways takes places, which is reflected in the expression of genes at both a and b loci. In the case of the tup1 mutant cross, however, we observed reduced mfa1 and bE1 expression (Figure 7A), indicating that tup1 is necessary for wild-type expression of these genes. Accordingly, FB1Δtup1 and FB2Δtup1 strains drastically reduced conjugation hyphae formation upon stimulation with synthetic pheromones of the opposite mating type (Figure 7B and 7C).
Thus, tup1 is required for signal transduction upon stimulation with pheromone and expression of genes at both a and b loci, which is reflected in the observed pre and post-fusion defects of Δtup1 cells.
The expression of a and b loci genes is controlled by the cAMP and MAP kinase pathways through their common effector Prf1. To situate tup1 within this genetic context, we used the FB1Pcrg1:fuz7DD strain, which harbors a constitutively active allele of fuz7 MAPKK under the control of the arabinose inducible promoter crg1 [51] (see Figure 1 for components of the MAP kinase pathway). Upon induction, this strain promotes the expression of a and b loci genes via the Prf1 transcription factor. After deleting tup1 from this strain, we checked for a and b loci gene expression under inducing conditions. As expected, increased transcription for genes at both loci was observed in the wild-type strain; however, this was not the case for the tup1 mutant, indicating that Tup1 regulates a and b gene expression downstream of Fuz7 MAPK kinase (Figure 8A). Since Tup1 is involved in regulating the expression of genes related to glucose metabolism, the expression level of Fuz7 under the control of the crg1 promoter was also examined. No difference in fuz7DD expression was observed between the wild-type and the Δtup1 strains (Figure 8A).
Apart from its effect on the expression of the previously mentioned genes, induction of the fuz7DD allele, promotes conjugation tube formation through a Prf1 independent pathway that also requires the action of Kpp2 MAP kinase [51]. Thus, we wondered whether the induction of fuz7DD in the tup1 deletion strain could also induce conjugation tube formation. As shown in Figure 8B, tup1 mutants in this background were able to form conjugation hyphae at similar levels to wild-type fungi in inducing conditions (Figure S9 for quantification). This result makes it unlikely that Tup1 is regulating conjugation tube formation downstream of the MAP kinase cascade and, at the same time, strongly suggest that tup1 regulates mating-type genes downstream of Kpp2 MAP kinase.
We have shown that tup1 seems to regulate the expression level of a and b loci genes acting downstream of the MAP kinase cascade. Since the Prf1 transcription factor is the genetic element connecting the MAP kinase cascade and the mating-type genes, we measured prf1 expression level in a FB1Pcrg1:fuz7DD background under inducing conditions. The removal of tup1 prevented the increase in prf1 expression (Figure 8A), indicating that tup1 is required for prf1 expression upon MAP kinase cascade induction. Moreover, the filamentation defects on charcoal-containing media as well as on the plant surface were rescued with the constitutive expression of prf1 (Figure 8C and 8D). These results strongly suggest that tup1 affects mating and b-dependent filament formation through control of prf1 transcription factor expression level rather than by controlling the expression of a and b loci genes directly.
As constitutive bE/bW expression did not fully complement Δtup1 phenotypes, we were interested in identifying other Tup1 regulated genes, that might also have roles in the dimorphic transition and virulence in U. maydis. For this purpose we performed a microarray analysis with custom Affimetrix array (MPIUstilagoA), covering 5823 of the 6787 predicted U. maydis genes, and compared the gene expression of SG200 and SG200Δtup1 strains grown on MM-charcoal array plates for 48 hours (see Methods). We identified a total of 115 genes (around 2 % of the covered genes) with altered expression in the tup1 mutant strain. Of these, 59 were upregulated and 56 downregulated. Within this list appear the bE and bW genes together with 34 genes that have also been described as b regulated genes [63], and 17 genes described as pheromone regulated [64] (Table S3). Thus, around 36% of the genes directly or indirectly regulated by tup1 are also regulated upon bE/bW heterodimer and/or pheromone/fuz7DD induction, in agreement with our earlier results and supporting the quality of our dataset. Additionally, in order to experimentally validate our microarray data, the differential expression of some of the genes was confirmed by Northern blot analysis (Figure 9A).
All the 115 Tup1-regulated genes were classified in functional categories using the Blast2Go tool [65]. Enrichment analysis of genes up-regulated by the deletion of tup1 did not reveal a significant over-representation in any of the GO categories (Table S4). Of the genes down-regulated upon tup1 deletion our analysis revealed a significant over-representation in two GO categories: “Carbohydrate metabolic process” (GO:0005975; 8 genes) and “Antioxidant activity” (GO:0016209; 3 genes) (Table S4). 4 of the 8 genes belonging to the first category were also b-regulated genes, with two of them defined as strictly b-dependent (Table S3). The second category comprises proteins involved in the inhibition of dioxygen or peroxide-induced reactions and could be related to pathogenicity since production of these compounds is a well-characterized plant defense mechanism [66], [67], and H2O2 detoxification is required for U. maydis virulence [68].
Interestingly, several tup1-regulated genes are associated with processes that could be related to the morphological switch from yeast-like to filamentous growth. Almost 10% of these genes are potentially involved in cell wall synthesis or modification, revealing that the altered yeast-to-hypha transition, promoted by deletion of tup1, results in a different cell wall composition.
Significantly, we found that rop1, that encodes a direct activator of Prf1, was down-regulated in the tup1 deletion strain (Table S3). This suggests an indirect role for tup1 in controlling prf1 expression. Rop1 has been described as being required for the mating of compatible strains on charcoal containing media, with a post-fusion role, due to the inability of SG200Δrop1 to form white fuzzy colonies on charcoal plates. It is essential for conjugation tube formation upon pheromone stimulation, and for expression of pheromone-responsive genes [61]. These phenotypes clearly resemble the situation described for tup1 mutants; however, rop1 mutants are fully pathogenic, with no mating or filamentation defects described on the plant leaf surface [61].
In addition to rop1, we identified an interesting candidate gene, um15096, that could be related to the tup1 mutant phenotypes. In Schizosaccharomyces pombe, a homologue of um15096, named pac2, has been shown to be a repressor of ste11 (the putative functional homologue of prf1) [69]. Interestingly, um15096/pac2, herein referred to as pac2, appeared over-expressed in the tup1 deletion strain. To check whether this putative prf1 repressor could also be playing a role during filamentation and pathogenic development, we over-expressed pac2 by integrating an extra copy of the gene under the control of the otef constitutive promoter in the ip locus of the SG200 strain. Filament formation of SG200pac2con was reduced on charcoal containing media (Figure 9B) and, more importantly, pathogenicity was reduced to levels comparable to tup1 mutants (Figure 9C). The fact that pac2 is over-expressed in tup1 mutants together with the observation that ectopic pac2 expression decreases filamentation and virulence in the wild-type strain, strongly suggest that pac2 expression contributes to the filament formation and pathogenic defects of Δtup1 cells. Consistent with this, the deletion of pac2 from SG200 resulted in wild-type filamentation and infection rates (Figure 9C). When prf1 expression was induced by constitutively activating the MAPK pathway at Fuz7 level, overexpression of pac2 abolished its expression, while deletion of pac2 did not apparently affect it. Similar results were observed for mfa1 and bE1 genes. The double Δtup1Δpac2 mutant showed the same level of expression as the single Δtup1 strain (Figure 9D); probably as consequence of the regulation of rop1 via Tup1. Surprisingly, pac2 deletion, weakly restored the filamentation and infection defects shown by SG200Δtup1 strain (Figure 9B and 9C), indicating that Pac2 contributes to tup1 deletion strain phenotypes.
In summary, our microarray data reveal that at least 36% of the genes whose expression is affected by deletion of tup1 seems to be a consequence of tup1-dependent regulation of a and b loci genes through prf1. Moreover, the role of Tup1 in the control of prf1 expression could be explained by the altered expression of rop1 and pac2 observed in the tup1 mutant strain.
As Tup1 seems to have an indirect effect on prf1 transcription level through Rop1 and, putatively, Pac2, we wondered whether the expression of other known prf1 regulators could be affected in tup1 deletion strains. Apart from Rop1, prf1 is known to be directly regulated by Hap2 [70] and indirectly through the MAP kinase Crk1 [71]. Northern blot assays of SG200 and SG200Δtup1 grown on charcoal media showed that the expression level of crk1 was unaffected in tup1 deleted strain. In contrast, the levels of rop1 and hap2 were reduced in comparison to the wild-type strain (Figure 10). However, as Crk1 acts on prf1 indirectly, and since it has been previously reported that the effect of Crk1 on prf1 depends on the prf1 promoter UAS [71], we tested whether Tup1 could regulate prf1 via its UAS. For this purpose, we used the HA232 strain, which harbors a GFP reporter gene under the control of the prf1 promoter UAS (see [53] for details). In this strain, GFP is strongly expressed when grown on glucose-containing media, while its expression is reduced on a maltose containing media [53]. As is shown in Figure S10, the expression levels of the reporter gene were indistinguishable in Δtup1 mutants from the wild-type in all the conditions tested. This indicates that Tup1 is unlikely to act via the prf1 promoter UAS, in contrast to Crk1. Thus, the effect of Tup1 on prf1 expression seems to be mediated via Rop1 and Hap2 but not through the Crk1 pathway.
To sum up, although other factors may be implicated in tup1 mutant phenotypes, Tup1 seems to control the dimorphic transition and participates in the virulence program of U. maydis by indirectly regulating prf1 expression via altered rop1 and hap2 expression levels, and possibly also through pac2, which would lead to a down-regulation of prf1-dependent expression of a and b loci genes and their related phenotypes.
In the basidiomycete phytopathogen U. maydis, the switch from non-infective yeast-like growth to an infective filament formation occurs in response to different environmental cues, and is tightly controlled by complex genetic pathways in order to ensure the coordination and timing of the different processes associated with dimorphism. In this work, we have shown that the highly conserved general transcriptional repressor Tup1 plays a central role in controlling the proper expression of the genes implicated in the genetic control of mating, filamentation, and pathogenic development of this corn smut fungus.
Tup1 has been shown to be important during growth of vegetative cells in other fungi such as S. cerevisiae, C. neoformans or P. marneffei [12], [27], [72]. In the case of Ustilago maydis, differences could be observed in the tup1 mutants, although none of these were statistically significant. Interestingly the normal growth of Δtup1 strains contrasts with the poor growth capacity described for U. maydis strains harboring a partial deletion of sql1, the functional homolog to S. cerevisiae SSN6. However because these strains were not stable, the role of Sql1 could not be completely analyzed [34]. Thus a comparison between Tup1 and Sql1 of their growth capacity on U. maydis vegetative cells cannot be properly established. In other fungi, single deletions of tup1 and ssn6 have been reported to result in different phenotypes [73]-[76]. For example, the deletion of SSN6 but not of TUP1 homologues is lethal in S. pombe [75] and Aspergillus nidulans [76]. Moreover, Tup1 and Ssn6 have been shown to regulate different set of genes [74] and to form independent complexes in C. albicans [77].
A central question in this study was whether tup1 is involved in the infectious process of plant pathogenic fungi. We have observed that infections with Δtup1 cells lead to a reduction in tumor formation, plant death, and a failure of spore formation, indicating that Tup1 is required for full pathogenic development in U. maydis, and making tup1 mutants unlikely to cause damage in natural environments. Thus, tup1 seems to play a conserved role in virulence of animal and plant fungal pathogens.
The next key question was to try to understand the mechanism by which tup1 is required for normal tumor formation. Our results suggest that the virulence phenotype of Δtup1 cells has two main causes: (i) a recognition problem between compatible partners, due to the inability of tup1 mutants to form conjugation hyphae upon pheromone stimulation, and (ii) a filamentation defect, due to the inability of SG200 to form filaments at wild-type levels both on PD-charcoal plates and on the plant leaf surface. Additionally, the fact that the differences on conjugation hypha formation between FB1Δtup1 and FB2Δtup1 strains, though not statistically significant, together with the differential filamentation showed by crosses of these strains with their respective compatible wild-type strains on charcoal plates, suggest also a role for Tup1 in cell fusion, at least in the FB2 background. These defects result in tup1 mutants being unable to properly undergo dimorphic transition. These findings suggest that the impaired pathogenicity of tup1 mutant animal and plant fungi may also depend on a conserved role in the yeast-to-hypha transition.
Consistent with the conjugation and filamentation phenotypes of tup1 mutants, the expression of a and b loci mating-type genes was reduced in tup1 deletion strains, most likely as a consequence of Tup1-dependent regulation of the prf1 transcription factor. Microarray analysis of SG200Δtup1 during filamentation on charcoal media revealed a number of mis-regulated genes whose expression was also affected upon b-compatible heterodimer and/or pheromone/fuz7DD induction, including the b locus genes themselves, supporting the proposed role for tup1 during U. maydis mating and dikaryotic filament formation. On the other hand, in our microarray analysis we did not detect tup1-dependent changes in gene expression for any of the b-dependent genes previously described as being essential for pathogenicity [60], [63], [78], which is consistent with the ability, albeit reduced, of tup1 mutants to induce tumors in maize.
Interestingly, the main effector that links tup1 to the control of dimorphism seems to be conserved between U. maydis and C. albicans. In contrast, the genetic pathways by which tup1 acts on filamentation seem to differ, depending on the genetic control of hypha-specific genes in each organism. In C. albicans, Tup1 is proposed to control filamentous growth through the repression of hypha-specific genes by forming complexes with the transcriptional repressors Rfg1 and Nrg1, rather than affecting the elements in the Cph1-mediated MAPK and Efg1-mediated cAMP pathways [7], [54]–[56]. Moreover, expression analysis of filament-specific genes in Δcph1/Δcph1, Δefg1/Δefg1 and Δtup1/Δtup1 strains revealed common and divergent target genes [7]. Thus, Tup1 integrates into the network system proposed for the control of filament-specific genes in this fungus [7], [10]. On the other hand, in U. maydis, Tup1 controls infective filament-specific gene expression via a central regulatory, the Prf1 transcription factor, which is transcriptionally and post-translationally regulated by the cAMP and MAPK pathways [47], [51]–[53]. Interestingly, U maydis Prf1 is a High Mobility Group (HMG) transcription factor, similar to C. albicans Rfg1. Thus, an analogous mechanism, implicating a Tup1-Prf1 complex, could explain the roles of Tup1 in the regulation of hypha specific genes in U. maydis. Moreover, in S. cerevisiae, a complex between Tup1p and the HMG-transcription factor Rox1p has also been proposed [19], [79]-[81]. S. cerevisiae ROX1, whose deletion can be complemented by C. albicans RFG1 [33], is known to control hypoxic gene expression in a TUP1 dependent manner [19], [79]–[81]. Additionally, the deletion of TUP1 increases the expression of ROX1 [82], [83], but Rox1p itself is also able to regulate its own expression [83]. In aerobic conditions these observations can be explained by the proposed Tup1p-Ssn6p-Rox1p complex which would regulate ROX1 expression and Rox1p-dependent hypoxic gene expression. In anaerobic conditions, however, the regulation of ROX1 expression seems to implicate an anaerobic repressor that requires Tup1p for its function [83]. Similarly, in U. maydis, the expression of prf1 is dependent on Tup1 and prf1 is also self-regulated [52]. However, when we analyzed the effect of Tup1 on prf1 expression level more deeply, we observed that at least two direct activators of Prf1 were also down-regulated upon tup1 deletion, rop1 and hap2. This finding, although not excluding a putative Tup1-Prf1 complex, points to an indirect effect of Tup1 on the expression of prf1 and its regulated genes. Rop1 is required for pheromone response and for fuzzy colony formation on charcoal-containing plates, but is dispensable for mating and filamentation on the plant leaf surface. In the case of hap2, it is known to be essential for the pheromone response and has also an effect on the filamentation capacity of SG200 that seem to be conserved on planta. Thus, we propose that the effect of Tup1 on prf1 is the sum of the effects of Tup1 in both rop1 and hap2 on artificial media, while only the effect on hap2 would be responsible for the on planta phenotypes. The drastic effect of tup1 deletion on prf1 expression levels on charcoal plates may be diminished on the plant leaf surface as rop1 is dispensable in this situation.
In this work, we have also described a new gene, pac2, which is likely to be playing a role in the tup1 mutant virulence phenotype, since its over-expression causes a decrease in the pathogenic capacity of U. maydis SG200 strain and its expression is increased in the SG200Δtup1 strain. Since the homologue of this gene in S. pombe is a repressor of ste11 [69], the putative functional homologue of prf1, we analyzed the relationship between Pac2 and Prf1 in U. maydis. We found that over-expression of pac2 in a FB1Pcrg1:fuz7DD strain abolished the prf1 expression observed in the wild type strain establishing Pac2 as a repressor of Prf1. Accordingly, the deletion of pac2 in a SG200Δtup1 strain partially restored its filamentation and virulence defects. However, the double Δtup1Δpac2 mutant in the FB1Pcrg1:fuz7DD background shows the same prf1 expression level than the single Δtup1 strain, probably because of Tup1 control of rop1 and hap2. Nevertheless since prf1 regulation on charcoal plates or during virulence integrates several imputs besides the MAPK pathway the relationship between pac2 and prf1 in the regulation of filamentation and pathogenicity cannot be fully established. Thus, the final role of tup1 in U. maydis virulence is also likely to be linked to its control of hap2 and pac2 mRNA levels (Figure 11).
Surprisingly, although Tup1 is described as a general transcriptional repressor, the deletion of tup1 from U. maydis leads to the down-regulation of the genes that control the dimorphic transition, suggesting an activator role for tup1 in controlling them. On the other hand, determining how Pac2 controls prf1 gene expression would help to determine the role of tup1 as an activator and/or repressor during dimorphism. The way Tup1 seems to control the expression of the prf1 transcription factor, through hap2 and rop1 and, putatively, pac2, clearly reflects the complex genetic regulation that prf1-related processes require.
Similarly, the number of genes that we found to be up- or down-regulated following tup1 deletion when cultured on charcoal-containing media was equivalent. Thus, under the conditions tested, the loss of tup1 causes a similar effect on both the de-repression and repression of genes. Although this could reflect indirect changes in genes expression resulting from the repression of Tup1-gene targets, it is nevertheless an intriguing observation. Regarding an activating role for Tup1, previous studies have also shown that Tup1 can behave as an activator as well as a repressor of the same target gene in different conditions [84] or different genetic backgrounds [85] in S. cerevisiae.
Finally, we have shown that tup1 seems to be required for spore production inside maize tumors. Roles for Tup1 in sporulation have been previously reported in other fungi. In S. cerevisiae, the sporulation-specific genes DIT1 and DIT2, which are required for spore wall formation, are regulated by Tup1p [86]; in Neurospora crassa, mutants for rco-1, the homologue of TUP1, are aconidial [87]; and in C. neoformans, tup1 deletion considerably reduces spore production [27].
In summary, our work provides new insights into the complex regulatory circuits for sexual and pathogenic development of U. maydis. We have identified for the first time a requirement for tup1 at several steps of the life cycle of a pathogenic plant fungus, including in the genetic pathways controlling dimorphism and virulence. Our findings contribute to a better understanding of the role of this general transcriptional repressor in pathogenic fungi and of the precise genetic control that these pathogenesis-related processes require. We consider that the roles and mechanisms of action described for U. maydis tup1 in this work will also be extremely valuable for studying the roles of tup1 in the transcriptional regulation of morphogenetic processes in other organisms.
Escherichia coli DH5α was used for cloning purposes. Growth conditions for E. coli [88] and U. maydis [42], [89] and the quantification of appressoria formation on the plant leaf surface [60] have been described previously. Quantification of filaments was performed as for the appressoria. For studies of growth rates and morphology, cells were grown on YEPSL liquid media for 12 hours, then diluted in the same media to an OD600 of 0.05 and grown until an OD600 of 0.8-1. Exponential growth cultures were examined under the microscope and transferred to solid plates for colony morphology studies. Growth rates on liquid media were determined by counting cells at different time-points. For charcoal mating and filamentation assays, cells were grown on YEPSL until exponential phase, washed twice with water, spotted onto PD-charcoal plates and grown for 24–48 hours at 25°C. For charcoal-grown cells used for RNA extractions, cells were spread out on charcoal plates at a concentration of OD600 = 0.1 per cm2. For DNA array charcoal media see below. U. maydis strains relevant to this study are listed in Table S5. Induction of nar promoter in AB33 [62] and crg promoter in FB1Pcrg1:fuz7DD [51] strains, and their derivatives, were done as previously described. Mating assays were performed as previously described in [90]. Pheromone stimulation was performed following the protocol of [51]. For pathogenicity assays, U. maydis strains were grown to exponential phase and concentrated to an OD600 of 3, washed twice in water, and injected into 7 days old maize (Zea mays) seedlings (Early Golden Bantam). Tumor formation was quantified 14 to 21 days post infection. Data are expressed as means ±SD of triplicate samples. Statistical significance was assessed using Statistical Calculators (http://www.graphpad.com/quickcalcs/index.cfm) and considered significant if p values were <0.05.
Molecular biology techniques were used as described by [88]. U. maydis DNA isolation and transformation procedures were carried out following the protocol of [91]. Deletion constructs were generated according to [36]. To generate single deletion U. maydis mutants for tup1 (Um03280), pac2 (Um15096) and um04807 genes, fragments of the 5′ and 3′ flanks of their open reading frames were generated by PCR on U. maydis FB1 genomic DNA with the following primer combinations: UmTUP1KO5-1/UmTUP1KO5-2 and UmTUP1KO3-1/UmTUP1KO3-2; UmPAC2KO5-1/UmPAC2KO5-2 and UmPAC2KO3-1/UmPAC2KO3-2; Um04807KO5-1/Um04807KO5-2 and Um04807KO3-1/Um048071KO3-2; (Sequences in Table S2). These fragments were digested with SfiI and ligated with the 1.9 Kb SfiI carboxin resistance cassette, 2.7 Kb SfiI hygromycin resistance cassette, or 1.5 Kb SfiI neourseotricin resistance cassette as described previously [35]. Ligation products were then clone into pGEM-T-EASY vector (Promega). PCR generated linear DNA for each construct was used for U. maydis transformation.
For complementation of the tup1 deletion, the p123-tup1 plasmid was generated. p123-tup1 is a p123 [92] derivative in which the eGFP fragment has been substituted with the tup1 open reading frame . For this purpose, the tup1 open reading frame was amplified by PCR with the oligonucleotides Tup1-Start and Tup1-Stop, which contain NcoI and NotI restriction sequences respectively. Phusion high fidelity DNA polymerase (Invitrogen) was used. The PCR product was digested with NcoI and NotI, purified, and cloned into a p123 vector digested with the same restriction enzymes. Positive cloning was verified by restriction analysis and sequencing. To generate SG200Δtup1Potef:tup1 strain, p123-tup1 was linearized with SspI and integrated into SG200Δtup1 ip locus by homologous recombination.
For over-expression of pac2, the p123-pac2 plasmid was generated by replacing the eGFP fragment from p123 with the pac2 open reading frame. The Pac2 open reading frame was amplified using the oligonucleotides UmPac2ATGSmaXma y UmPac2StopNotI, digested with XmaI and NotI restriction enzymes and ligated into the p123 vector digested with the same enzymes. Successful cloning was verified by restriction analysis and sequencing. To generate SG200pac2con, p123-pac2 was linearized with SspI and integrated into SG200 wild-type strain ip locus.
For constitutive expression of pac2 in FB1Pcrg1:fuz7DD, we constructed the plasmid p5HOP2. This plasmid consists in 1 kb fragment of the upstream sequence of pac2 open reading frame (ORF) followed by the otef constitutive promoter, the hygromycin resistance cassette and 1 kb of the pac2 ORF integrated in a pGEM-T-EASY vector. For this purpose 1 kb fragment of the upstream sequence of pac2 was amplified with the primers Umpac2-5UTR-1 and Umpac2-5UTR-2, using FB1 genomic DNA; the otef constitutive promoter followed by 1kb of pac2 ORF was amplified with the primers Umotefpac2 and Umpac2-+1kb, using the plasmid p123-pac2 as template. Both flanks where then digested with SfiI restriction enzyme and ligated with the hygromycin resistance cassette. This construction was ligated to a pGEM-T-EASY vector. FB1Pcrg1:fuz7DDpaccon was generated by transformation of the wild-type FB1Pcrg1:fuz7DD with the mentioned construct.
Single homologous integration of the linear plasmids or PCR products transformed was verified by PCR and Southern blot.
In the expression analysis, cells grown on liquid culture were recovered by centrifugation, washed with cold water, and total RNA was isolated with QIAGEN (Valencia, CA) RNeasy mini kit. For charcoal grown cells, biomass was recovered and transferred to liquid nitrogen pre-chilled mortars. Total RNA was then extracted from the crushed powder with trizol reagent (Invitrogen) and with the QIAGEN RNeasy mini kit. Isolated RNA was separated by formaldehyde denaturing agarose gel electrophoresis, and transferred overnight by capillary action to nylon membranes. Probes were obtained by PCR with the oligonucleotides indicated in Table S6. Radioactive labelling of PCR generated probes was carried out. Radioactive bands were visualized and quantified using a Molecular Dynamics PhosphoImager.
For qRT-PCR first strand cDNA synthesis was performed using the Transcriptor First Strand cDNA Synthesis Kit (Roche) according to the manufacturer's protocol. As a template for the reaction 1 µg of total RNA was used. Samples were incubated at 50°C for 1 hour. Real-time PCR was performed in a ABIPRISM 7000 Sequence Detection System (Applied Biosystems) using the Power SYBR Green PCR Master Mix according to the manufacturer's protocol. Primers used for detection are shown in Table S6.
U. maydis Tup1 sequence was obtained from MIPS U. maydis DataBase (http://mips.gsf.de/genre/proj/ustilago/). S. cerevisiae and C. albicans Tup1 sequences were obtained from SGD (http://www.yeastgenome.org/) and CGD (http://www.candidagenome.org/) databases, respectively. The rest of the Tup1 sequences were obtained from the NCBI. Multiple sequence alignments were made with ClustalW2. Domain structure analysis was performed using InterProScan Sequence Search tool from the European Bioinformatics Institute (http://www.ebi.ac.uk/). Pfam retrieved domains were used. Schematic representation of the retrieved domains was performed maintaining proportions of each domain with respect to the whole protein sequence length.
Cells were grown on nitrate minimal media containing 1% glucose or 1% maltose to an OD600 of 0.6–0.8, then pelleted and resuspendend in sterile water to an OD600 of 1.0. Fluorescence from 200 µl of cell suspension transferred to a microtiter plate was measured by using a POLARstar Omega fluorescence reader (BMG LABTECH). GFP fluorescence was measured at a wavelength of 485 nm for excitation and 520 nm for emission. Fluorescence was normalized to OD600. At least three independent experiments were performed, each measured in triplicate.
Cell morphology of WGA-stained cells, conjugation tube and b-dependent filament formation were analyzed with a Zeiss Apotome microscope.
For on planta quantification of filament and appressoria formation in co-infection experiments with U. maydis CFP and YFP labelled strains, leaf samples were stained with calcofluor white (Sigma) to visualize fungal material and then checked for CFP or YFP fluorescence. Quantification of filament formation on charcoal plates was performed by fluorescence analysis of colony samples from co-spotted YFP and CFP strains. Post-penetration stages were visualized by WGA-AF 488 and Propidium Iodide (Sigma) staining of infected leaf samples as previously described [93]. Samples were examined using a Leica fluorescence microscope, equipped with a PlanApo x 100 lens and a Deltavision widefield microscope (Applied Precision, Issaquah, WA) equipped with 20, 40, 63 and 100 x lens. Image processing was carried out using Adobe Photoshop CS2.
SG200 and SG200Δtup1 cells were grown on YEPSL until exponential phase, then washed twice with sterile water and cultured on minimal charcoal array plates (12.5% Holliday salts, 2% vitamins, 30 mM L-glutamine, 2% glucose, 4% agar and 2% charcoal, pH 7) during 48 hours at 25°C. 144 cm2 plates and a cell density of OD600 of0.1/cm2 was used. DNA-array analysis was performed using custom-designed Affymetrix chips (UstilagoA). Probe sets for the individual genes can be obtained from http://mips.helmholtz-muenchen.de/genre/proj/ustilago/. Target preparation, hybridization and data analysis was performed as described before [94], with the following alterations: total RNA was extracted as commented in DNA and RNA procedures for charcoal growing cells; 5 µg RNA were used for first strand cDNA synthesis at 50°C with Superscript II (Invitrogen); an adjusted P-value of ≤0.01 for the false discovery rate [95] and a change in expression of ≥2 was used for filtering. Expression values were calculated as mean of two biological replicates. Array data can be accessed at GEO/NCBI database (accession number GSE29591).
U. maydis sequence data can be found in the GenBank/EMBL data libraries under accession numbers XP_759427 for Tup1, XP_762643.1 for Pac2,, XP_756724 for bE1, XP_756725 for bW1, XP_758529 for Mfa1, XP_760967 for Acf1, XP_762479 for Egl1, XP_762172 for Rop1, XP_762530 for Hap2, XP_758660 for Crk1, XP_758860 for Prf1, XP_757661 for Fuz7, XP_760954 for um04807, XP_758669 for um11413, XP_756174 for um00027, XP_759558 for um03411 and XP_758874 for um02727. Other sequences used in this study have the following accession numbers: S. cerevisiae Tup1p, NP_010007; C. albicans Tup1, AAB63195; C. neoformans Tup1, XP_570974; P. marneffei TupA, AAL99251; N. crassa Rco-1, AAB37245; A. nidulans TupA ACD46267; S. pombe Tup11, NP_592873; S. pombe Tup12, NP_592910.
|
10.1371/journal.pgen.1004410 | Speciation and Introgression between Mimulus nasutus and Mimulus guttatus | Mimulus guttatus and M. nasutus are an evolutionary and ecological model sister species pair differentiated by ecology, mating system, and partial reproductive isolation. Despite extensive research on this system, the history of divergence and differentiation in this sister pair is unclear. We present and analyze a population genomic data set which shows that M. nasutus budded from a central Californian M. guttatus population within the last 200 to 500 thousand years. In this time, the M. nasutus genome has accrued genomic signatures of the transition to predominant selfing, including an elevated proportion of nonsynonymous variants, an accumulation of premature stop codons, and extended levels of linkage disequilibrium. Despite clear biological differentiation, we document genomic signatures of ongoing, bidirectional introgression. We observe a negative relationship between the recombination rate and divergence between M. nasutus and sympatric M. guttatus samples, suggesting that selection acts against M. nasutus ancestry in M. guttatus.
| While speciation is often depicted as a simple population split, in many cases it is likely more complex. Recently, whole genome sequencing and computational methods to interpret patterns of genomic variation have facilitated the inference of complex speciation histories. We present and analyze genomic data to infer the speciation history of an ecological and evolutionary model species pair - Mimulus guttatus/M. nasutus. We infer that M. nasutus split from a central Californian M. guttatus population approximately 200–500 kya, roughly corresponding to M. nasutus’ shift to self-fertilization. We document ongoing gene flow between these species where they co-occur. Finally, we present patterns genomic divergence suggesting that natural selection disfavors introgression of M. nasutus ancestry in M. guttatus.
| While speciation is often depicted as a simple event in which a single species splits into two, there is increasing evidence that this process is often more complex. In particular, speciation reflects a tension among divergence, the assortment of ancestral variation, ecological interactions and in some cases introgression that play out across the environment of the incipient species. Historically, a population genetic view of the process of speciation has been limited to few loci, where stochasticity in ancestral processes can prevent strong inferences about isolation and gene flow. By contrast, whole genome resequencing (even of only a few individuals) reveals many genealogical histories across contiguous genomic regions to provide well-resolved views of population history, divergence and introgression [1]–[5]. Here, we present a population genomic investigation of the speciation history of two closely related species of yellow monkeyflowers, the primarily outcrossing Mimulus guttatus, and the self-pollinating M. nasutus – an evolutionary model system for which the genetic and ecological basis of reproductive isolation is reasonably well characterized [6].
In flowering plants, speciation often involves a shift in pollinator (e.g., [7]–[9]) or mating system (e.g., [7], [10]–[12]), with concomitant divergence in key floral traits causing reproductive isolation between lineages. The evolutionary transition from outcrossing to self-fertilization, as occurred in M. nasutus, is of particular interest because the expected reduction in both the effective population size and effective recombination rate [13], [14] can dramatically alter population genetic processes and patterns of genomic variation [15], [16]. Recent evidence for elevated levels of putatively deleterious alleles in selfing taxa [17]–[19] is consistent with the idea that inbreeding reduces the effectiveness of purifying selection (due to a lowered effective population size). However, we still have few examples of the effects of self-fertilization on patterns of diversity across the genome, particularly in the context of recently diverged and potentially hybridizing species. Genomic datasets from young selfing species can uniquely inform the process of mating system divergence by allowing us to compare regions of the genome that share a common ancestor before or after the origin of self-fertilization and thus understand the assortment of ancestral variation [20].
The M. guttatus – M. nasutus species pair is an excellent model for investigating the causes and consequences of mating system evolution and species divergence. M. guttatus is primarily outcrossing (although the outcrossing rate varies across populations [21]–[23]) with large, bee-pollinated flowers and occupies diverse ecological habitats throughout western North America. M. nasutus is highly selfing with reduced, mostly closed flowers. Although these species are often found in different microhabitats, the range of M. nasutus is broadly nested within that of M. guttatus and the two species do co-occur. In sympatry, M. nasutus and M. guttatus are partially reproductively isolated by differences in floral morphology, flowering phenology, and pollen-pistil interactions [24]–[26]. Although early-generation hybrids occur in nature [24], [27], numerous intrinsic hybrid incompatibilities decrease hybrid fitness [28]–[30]. Based on the most detailed population genetic analyses of Mimulus to date (two and six nuclear loci, respectively [30], [31]), M. nasutus exhibits reduced diversity compared to M. guttatus, and some M. guttatus sequences are nearly identical to M. nasutus, suggestive of historical introgression. However, this limited view of the genome cannot resolve the timing and genomic consequences of divergence between Mimulus species, nor can it inform the extent or consequences of introgression between them.
We present the first population genomic analysis of M. guttatus and M. nasutus, spanning diverse ecotypes collected from throughout the species' ranges. We use these dense and contiguous population genomic data to estimate the population-split time, quantify rapid loss of ancestral variation accompanying the transition to selfing in M. nasutus, and identify ongoing, bidirectional introgression. Additionally, we observe a negative correlation between the recombination rate and interspecific divergence between M. nasutus and sympatric M. guttatus, a result best explained by selection against introgressed M. nasutus ancestry in M. guttatus. Our approach provides a detailed view of differentiation and introgression in a tractable ecological, genetic, and evolutionary model system.
We present and analyze a population genomic dataset of nineteen lab and/or naturally inbred (see Table S1) Mimulus samples – thirteen M. guttatus, five M. nasutus, and one of M. dentilobus, an outgroup. We generated sequence data for five of these samples (four M. nasutus and one M. guttatus), and accessed data for the other 14 samples from previously existing resources (see Materials and Methods for sequence sources and processing details). Collections spanned the ecological and geographic ranges of each species (Figure 1A and Table S1). Many of our analyses focus on four M. guttatus and four M. nasutus collections sequenced to relatively high depth (13.8×–24.7×) and with identical read lengths (100 bp, paired-end reads). Pairwise comparisons of nucleotide diversity among all nineteen samples are presented in Tables S2A and S3. Individual heterozygosity in the eight focal samples was relatively low (see METHODS) and was not clustered in regions of residual heterozygosity (contra the observation in naturally inbred Capsella rubella [20]) suggesting that natural and lab inbreeding has resulted in near total homozygosity by (recent) descent. Throughout this manuscript, we present results from samples aligned with bwa [32]. In Table S2B, we show that qualitative patterns of relative differentiation in focal samples are consistent when analyzed with a different alignment program, Stampy [33], demonstrating that our results are robust to the choice of bioinformatic pipeline.
Of our focal M. guttatus samples, CACG and DPRG are narrowly sympatric with M. nasutus populations from the northern and southern portion of both species' ranges, respectively. Our focal northern allopatric M. guttatus collection, AHQT, is well outside the geographic range of M. nasutus. By contrast, the southern allopatric collection (SLP) is geographically close to M. nasutus populations. Our focal M. nasutus collections also include sympatric and allopatric samples from the north and south (Table S1).
Overall patterns of genomic differentiation show deep population structure in M. guttatus, with M. nasutus diverging from a central Californian M. guttatus population approximately 200 kya.
To visualize pairwise relatedness, we constructed a rate-smoothed neighbor-joining (nj) tree (see METHODS for a discussion of the nj approach in population genetics). This tree clearly displays a deep phylogeographic split within M. guttatus, roughly corresponding to northern and southern parts of its range; however, geography is an imperfect predictor of genetic structure within M. guttatus (e.g., DUN is from a northern latitude yet clusters with our southern M. guttatus samples). The tree places all M. nasutus samples as a node within the southern M. guttatus cluster (Figure 1B). The fact that M. guttatus is paraphyletic suggests that M. nasutus budded from within a structured ancestral M. guttatus population. A principle component analysis (PCA, Figure 1C) also reveals the genetic structure within M. guttatus – PC2 differentiates northern and southern M. guttatus groups. Consistent with the single origin of M. nasutus, PC1 separates M. guttatus from the strongly clustered M. nasutus, presumably as a consequence of a shared history of genetic drift among these M. nasutus samples. Down-sampling to any one M. nasutus sample controls for this shared drift, and places M. nasutus within southern M. guttatus (Figure S1).
To support these qualitative inferences we generated a quantitative description of genetic structure within M. guttatus, focusing on our high-coverage (focal) samples. Pairwise sequence diversity at synonymous sites within northern (πS AHQT×CACG = 3.97% [3.89%–4.06%]) and southern (πS DPR×SLP = 4.45% [4.39%–4.52%]) M. guttatus samples is significantly lower than that within M. guttatus overall (πS = 4.91% [4.85%–4.96%]), and between the north and south (πS = 5.26% [5.20%–5.30%], Figure 1D). Diversity within the northern and southern clades is consistent with a very large effective population size (Ne) of approximately one and a half million chromosomes for both groups (assuming the per generation per base mutation rate, μ = 1.5×10−8 [following Koch et al. 2001]). As a simple estimate of the population split time (τ generations), ignoring possible introgression, we assume that the divergence between populations is the sum of pairwise diversity (π) within an ancestral population and the product of the per-generation mutation rate, μ, and two times the split time [34]. Using this relationship, and representing ancestral diversity by the southern M. guttatus samples, we set τ = (πS NorthGut×SouthGut−πS SouthGut)/2μ and estimate a split between northern and southern Mimulus populations more than a quarter of a million years ago (265 ky [251 ky–280 ky], assuming an annual life history). As above, this estimate assumes μ = 1.5×10−8/bp/generation but can be linearly rescaled by alternative estimates of μ. For example, readers can multiply divergence time estimates by a factor of two if they prefer the estimate of μ = 7×10−9/bp/generation [35].
Interspecific divergence between M. guttatus and M. nasutus (dS = 4.94% [4.88%–5.00%]) is comparable to overall M. guttatus diversity, and exceeds diversity within northern or southern M. guttatus collections (Figure 1D). We derive a simple estimate of split time between M. guttatus and M. nasutus as we did above to estimate the split between focal northern and southern M. guttatus samples. Using the difference between divergence of M. nasutus from the southern, allopatric M. guttatus sample (to minimize the influence of recent introgression and historical divergence between M. guttatus’ genetic clusters) and a proxy for diversity in an ancestral population (southern M. guttatus), we estimate that M. nasutus and M. guttatus split approximately 200 ky ago, τ = (πS Nas×AlloSouthGut−πS SouthGut)/2μ = 0.5875%/2μ = 196 ky [181 ky–212 ky].
As a complementary inference of historical patterns of divergence within M. guttatus and between species, we applied Li and Durbin's implementation of the pairwise sequentially Markovian coalescent (PSMC) [36] to pairwise combinations of focal haploid genomes (Figures 1E and S2, S3, S4, S5, S6). The PSMC analysis infers large population sizes within both northern (CACG×AHQT) and southern (SLP×DPRG) samples, with an apparent bout of strong recent population growth. However, as we have sampled from a structured population the inferred larger recent population sizes likely represent reduced coalescent rates caused by population structure, rather than dramatic recent increases in Ne. Likewise we infer a very low rate of coalescence (a very large effective population size) in the recent past between northern and southern M. guttatus (SLP×AHQT) compared to within these groups (SLP×DPRG and CACG×AHQT, Figures 1E and S2) likely reflecting the strong genetic structure within range-wide M. guttatus.
We also use this PSMC analysis for an additional estimate of the approximate split time, by assessing when the inferred coalescent rate between species decreases (i.e., the population size estimate increases) relative to the rate within M. guttatus [see 36]. In doing so, we focus on the southern M. guttatus samples that fall closest to M. nasutus in our nj tree. The inferred coalescent rate between M. nasutus and southern M. guttatus (SLP×KOOT, gray line) decreases relative to the rate within southern M. guttatus (SLP×DPRG, dark blue/navy line), i.e., the lines diverge, from ∼500 to ∼300 kya, suggesting either a gradual split between species over that time span, or a hard split sometime within that range (Figures 1E and S3). This result, which represents an upper bound on time since speciation, is qualitatively similar to our lower estimate based on synonymous nucleotide variation among these samples. We note that for both analyses, historical introgression of M. nasutus into SLP would make this split seem more recent than it actually was.
Patterns of genomic variation within M. nasutus reflect the genomic consequences of a recent transition to selfing. Synonymous diversity within M. nasutus (πS = 1.09% [1.03%–1.14%], Figure 1D) is one fifth that observed within M. guttatus, consistent with a high rate of genetic drift since M. nasutus’ origin. Moreover, most ancestral variation in M. nasutus has been homogenized: of the fixed differences between M. nasutus and M. guttatus, 90% are derived in M. nasutus and 10% are derived in M. guttatus (when polarizing by M. dentilobus). Although M. nasutus has lost much of its ancestral variation, shared variants still constitute a much higher proportion of its polymorphism (50%) relative to an equally sized sample of M. guttatus (10%). This pattern reflects both the paraphyly of M. guttatus and the incomplete sorting of variation present in M. nasutus’ founders.
Consistent with this reduction in nucleotide diversity and incomplete sorting of ancestral variation, PSMC analyses infer a dramatic decline in M. nasutus’ effective population size after it split from M. guttatus (compare red and black-gray lines in Figure 1E, see also Figure S4), suggesting that the evolution of selfing roughly coincided with M. nasutus’ split from M. guttatus. We caution, however that interpretation of PSMC's estimated population size in M. nasutus is not straightforward. This is because the transition to selfing reduces the population recombination rate more than the population mutation rate [14]; however, Li and Durbin's [36] implementation of the PSMC assumes that both these values change proportionally with the historical effective population size.
Relative to expectations under selective neutrality and demographic equilibrium, M. nasutus contains an excess of high-frequency derived synonymous alleles (Figure S7). We interpret this observation as a reflection of a recent population contraction. This interpretation is in agreement with the decreased synonymous diversity in M. nasutus relative to M. guttatus and our PSMC-based inference of a reduction in Ne. However, population structure within M. nasutus may also contribute to this excess of high frequency derived alleles [37], [38]. By contrast, in M. guttatus we observe slightly more rare synonymous alleles than expected under a neutral equilibrium model, reflecting recent growth, population structure, and/or weak selection against unpreferred codons.
The distribution of synonymous diversity across the genome (overlapping 5 kb windows with a 1 kb slide, Figure 2A) bolsters the view that M. nasutus’ genomic diversity is a mixture of closely related genomic regions that rapidly coalesce in the small M. nasutus population, and distantly related regions that do not coalesce until joining a large M. guttatus-like ancestral population. In pairwise comparisons of sequence diversity within M. nasutus, half of the genomic windows are differentiated by πS<0.5% (corresponding to ∼170 thousand years of divergence), reflecting recent common ancestry since the species split. On the other hand, one third of such windows are differentiated by πS>2.0%, reflecting deep ancestry in a large ancestral population (Figure 2A, see Figure S8 for different window sizes).
These findings contrast sharply with comparisons within M. guttatus, as well as between M. nasutus and allopatric M. guttatus samples, for which recent common ancestry since the species split is rare (πS<0.5% for less than 1.5% of 5 kb windows) and deep coalescence is the norm (mode πS = 4%, Figure 2A, a result roughly consistent across window sizes, Figure S8). Under the neutral coalescent, a pair of lineages will fail to find a common ancestor with each other by generation t with probability e−t/Ne*, where Ne* is the (constant) effective number of chromosomes. Therefore, the observation that half of our M. nasutus windows share a common ancestor in the past ∼170 ky, by an admittedly crude calculation, predicts a population size between 150k and 250k effective chromosomes (compared to the estimated Ne* of 1.5 million in M. guttatus from synonymous diversity, above). This ten-fold reduction in effective population size as compared to M. guttatus far exceeds both the two-fold decrease in Ne expected to accompany the evolution of selfing and the four-fold decrease calculated by the difference in intraspecific variation.
Across the genome, the mosaic nature of ancestry within M. nasutus is apparent as long contiguous regions of recent common ancestry (colored windows in Figures 2B and S9) interrupted by regions of deep ancestry, due to incomplete lineage sorting and/or historical introgression (white windows in Figure 2B and S9). This block-like ancestry structure results in extensive linkage disequilibrium (LD) in M. nasutus. In contrast to M. guttatus, for which the sample pairwise LD drops halfway towards its minimum values within only 15–20 base-pairs, LD in M. nasutus decays much more slowly, not dropping halfway towards its minimum values until 22 kb (Figure 2C). This represents a thousand-fold difference in the decay of LD, as compared to a more modest ten-fold reduction in the effective population size between M. nasutus and M. guttatus. This dramatic difference in the scale of LD between Mimulus species is likely due to a major reduction in the effective recombination rate within the selfing M. nasutus. We use this difference to derive a simple estimate of M. nasutus’ selfing rate using Eq. 1 of Nordborg [14]. Nordborg showed that the ratio of the population-scaled recombination to mutation rate in selfers is reduced by a factor of 1-F compared to the same population if it was outcrossing, where F is the inbreeding coefficient. Substituting in the hundred fold difference in the ratios of effective population size and decay of LD between M. nasutus and M. guttatus, we arrive at 1−F = 0.01. Assuming a constant selfing rate s, F = s/(2−s), M. nasutus’ selfing rate is approximately 99%.
Patterns of sequence variation suggest a reduced efficacy of purifying selection in M. nasutus, a result consistent with extensive genetic drift and/or linked selection within M. nasutus. All M. nasutus samples contain more premature stop codons than any M. guttatus sample (M. nasutus: mean 124.0, range = 121–126, M. guttatus: mean 95.5, range = 86–102), and a large proportion of these premature stops are at high frequency in M. nasutus (Figure 2D). For 27 of the 29 fixed differences for a premature stop codon, M. nasutus carries the premature stop and M. guttatus carries the intact allele. We acknowledge that errors in annotation could underlie some of the excess of premature stop codons inferred in M. nasutus. However, M. nasutus and the focal southern M. guttatus samples are equally diverged from the reference and yet our southern M. guttatus samples do not show this excess. As such, annotation error is likely not a strong contributor to the large and consistent interspecific differences in premature stop codons observed.
Additionally, after standardizing by synonymous variation, we observe an excess of putatively deleterious, non-synonymous variation in M. nasutus relative to M. guttatus (πN/πS = 0.197 [0.192–0.203] and 0.157 [0.155–0.160], respectively). However, this difference is not yet reflected in divergence between the species (dN/dS = 0.156 [0.154–0.159]), presumably because interspecific sequence differences largely reflect variation that predates the origin of selfing in M. nasutus rather than the relatively few mutations accrued within the past ∼170 ky. This pattern of elevatedπN/πS in selfing species but only modest dN/dS between selfers and their close relatives is common [15], even in genome-wide analyses (e.g., [20]). We note that elevated πN/πS in selfing species may reflect the faster rate at which nonsynonymous diversity approaches equilibrium after a reduction in diversity compared to synonymous mutations [39], [40], rather than a consequence of a reduced efficacy of purifying selection. However, this interpretation cannot explain the absolute excess of premature stop codons in M. nasutus.
Genetically, M. nasutus clusters with central Californian M. guttatus samples, suggesting that speciation post-dated the differentiation of some M. guttatus populations. Thus, speciation in this pair is best described as a ‘budding’ of M. nasutus from M. guttatus, rather than a split of an ancestral species into two. We observe an approximate coincidence between the timing of divergence and the decline in population size in M. nasutus (as inferred from our PSMC analysis). This observation could be a result of the transition to selfing being linked to speciation (see [11] for phylogenetic evidence of this link in the Solanaceae and [51], [52] for a likely case in Capsella). However, given the misspecification of the PSMC model for the transition to selfing (see above), this observation should be treated with caution. More work is needed to develop methods to test whether split times and changes in selfing rate occur concurrently to see if this is indeed a general pattern.
Future genomic analyses across the M. guttatus complex and other species groups will facilitate an in-depth view of the causes and consequences of speciation by the budding of selfing and/or endemic populations from widespread parental species. We note that recent phylogenetic analyses of species' ranges suggest that this mode of speciation is common in Mimulus [53] and other flowering plants [54].
We estimate that M. nasutus split from a M. guttatus population within the last two hundred to five hundred thousand years (with our estimate of ∼200 ky, inferred from differences in synonymous sequence differences within and between species, and the estimate of ∼500 ky corresponding to conservative estimates of population splits from the PSMC). This lies between the ∼50 ky separating selfing Capsella rubella from outcrossing C. grandiflora [20], [55] and Arabidopsis thaliana which has potentially been selfing for over a million years ([56], having split from A. lyrata ∼3–9 Mya [57]). Although 200 ky represents a relatively short time evolutionarily, it implies that M. nasutus managed to survive numerous dramatic bioclimatic fluctuations.
The transition from outcrossing to self-fertilization in M. nasutus has had clear consequences on patterns of genomic variation. In M. nasutus, linkage disequilibrium exceeds that in M. guttatus by three orders of magnitude. This result suggests a high selfing rate in M. nasutus (estimated above at 99%), consistent with direct estimates from field studies [24]. We observe a four-fold drop in diversity and infer a ten-fold reduction in the recent effective population size in M. nasutus compared to M. guttatus, values far exceeding the two-fold decrease in Ne expected as a direct consequence of selfing [58], [59]. This more than two-fold reduction in Ne of selfing populations relative to their outcrossing relatives has been identified in other plant [17], [55] and animal [60]–[62] species pairs, and may be partially due to extreme founding bottlenecks, frequent colonization events and/or demographic stochasticity that further increase the rate of genetic drift [13], as well as a heightened influence of linked selection in selfing taxa [60], [63]–[65].
Selfing populations are expected to experience a reduced efficacy of purifying selection accompanying the drop in effective population size and recombination rates [15], [65], [66]. Consistent with these predictions, M. nasutus has accumulated numerous putatively deleterious mutations, including nonsynonymous variants and premature stop codons. Presumably, this elevation in radical genetic variants reflects a reduction in the efficacy of purifying selection due to a high rate of genetic drift and linked selection, as well as perhaps the escape of some genes (e.g., loci involved in pollinator attraction) from the selective constraints they faced in an outcrossing population (e.g., [17]).
Despite multiple reproductive isolating barriers, including mating system differences, we find ongoing, bidirectional introgression between M. guttatus and M. nasutus.
Evidence of ongoing introgression from the selfer, M. nasutus, into the outcrosser, M. guttatus, is particularly stark. There are numerous evolutionary implications of introgression from selfers to outcrossers. Introgression of deleterious mutations accumulated in selfers may introduce a genetic load to outcrossers. This burden would result in selection against genetic material from selfers in hybridizing outcrossing populations, and could ultimately favor reinforcement of reproductive isolation. Alternatively, such introgression could provide a multi-locus suite of variation facilitating self-fertilization, and other correlated traits (e.g., drought resistance and rapid development), in favorable environments, as appears to be the case in introgression between wild and domestic beets (Beta vulgaris, [67]).
Evidence of introgression from M. guttatus into M. nasutus is subtler, but is potentially critically important. Even relatively low levels of introgression into a selfer may rescue the population from a build up of deleterious alleles, and reintroduce adaptive variation, and so may lower its chances of extinction, a fate considered likely for most selfing lineages [68], [69]. However, before potentially rescuing a selfing population from extinction, genomic regions introduced from outcrossing species must themselves survive a purging of deleterious recessive alleles.
Higher rates of introgression from M. nasutus to M. guttatus would be consistent with the prediction that backcrosses should be asymmetric – because bees preferentially visit plants with larger flowers [70], [71] and/or larger floral displays [72], [73], both features of M. guttatus, visits to M. nasutus and F1 hybrids are likely preceded and followed by visits to M. guttatus [24], [30]. Consistent with this prediction, direct estimates of hybridization in the DPR sympatric population reveal that F1 hybrids are the product of M. nasutus maternal and M. guttatus paternal parents, respectively [24]. However, we caution that it is considerably more challenging to identify introgression into M. nasutus than into M. guttatus, as the similarity between interspecific divergence and diversity in M. guttatus makes historical admixture difficult to separate from the incomplete sorting of M. nasutus’ ancestral variation. We further note that, although asymmetrical introgression from selfers to outcrossers has been detected in other systems (Pitcairnia [74], and potentially in Geum [75], [76]), the relative contribution of selfing vs. other isolating barriers and/or selection is unclear. Dense sampling of sympatric and allopatric populations of outcrossing species experiencing ongoing gene flow with selfing relatives will allow for tests of these hypotheses. Importantly, the number, location and length-distribution of admixture blocks identified from genomic analyses provide information about the longer-term consequences and pace of introgression between selfers and outcrossers.
The numerous short blocks (in addition to long blocks) of M. nasutus ancestry observed in M. guttatus suggest that M. nasutus ancestry can potentially persist in an M. guttatus background for many generations. Despite this, M. guttatus and M. nasutus are still ecologically and genetically distinct.
We identified a genome-wide signature of selection against introgression of M. nasutus ancestry in M. guttatus, in the form of a negative relationship between the local recombination rate and absolute divergence. This relationship was highly significant in both sympatric comparisons, but only weakly significant in parapatry, and insignificant in allopatry. Additionally, we did not find a relationship between recombination and diversity within either species. Moreover, unlike a negative relationship between the recombination rate and relative measures of differentiation, such as FST or the number of fixed differences [e.g.], [ 77], [78,79], this finding cannot also be explained by a high rate of hitchhiking or background selection within populations since the species split [46], [48], [49]. Instead, it seems more consistent with M. nasutus ancestry being selected against more strongly in regions of low recombination due to linkage with maladaptive alleles that introgression would introduce. This suggests that the genome has potentially congealed as a barrier to gene flow in low recombination regions. We note that this ‘congealing’ (sensu Barton [80], [81]) requires a threshold density of locally adaptive mutations, measured in recombination distance, and does not require a complex model of multi-locus coadaptation. Previous reports of absolute divergence near the breakpoints of inversions (e.g., [82], [83]), or in centromeres relative to telomeres [84], suggested this result; however, genome-wide evidence for this basic prediction is scarce.
Further work, including experiments measuring selection on genetic variants in the wild, and larger sample sizes from both allopatric and sympatric populations, is needed to pinpoint which (if any) genomic regions are particularly strongly selected against in hybrids. Genetically mapped loci for adaptive interspecific differences [85] and hybrid inviability and sterility [29] are promising candidates. Indeed, recent analyses of the distribution of Neanderthal haplotype blocks in ∼1000 human genomes has identified genomic candidates for adaptive introgression from Neanderthals to humans and an apparent paucity of introgression at loci putatively influencing male fertility [86].
Our analyses of whole genomes from the M. nasutus - M. guttatus species pair provide a broad view of both the historical divergence in this group and the ongoing processes by which they remain distinct. Less than a half million years ago, a semi-isolated M. guttatus population evolved self-pollination and ultimately transformed into modern day M. nasutus. In the intervening time, this population experienced a contraction in effective population size, and accumulated deleterious mutations while spreading geographically across western North America. More broadly, our work demonstrates that much can be learned about population history from resequencing relatively few samples in a group with an annotated genome and an integrated physical-genetic map.
Despite numerous reproductive isolating barriers [24], [25], [27]–[29], sympatric populations of M. guttatus and M. nasutus are still exchanging genes. The low diversity and extensive linkage disequilibrium in M. nasutus facilitates straightforward identification of M. nasutus-like ancestry in M. guttatus, and we use the length-distribution of these blocks to parameterize the recent history of introgression. The many short M. nasutus ancestry blocks suggest that its ancestry can persist in M. guttatus beyond early generation hybrids, and the length distribution of this ancestry is consistent with more than one pulse of introgression. The genomic distribution of introgression is non-random – in sympatry, absolute interspecific divergence is greater in regions of reduced recombination, suggesting selection against long blocks of M. nasutus ancestry in M. guttatus. Additional sequencing of individuals in sympatry will help better parameterize the dynamics and extent of introgression from M. guttatus to M. nasutus and clarify the action of selection for or against admixed ancestry across the genome.
We utilized a combination of existing [downloaded from the NCBI SRA, sequenced by 87] and newly generated whole genome sequence data from 19 different lab and/or naturally inbred Mimulus accessions, including 13 M. guttatus, 5 M. nasutus, and 1 M. dentilobus individual as an outgroup (Table S1). Samples varied in their geography and life history. Mean sequencing depths range from 2× to 25×, and read lengths include 36, 76, and 100 base pair paired end reads. We present SRA accession numbers as well as depth, read length and additional sample information in Table S1, and note that we obtained the DPRG sequence data directly from the U.S. Department of Energy Joint Genome Institute. Our analysis included newly generated whole genome sequences from five lines (CACG, CACN, DPRN, NHN, and KOOT), and we present details of sequence generation in Text S1.
We aligned paired end reads to the M. guttatus v2.0 reference genome [87] using Burrows-Wheeler Aligner (bwa [32]) with a minimum alignment quality threshold of Q29 (filtering done using SAMtools [88]). Alignment-processing details can be found in Text S1. We produced a high quality set of invariant sites and SNPs simultaneously for all lines using the GATK Unified Genotyper, with a site quality threshold of Q40 [89], [90]. For all analyses described below, we exclusively used genotype calls from reference scaffolds 1–14, corresponding to the 14 chromosomes in the Mimulus genome. For all analyses (except PSMC, which requires a consistently high density of data, see below), we set also set a strict minimum depth cutoff of 10 reads per site. To assign genotypes at heterozygous sites, we randomly selected one of two alternate alleles. Such heterozygous sites are not concentrated in long genomic regions and account for approximately 1% and 2% of synonymous SNPs in average focal M. nasutus and M. guttatus samples, respectively. This translates to individual synonymous heterozygosity of approximately 0.2% and 0.5% in M. nasutus and M. guttatus, respectively, even before additional filtering to remove misaligned sites (see below). Because sequence diversity is relatively high in our sample, we also analyzed patterns of pairwise sequence diversity using reads aligned with Stampy [33] (expected divergence set to 5%) and an otherwise identical pipeline to that described above. We describe the number of reads mapped with bwa and Stampy for our focal, high coverage lines in Table S7.
To minimize misclassifying mismapped paralogs as SNPs, we then removed triallelic sites and censured genotypes at sites where individual depth was two standard deviations away from mean depth. After these filtering steps, we classified remaining genic loci as zero, two, three, or fourfold degenerate using the Mimulus guttatus v2.0 gene annotations provided by phytozome [87]. Compared to bwa, the Stampy pipeline generated quantitatively larger estimates of sequence diversity, but qualitatively similar results (i.e., rank order of pairwise πS, Table S2B). Because Stampy doubled individual heterozygosity at synonymous sites, and increased πN/πS, we believe that it may have mismapped a greater proportion of our reads. Therefore, although Stampy aligned a greater number of reads to the reference genome than bwa (Table S7), we conservatively focus on our bwa alignments for our major analyses. As noted above, none of our qualitative conclusions depend on the read alignment pipeline used.
In addition to descriptions of our analyses, below and in Text S1, we recreate many analyses, including our PCA, and HMM in a file submitted to Dryad. These analyses can be run on the processed genotypic data for all samples at SNPs used in nj and PCA analyses as well as comparisons between focal samples in 1 kb windows across the genome, all available from doi:10.5061/dryad.vp645.
|
10.1371/journal.pntd.0001590 | Detection of Mycobacterium ulcerans by the Loop Mediated Isothermal Amplification Method | Buruli ulcer (BU) caused by Mycobacterium ulcerans (M. ulcerans) has emerged as an important public health problem in several rural communities in sub-Saharan Africa. Early diagnosis and prompt treatment are important in preventing disfiguring complications associated with late stages of the disease progression. Presently there is no simple and rapid test that is appropriate for early diagnosis and use in the low-resource settings where M. ulcerans is most prevalent.
We compared conventional and pocket warmer loop mediated isothermal amplification (LAMP) methods (using a heat block and a pocket warmer respectively as heat source for amplification reaction) for the detection of M. ulcerans in clinical specimens. The effect of purified and crude DNA preparations on the detection rate of the LAMP assays were also investigated and compared with that of IS2404 PCR, a reference assay for the detection of M. ulcerans. Thirty clinical specimens from suspected BU cases were examined by LAMP and IS2404 PCR.
The lower detection limit of both LAMP methods at 60°C was 300 copies of IS2404 and 30 copies of IS2404 for the conventional LAMP at 65°C. When purified DNA extracts were used, both the conventional LAMP and IS2404 PCR concordantly detected 21 positive cases, while the pocket warmer LAMP detected 19 cases. Nine of 30 samples were positive by both the LAMP assays as well as IS2404 PCR when crude extracts of clinical specimens were used.
The LAMP method can be used as a simple and rapid test for the detection of M. ulcerans in clinical specimens. However, obtaining purified DNA, as well as generating isothermal conditions, remains a major challenge for the use of the LAMP method under field conditions. With further improvement in DNA extraction and amplification conditions, the pwLAMP could be used as a point of care diagnostic test for BU
| In order to develop a simple and rapid test that can be used to diagnose Buruli ulcer under field conditions, we modified the conventional LAMP assay by using a disposable pocket warmer as a heating device for generating a constant temperature for the test reaction and employed the use of crude sample preparations consisting of boiled and unboiled extracts of the clinical specimen instead of using purified DNA as the diagnostic specimen. Thirty clinical specimens from suspected Buruli ulcer patients were investigated by the modified LAMP (or pocket warmer LAMP) and the conventional LAMP, as well as IS2404 PCR, a reference method for the detection of Mycobacterium ulcerans. There was no significant difference in the detection rate (63–70%) in all of the methods when purified samples were used for the tests. On the other hand the use of crude specimen preparation resulted in a drop in detection rate (30–40%). This study demonstrates that the LAMP test can be used for rapid detection of M. ulcerans when purified DNA preparations are used. With further improvements in the sample reaction, as well as in specimen purification, the pocket warmer LAMP may provide a simple and rapid diagnostic test for Buruli ulcer.
| Buruli ulcer (BU) caused by Mycobacterium ulcerans (M. ulcerans) is a necrotizing skin disease endemic mostly in rural wetland of tropical countries of Africa, America, Asia and Australia. The disease also occurs in non-tropical areas of Australia, China and Japan. Globally, BU has been reported in over 30 countries [1]–[3]. The burden of BU is however most severe in sub Saharan Africa where the true incidence of the disease is difficult to determine as a result of poor surveillance measures and case confirmation [2]. Available data however reveals an increase in BU incidence over the last several years in the west African countries of Ivory Coast, Ghana and Benin. In these countries BU has replaced leprosy as the second most prevalent mycobacterial disease [1], [3]–[5].
BU begins as painless nodule, papule, plaque or edema that evolves into characteristic ulcers with undermined edges. If untreated, extensive ulceration (that can cover 15% of the body), scarring and contractures may cause serious functional disabilities in patients [5]–[7].
Unfortunately most patients seek treatment late and present with large ulcers [8]–[10]. Previously treatment of such lesions involved surgical removal of all the affected tissue and part of the surrounding tissues, eventually followed by skin grafting [9]–[12].
In 2004 antimycobacterial treatment alone (if necessary in combination with surgery) was introduced and has since been considered as the treatment of choice for BU [6], [13]–[16]. Laboratory confirmation of clinically suspected BU cases has therefore become crucial for the clinical management of BU [17].
Four laboratory tests are recommended for the diagnosis of BU. These include microscopic examination, culture, IS2404 PCR and histopathological analysis. Microscopic examination detects 29%–78% of clinically suspected BU cases and is currently the only rapid and affordable test available for BU diagnosis in many endemic areas. The detection rate of culture is between 34%–79% and takes an average of 9–12 weeks to yield positive results. Culture therefore cannot be used for rapid laboratory confirmation of BU. Histopathological analysis is reported to detect 30% additional cases than other confirmatory tests, however this technique is restricted to external reference laboratories and are unavailable in peripheral health centres or district or regional hospitals. IS2404 PCR has close to 96% sensitivity and is considered the method of choice for laboratory confirmation of BU [16]–[25].
The WHO recommends that at least 50% of cases must be confirmed by IS2404 PCR before commencement of antibiotic therapy [22]–[25].
However technical difficulties (eg, cold chain requirement, stable power supply and qualified laboratory staff) limit the use of this diagnostic test in BU endemic areas.
A dry reagent PCR consisting of lyophilized PCR mix which is reconstituted with water for testing DNA was developed to simplify BU diagnosis by PCR [26] but this method also requires the use of a thermocycler, electrophoresis and gel imaging equipment and therefore similarly makes the use of this diagnostic test for BU diagnosis in endemic areas unlikely.
The Loop Mediated Isothermal Amplification (LAMP) is a novel nucleic acid amplification method for molecular detection and identification [27]. The principle of LAMP is autocycling strand displacement DNA synthesis in the presence of Bst DNA polymerase with high strand displacement activity under isothermal conditions between 60–65°C within 60 minutes [28]. The assay is highly specific due to the recognition of target DNA by 4 to 6 independent sequences and the amplification efficiency of LAMP is equivalent to that of PCR based methods ([27], [29], [30]).
The LAMP reaction enables easy identification of positive tests due to the accumulation of high amounts of amplification products in the reaction tubes. Further improvement in visual identification can be realized through the addition of intercalating dyes such as SYBR green or hydroxynapthtol blue to reaction tubes [31]. This therefore precludes the need for post amplification analysis and hence reduces cost and labour. LAMP has also been shown to be less affected by a number of inhibitors of conventional PCR [32]. Additionally the closed tube format of this assay reduces problem of carry over contamination which is likely in less controlled environments [33]. With all of these characteristics LAMP of DNA has emerged as a powerful tool to facilitate point of care diagnostic test [31].
In order to develop a field applicable technique that offers high detection sensitivity and specificity for the diagnosis of BU, we explored the use of the pocket warmer LAMP (pwLAMP) technique, a DNA amplification method using isothermal conditions (60°C) provided by a disposable pocket warmer [34].
Ethical approval for analysing patients' specimens was obtained from the ethical review board of the Noguchi Memorial Institute for Medical Research. Specimens used were anonymously taken from an already existing collection of patients' specimens processed for diagnosis of BU from Agogo Presbyterian Hospital in Ghana.
Thirty clinical specimens consisting of 20 swabs and 10 fine needle aspirates taken respectively from ulcers and pre-ulcerative lesions of suspected BU patients were used in this study. The fine needle aspirate specimens were kept in 1 ml phosphate buffered saline (PBS) and swabs were stored dry in sterile tubes. Each swab was transferred into a tube containing 2 ml milli-Q purified water (Millipore Corporation, Billerica, MA) and gently vortexed for 5 sec and then removed. Portions 250 µl of the sample suspensions were transferred to separate new sterile eppendorf tubes containing 250 µl of lysis buffer (1.6 M GuHCl, 60 mM Tris pH 7.4, 1% Triton X-100, 60 mM EDTA, Tween-20 10%), 50 µl proteinase-K and 250 µl glass beads.
The mixtures were incubated horizontally in a shaker (200 rpm) at 60°C overnight. To capture the DNA, 40 µl of diatomaceous earth solution (10 g diatomaceous earth obtained from Sigma Aldrich Chemi GmbH in 50 ml of H2O containing 500 µl of 37% (wt/vol) HCl) was added to the suspensions and incubated at 37°C with shaking (200 rpm) for 60 min. The mixtures were centrifuged at 14,000 rpm for 10 sec and the resulting pellets were washed twice with 900 µl of 70% ethanol (2–8°C) followed by 900 µl of acetone. The pellets were dried at 50°C for 20 min and resuspended in 100 µl milli Q purified water and centrifuged at 14,000 rpm for 10 sec. The purified DNA was used as templates for both IS2404 PCR and LAMP assays to detect M. ulcerans.
To investigate the performance of the LAMP assay on crude DNA preparations, we obtained 2 types of DNA extracts for each clinical specimen. One crude extract consisted of 250 µl suspensions of the specimen boiled for 10 min followed by centrifugation at 14,000 rpm for 5 min (boiled extract). The other crude extract used was a 250 µl suspension of the unboiled specimen.
Ten M. ulcerans strains grown on LJ slants were harvested and DNA was extracted as previously described [35]. Serial dilutions of purified M. ulcerans DNA containing 300,000, 30,000, 300, 30 and 3 copies of IS2404 element per 5 µl were prepared. The number of copies of the insertion sequence element was determined based on the genome size of 5,806 kb and presence of an average number of 207 copies of IS2404. This was used to determine the detection limit of the LAMP assays.
PCR targeting IS2404 was performed as described previously [21]. The first and second round PCRs used primers pGp1: 5′-AGGGCAGCGCGGTGATACGG-3′and pGp2: 5′- CAGTGGATTGGTGCCGATCGAG-3′ and pGp3: 5′-GGCGCAGATCAACTTCGCGGT-3′ and pGp4: 5′-CTGCGTGGTGCTTTACGCGC-3′, respectively.
For the First round, the 30 µl reaction volume contained 3 µl DNA, 25 pmol/µl of each primer (pGp1 and pGp2), 3 µl of 10× PCR buffer (containing 1.5 mM magnesium chloride), 6.0 µl Q-solution, 0.2 mM deoxynucleotide triphosphates (dNTPs) and 1.0 U HotStar Taq polymerase (QIAGEN).
For the second run, 1 µl of the first run product was added to 24 µl reaction volume containing, 25 pmol/µl of each primer (pGp3 and pGp4), 2.5 µl of 10× PCR buffer, 5.0 µl Q-solution, 0.2 mM dNTPs and 1.0 U HotStar Taq polymerase. Amplification for both rounds for 40 and 35 cycles respectively was carried out in an Eppendorf mastercycler thermal cycler as follows: denaturation at 95°C for 15 min, 94°C for 30 sec, 64°C for 1 min, 72°C for 1 min, 30 sec and a final extension at 72°C for 10 min.
The second round PCR products were electrophoresed in a 2% TAE (0.04 M Tris-acetate and 0.001 M EDTA pH 8.0) agarose gel with ethidium bromide. The size of amplicons was estimated by comparison with 100 bp plus DNA ladder (Fermentas Life Sciences, EU) and visualized using Kodak Gel logic 100 Molecular Imaging System.
The LAMP assay was performed using a set of 6 primers comprising 2 outer primers (Buruli- F3: CGAGAACAGCCTGCACTG, and Buruli- B3: CGGTTGGCGGTCAAAGC).
Two inner primers (Buruli-FIP:GTGCGCCGTGTCCGGTATGGATACGCGATGTCACCTTC and Buruli- BIP: AGGTCCTAGCAACGCTACGCAAATCCGGCAGGCTTCGG), 2 loop primers Buruli-LF: GCCTTTGACGGTCTTCGTC, and Buruli- LB: (CACCGCGATCAATCTGCAC). The primers were designed using Primer Explorer (version 4; EikenChemical, Tokyo, Japan; http://primerexplorer.jp/elamp4.0.0/index.html).
Pocket warmer LAMP (pwLAMP) was performed using a loopamp DNA amplification kit (Eiken Chemical) described previously [32]. Each 25 µl reaction mixture contained 1.6 µM each of FIP and BIP, 0.2 µM each of F3 and B3, 0.8 µM each of LF and LB, 2× reaction mixture (12.5 µl), 1 µl of Bst DNA polymerase, 1 µl of fluorescence detection reagent (Eiken Chemical), 3.5 µl distilled water and 1 µl sample.
Reaction tubes were incubated at 60°C for 60 min in the heat block (GeneAmp 9700, Applied Biosystems, Foster City, CA) while with the pwLAMP, the tubes were sandwiched in a twofold pocket warmer (Hokaron Haru-type, Lotte Health Products, Tokyo, Japan) surrounded by a paper towel, and put in a Styrofoam box for 120 min (60 min reaction incubation). The reaction was terminated at 85°C for 5 min and the results were read by eye in ambient light and also using UV illumination.
To evaluate the specificity of the LAMP method for M. ulcerans, DNA extracts of eight Mycobacterium sp. (Mycobacterium marinum, Mycobacterium tuberculosis, Mycobacterium avium, Mycobacterium intracellulare, Mycobacterium kansasii, Mycobacterium abscessus, Mycobacterium chelonae), two Mycobacterium ulcerans strains (Mycobacterium shinsuence (Japanese strain) and one (African strain)) and Jurkat (human T cell line) were examined.
A chi-squared test was performed to reveal the statistical difference using SPSS (version 16.0; SPSS Inc., Chicago, IL) software.
LAMP reaction requires a constant temperature of about 60°–65°C for 60 min for amplification of DNA [27]–[34]. In a previous study a pocket warmer reached 58°C in 30 min and stayed around 60°C for more than 60 min in a Styrofoam box [34]. The 3 pocket warmers (of a pack of 30 hand warmers) tested in this study achieved a temperature of 60°C after 60 min and maintained this temperature for about 90 min. The pocket warmer thus provided a suitable temperature (60°C) and time range (60 min) for amplification. Both pwLAMP and the conventional LAMP assays were able to detect to the limit of 300 copies of the target sequence after 60 min of amplification. This limit improved to 30 copies when the conventional LAMP was carried out at 65°C (the pocket warmer was not able to attain this temperature and was therefore not investigated).
Observation under ambient as well as UV illumination demonstrated clearly that the LAMP reaction produced positive signal specifically in DNA from M. ulcerans, but not in DNA extracts of M. marinum, M. tuberculosis, M. avium, M. intracellularie, M. kansasii, M. abscessus, M. chelonae, and Jurkart, a human T cell line (Figure 1).
The sensitivity and specificity of the LAMP assays for the detection of M. ulcerans is shown in tables 1 and 2. Under ambient illumination, positive specimens in the LAMP assay produced greenish colouration (Figure 2). When purified DNA extracts were used, 21 (16 swabs, 5 fine needle aspirates) (70%) of 30 clinical specimens were positive by IS2404 PCR as well as by the conventional LAMP. None of the PCR positive specimens were negative by conventional LAMP. However 19 samples of purified DNA extracts were positive with the pwLAMP, but the 90.5% sensitivity (19/21) of the pwLAMP compared to the results by IS2404 PCR was not statistically significant (p = 0.58, Chi-square test).
All negative specimens in IS2404 PCR were negative in both LAMP assays, indicating specificities of both LAMP assays to the reference method were 100%. Twelve unboiled (9 swabs and 3 fine needle aspirates) and 9 boiled (6 swabs and 3 fine needle aspirates) extracts were positive by all 3 detection assays with sensitivities of 57.1% (unboiled, 12/21) and 42.9% (boiled, 9/21) compared to results using purified DNA extracts respectively for both LAMP and IS2404 PCR assays. The positivity of swabs was found to be in the range of 30% to 80% compared to 30% to 50% for fine needle aspirates.
When the positivities in crude DNA specimens were compared with those in purified DNA, the differences were statistically significant by chi-square test (unboiled vs purified DNA, p = 0.0195, and boiled vs purified DNA, p = 0.0019). None of the IS2404 PCR negatives was positive in the LAMP assays irrespective of the DNA extracts type used. These data suggest that sensitivity of LAMP and PCR assays for the detection of M. ulcerans in clinical specimens is enhanced when purified DNA extracts are used.
BU is a neglected tropical disease that mostly affects the poor in resource limited communities in sub-Saharan Africa [1]–[3]. IS2404PCR, the method of choice for confirmation of BU diagnosis cannot be operational in BU endemic areas [16],[17],[24]–[25].
The development of rapid and reliable point of care diagnostic assays is of high priority to BU management and prevention.
This study explored the potential use of the LAMP method for field diagnosis of BU. Some important limitations to the use of this assay in the field include specimen purification and difficulty in maintaining isothermal condition for the reaction. A study suggested that omission of DNA extraction or the use of crude DNA extracts have no effect on the LAMP test [32]. Hatano et al used a disposable pocket warmer to provide isothermal condition for LAMP reaction [34].
Based on this knowledge, we applied the pwLAMP to crude and purified DNA extracts in order to determine whether this method will be suitable for use as a point of care diagnostic test for BU.
Although the pocket warmers used in this study reached 60°C after 1 hr (instead of 30 min in previous studies [34]), both devices achieved the requisite temperature and holding time for executing LAMP reaction, a major advantage in the use of amplification based assay for the detection of an infectious agent under field condition. The pwLAMP did not cross-react with other mycobacteria (Figure 1). Moreover, the experiment on clinical specimens demonstrated that the pwLAMP had a 100% specificity in clinical specimens of BU.
The pwLAMP was found to have comparable sensitivity as the conventional LAMP at 60°C as both assays were able to detect 300 copies of IS2404 element (equivalent of 1.5 genomes of M. ulcerans). The detection limit of the conventional LAMP at 65°C improved to 30 copies of IS2404 and this level of sensitivity may probably be achieved with a pocket warmer that can attain a temperature of 65°C and maintain a holding time of 60 min. When applied to purified DNA extracts of clinical specimens, the pwLAMP, conventional LAMP and IS2404 PCR yielded concordant results (Tables 1 and 2).
However, 2 of the samples that were positive by IS2404 PCR/conventional LAMP were negative by the pw LAMP. None of the IS2404 PCR negative samples were positive in both types of LAMP assays. On the other hand we observed a drop in detection rate from 63–70% to 30–40% when crude extracts of clinical specimens were used (Tables 1 and 2).
This indicates that the use of crude DNA extracts as template may not be appropriate for the detection of M. ulcerans by the LAMP method. This observation contradicts a previous study that suggested omission of DNA extraction had no effect on sensitivity of the LAMP assay [32].
For the crude preparations however, it is noteworthy that the detection rate of the LAMP assay was significantly higher for the unboiled extracts than for the boiled extracts. Explanations for these results were not explored. The observation that the LAMP assay was not inhibited especially for the unboiled specimens is quite consistent with previous work that have shown LAMP to be tolerant to culture medium and to certain biological substances including phosphate buffered saline, serum, plasma, urine and vitreous [32].
In conclusion, the study demonstrates that the LAMP assay yields comparable results as IS2404 PCR when it is performed at 60°–65°C for 60 min on purified DNA extracts and further supports the use of the pocket warmer as a device for providing isothermal amplification condition for the LAMP assay. This therefore is a potential boost to the application of pwLAMP in resource poor settings.
Challenges of obtaining pure DNA extracts of clinical specimen as well as the use of a pocket warmer capable of maintaining 65°C for 1 hr, however needs to be addressed in order to improve the performance of the pwLAMP assay.
Further development and testing in larger numbers of specimens is therefore necessary to access the potential use of pwLAMP as a simple and rapid point of care diagnostic test for BU.
|
10.1371/journal.pgen.1005673 | Human β-D-3 Exacerbates MDA5 but Suppresses TLR3 Responses to the Viral Molecular Pattern Mimic Polyinosinic:Polycytidylic Acid | Human β-defensin 3 (hBD3) is a cationic host defence peptide and is part of the innate immune response. HBD3 is present on a highly copy number variable block of six β-defensin genes, and increased copy number is associated with the autoimmune disease psoriasis. It is not known how this increase influences disease development, but psoriasis is a T cell-mediated disease and activation of the innate immune system is required for the initial trigger that leads to the amplification stage. We investigated the effect of hBD3 on the response of primary macrophages to various TLR agonists. HBD3 exacerbated the production of type I Interferon-β in response to the viral ligand mimic polyinosinic:polycytidylic acid (polyI:C) in both human and mouse primary cells, although production of the chemokine CXCL10 was suppressed. Compared to polyI:C alone, mice injected with both hBD3 peptide and polyI:C also showed an enhanced increase in Interferon-β. Mice expressing a transgene encoding hBD3 had elevated basal levels of Interferon-β, and challenge with polyI:C further increased this response. HBD3 peptide increased uptake of polyI:C by macrophages, however the cellular response and localisation of polyI:C in cells treated contemporaneously with hBD3 or cationic liposome differed. Immunohistochemistry showed that hBD3 and polyI:C do not co-localise, but in the presence of hBD3 less polyI:C localises to the early endosome. Using bone marrow derived macrophages from knockout mice we demonstrate that hBD3 suppresses the polyI:C-induced TLR3 response mediated by TICAM1 (TRIF), while exacerbating the cytoplasmic response through MDA5 (IFIH1) and MAVS (IPS1/CARDIF). Thus, hBD3, a highly copy number variable gene in human, influences cellular responses to the viral mimic polyI:C implying that copy number may have a significant phenotypic effect on the response to viral infection and development of autoimmunity in humans.
| Defensins are classically known as antimicrobial peptides due to their ability to rapidly kill pathogens including bacteria, viruses and fungi. They are produced in the presence of infectious agents at body surfaces exposed to the environment. Increasingly, their functional repertoire is expanding, and they have been shown to modulate the immune system. In humans, there is a block of six β-defensin genes that varies in copy number in the population. Individuals with an increased number of β-defensin genes have an increased likelihood of developing the skin autoimmune disease psoriasis. It is not known how this increase in gene copy number influences development of the disease, and psoriasis is a complex interplay of genomic and environmental factors that trigger disease progression and include exposure to viruses. We examined whether a molecular pattern characteristic of viruses produces an altered immune response in the presence of the defensin human β-defensin 3 (hBD3). We find that hBD3 triggers a larger interferon defence response to this viral mimic by increasing accessibility to a cellular receptor that recognises viral patterns. Interferon is known to be important in autoimmunity and our work may explain why individuals with increased β-defensin number are predisposed to develop psoriasis.
| HBD3 is a member of the β-defensin multigene family. The amphipathic, antiparallel β-sheet structure, stabilised by disulfide bonds, via six canonical cysteines is conserved throughout evolution [1] and between family members despite significant sequence diversity [2]. These powerful cationic antimicrobials directly kill fungi, bacteria and viruses, and recently it has become clear that this gene family has roles in other processes including male fertility, immunomodulation and inflammatory disease [3]. Defensins are primarily expressed from mucosal surfaces, some exclusively in the reproductive tract and others in skin, intestine and gingival surfaces [4–6]. Of β-defensin genes, hBD3 is probably the most versatile and studies both in vitro and in vivo demonstrate its ability to chemoattractant immune cells [7]; encourage wound healing [8] and modulate innate signalling [9–11]. HBD3 (gene name DEFB103) and its mouse orthologue (Defb14) are promiscuous ligands with ability to bind the receptors CCR6, CCR2, CXCR4. In addition, a dominant mutation in the DEFB103 gene in dogs and wolves causes an increase in canine β-defensin 3 (CBD103) peptide level allowing off-target binding to melanocortin receptor 1 (MC1R), which results in black coat colour [7, 12–15].
DEFB103 is present on hypervariable clusters of six β-defensin genes and alteration in copy number may influence innate immune responses. Increased copy number of the cluster is associated with psoriasis [16, 17]. Increased defensin peptide level has been reported in serum of psoriasis patients, although the influence of defensins on the pathogenesis of the disease is not understood. Psoriasis is a T cell-mediated disease predominantly orchestrated by Th-17 cells. Amplification of the disease process is triggered by an initial phase modulated by an increase in innate immune signalling through pattern recognition receptors (PRR) such as toll like receptors (TLR) [18]. Psoriatic patients have an increase in dendritic cells and cationic antimicrobial peptides in the skin [19]. It has been shown that self and viral nucleic acids trigger an increase in the type I Interferon-α response of plasmacytoid dendritic cells (pDC) that are specialised cells for Interferon-α production through TLR9[20]. Blocking production of Interferon-α by these cells prevents T cell–dependent development of psoriasis in a xenograft model. The antimicrobial peptide LL-37 has been identified as a molecule that encourages recognition of self DNA and RNA through TLRs on pDC to induce release of Type I Interferon [21]. Recently antimicrobial peptides hBD2, hBD3 and lysozyme have also been shown to bind self-DNA and activate pDC through TLR9 to release Interferon-α. The presence of these peptides in psoriatic plaques suggests a concerted role for them in the pathogenesis of psoriasis [9, 19].
Here we determine the effect of hBD3 on the response of primary macrophages to known pathogen-associated molecular patterns (PAMPs) to investigate the influence of hBD3 on signalling from innate immune receptors. We have previously shown that hBD3 suppresses the TLR4-mediated response to bacterial lipopolysaccharide (LPS) which is mediated by both MyD88(Myeloid Differentiation Primary Response 88) and TICAM1 (Toll-Like Receptor Adaptor Molecule 1-also known as TRIF) [10]. We show that polyI:C in the presence of hBD3 has an exacerbated Interferon-β response and decreases CXCL10 production, in vitro and in vivo in both mice and human primary cells. PolyI:C is a synthetic double stranded RNA (dsRNA) and consequently a viral ligand mimic, which is recognised by endosomally located TLR3 and also by cytoplasmic RIG-I-like receptors (RLRs) [22]. Recently high molecular weight (HMW) or long poly:IC has been shown to preferentially access RLRs in conventional dendritic cells [23]. The RLRs include RIG-I (also known as DDX58 (DEAD (Asp-Glu-Ala-Asp) box polypeptide 58) and MDA5 (Melanoma Differentiation-Associated protein 5, also known as IFIH1 -interferon induced with helicase C domain 1) [24]. HMW polyI:C is recognised primarily by MDA5 (without the need for transfection) and secondarily (with transfection) by RIGI [23]. Activation of MDA5 and consequent interferon-β production has been shown to be associated with autoimmune disorders [25–28]. Both TLR3 and RLR types of receptor are MYD88 independent, with MDA5 and RIGI requiring the recruitment of the adaptor protein MAVS (mitochondrial antiviral signalling protein -also known as VISA, IPS-1 and CARDIF). Signalling through the MAVS pathway in macrophages results in IRF3/7 driven expression of type I Interferon and NF-ᴋB induction of inflammatory cytokines and chemokines [29, 30]. TLR3 specifically associates with the adaptor TICAM1 and mediates signal transduction which activates IRF3 and NF-ᴋB [30]. Our studies have been carried out using HMW polyI:C throughout. We dissect the effect on the pathways responsible for the altered response of macrophages to HMW polyI:C in the presence of hBD3 and reveal the mechanism that enables increased Interferon-β and decreased CXCL10 production.
We previously reported that hBD3 suppresses Toll Like Receptor 4 (TLR4)-induced signalling in response to LPS both through MYD88 and TICAM1 pathways [10]. To investigate the effect of hDB3 on other TLR pathways we exposed primary bone marrow derived macrophages from mice (BMDM) to a variety of TLR ligands and found that the response to TLR 2, 2/6, 1/2, 7 or 9 ligands were not significantly altered in the presence of hBD3 (Fig 1A). In agreement with our previous findings, hBD3 inhibited LPS-induced TNFα. In contrast, the response to HMW polyI:C was significantly exacerbated in the presence of hBD3, increasing TNF-α; Interferon-β and IL-6 production, however hBD3 significantly suppressed the CXCL10 (also known as IP10) response to polyI:C (Fig 1B). The enhancing effect of hBD3 on polyI:C cytokine induction, was also seen in the human monocytic cell line THP1, where hBD3 significantly enhanced polyI:C-induced TNFα, IL-6 and IL-8. However, in contrast to what we observe in mouse cells, there was no significant effect of hBD3 on polyI:C-induced CXCL10 in human cells (Fig 1C). In addition this enhancing effect was also seen on Interferon-β gene expression in primary human peripheral blood monocyte derived macrophages (PBMDM), measured by qRT-PCR (Fig 1D).
The amount of hBD3 required to induce an enhanced Interferon-β response to polyI:C differed from that required to decreased CXCL10 response (Fig 2A). PolyI:C-induced CXCL10 was inhibited by 4μg/ml hBD3, but no longer inhibited at 2μg/ml hBD3, whereas concentrations as low as 0.05 μg/ml hBD3 enhanced the Interferon-β response to polyI:C, although at this concentration the effect is beginning to diminish. We tested the importance of the hBD3 cysteine-stabilised structure using hBD3 with the canonical defensin motif of six cysteines (which form three intramolecular disulfide bonds) replaced with serines. This modified peptide (hBD3 Cys:Ser) did not augment cytokine production in response to polyI:C, suggesting that the enhancing effect of hBD3 on polyI:C signalling is dependent on the 3-dimensional structure of the hBD3 peptide (Fig 2B). This lack of effect was not due to the inability of the linear peptide to rapidly enter the cell as TAMRA(tetramethylrhodamine azide)-labelled hBD3 Cys:Ser peptide entered BMDM in 10 min (Fig 2B) similar to canonically folded hBD3, as shown in our previous studies[10].
To test whether these in vitro effects were relevant in vivo, we injected wild type mice with polyI:C in the presence of synthetic hBD3 peptide (Fig 3A). We found that TNF-α and IL12p40 responses to polyI:C were significantly increased in the presence of synthetic hBD3. Interestingly the cytokine CXCL10 was not decreased in the presence of hBD3 (Fig 3A) although there was a trend for CXCL10 to be lower in the presence of hBD3- in keeping with the significant reduction of this cytokine seen in the BMDM stimulated with polyI:C and hBD3.
There is a concern that unless synthetic hBD3 is correctly oxidised and the correct disulphide bonding achieved, the properties of the peptide may be altered [7]. The importance of structure is confirmed in our experiment with hBD3 Cys:Ser shown in Fig 2B. The oxidised synthetic peptide we used here (obtained from Peptide Institute, Japan) gives details of preparation and oxidation [31] and implies correct cysteine bonding (C1-C5; C2-4 and C3-C6). However, in order to investigate our effect with hBD3 oxidised in vivo, we expressed the gene (DEFB103) that encodes hBD3, as a transgene in mice. Transgenic mice were made using the pCAAG promoter (as described previously by Candille et al. [14]) by introducing a transgene containing the genomic copy of DEFB103 into ES cells. The vector (shown in S1A Fig) expresses hBD3 only after CRE recombinase-mediated deletion of the floxed DsRed, puromycin STOP spacer. After deletion, hBD3 and EGFP are expressed as a polycistronic mRNA and translated as independent proteins using an IRES site. We made several ES cell lines with the DEFB103 expression construct and isolated a clone with strong expression by virtue of DsRed expression and a single site of insertion of the transgene (located to the sub-telomeric region of chromosome 12 by FISH and chromosome painting (S1B Fig). The control DsRed transgenic mice were made with this ES clone. HBD3 transgenic mice were made from the same ES clone after treatment with CRE recombinase. These cells had strong expression of EGFP and hBD3 and only the ES cells containing the CRE-excised vector showed expression of DEFB103 mRNA (S1C Fig). HBD3 mRNA expression was detected in all tissues tested from mice made using the hBD3-Tg ES cells and hBD3 protein was detected in various tissues including BMDM by immunohistochemistry (S1D Fig) using an hBD3 monoclonal antibody [32]. No hBD3 was detected in DsRed-Tg control mice. The effect of polyI:C exposure in mice expressing physiologically secreted hBD3 was tested by exposing heterozygote hBD3-Tg and DsRed-Tg mice to polyI:C (100μg, i.p.). The transgenic animals expressing hBD3 demonstrated an increased level of both Interferon-β and TNF-α (Fig 3B). In addition, homozygous transgenic hBD3 mice had a significantly raised basal level of Interferon-β compared to controls (Fig 3C). A number of additional cytokines were investigated, including IL-6 and CXCL10. hBD3-Tg mice treated with polyI:C showed a trend towards enhanced IL-6 induction and reduced CXCL10 induction compared to controls, however these data did not reach significance (see S2 Fig).
HMW polyI:C enters macrophages without a transfection reagent, however complexing with the cationic lipid, Lipofectamine 2000, enhances the amount of polyI:C entering cells by endocytosis, allowing more ligand to be available to both endosomal and cytoplasmic receptors [33]. We hypothesised that hBD3, being positively charged, may form complexes with polyI:C, enhancing uptake in a similar way to lipofection. To investigate this, we compared cellular uptake of FITC-labelled polyI:C in the presence of Lipofectamine 2000 and/or hBD3, by flow cytometry. Compared to polyI:C alone, the addition of hBD3 significantly increased the amount of labelled ligand per cell (Fig 4A). Lipofectamine alone also significantly augmented the amount of polyI:C entering cells, with the amount of FITC-polyI:C uptake in the presence of Lipofectamine not differing significantly from the level of FITC-polyI:C in hBD3 treated cells. Uptake in the presence of both lipofectamine and hBD3 was similar to either hBD3 or lipofectamine alone, so no additive effect was apparent.
We further compared the effects of lipofectamine and hBD3 on the production of Interferon-β and CXCL10 by polyI:C. Transfecting polyI:C with lipofectamine into wildtype BMDM resulted in enhanced CXCL10 production (Fig 4B). In contrast hBD3 inhibited CXCL10 induction by polyI:C. In the presence of a combination of lipofectamine and hBD3, CXCL10 production was still inhibited compared to polyI:C and lipofectamine, but to a lesser extent than hBD3 alone. In contrast, hBD3 and hBD3/lipofectamine significantly increased Interferon-β production in response to polyI:C in BMDM (Fig 4C). Lipofectamine and polyI:C did not demonstrate an enhanced Interferon-β response and lipofectamine decreased the amount of Interferon-β produced in response to polyI:C.
To visualise the uptake of polyI:C by BMDM we used FITC-labelled polyI:C (green fluorescence). Stronger fluorescence, was observed in cells when either hBD3 or lipofectamine was also present, supporting the flow cytometry data in Fig 4A showing both hBD3 and lipofectamine increase the entry of polyI:C into the cells. HBD3 increased the intensity of cyloplasmic polyI:C staining compared to either lipofectamine or polyI:C alone. In the presence of lipofectamine, polyI:C had a more punctate pattern within the cell, consistent with endosomal location. In the presence of hBD3, polyI:C appeared to have increased cytoplasmic distribution in addition to foci of staining. These results show that lipofectamine and hBD3, both enhance the amount of polyI:C that enters the cell but the different signalling responses triggered by each, suggests that hBD3 is directing polyI:C towards cellular compartments, that are not those targeted by the polyI:C-lipofectamine complexes.
To further investigate the mechanism of the hBD3 effect on polyI:C we carried out fluorescence co-localisation studies using TAMRA-labelled hBD3, FITC-labelled polyI:C and immunohistochemistry for the early endosomal marker (EEA1). TAMRA-labelled hBD3 entered the cells by 10mins (Fig 5Aii), whereas FITC-polyI:C was not visible in the cells until 30 minutes (Fig 5Biii). FITC-polyI:C added to the cells without hBD3 (Fig 5B and 5C), demonstrated FITC fluorescence in punctate regions which stained to some extent with the early endosome marker EEA1-1 antibody (Pearson coefficient, r = 0.615; Fig 5Civ). HBD3 alone (Fig 5A and 5D) also localised to discrete regions however these did not co-localise with EEA1 staining (Pearson coefficient, r = 0.205; Fig 5Div). Adding TAMRA-hBD3 and FITC-polyI:C together onto BMDM, again showed more polyI:C entering the cell in the presence of hBD3 (Fig 5Biii, 5Cii vs 5Eiii). However we could see no evidence for co-localisation of hBD3 and polyI:C (Pearson coefficient, r = 0.073; Fig 5Evii). PolyI:C appeared more cytoplasmic in the presence of hBD3 than when it entered the cell alone and the co-localisation coefficient with EEA1-1 staining was lower than in the absence of hBD3. (Pearson coefficient, r = 0.562 compared to r = 0.615 Fig 5Civ versus Fig 5Evi). In the presence of polyI:C, hBD3 remained localised in the discrete foci similar to those seen with hBD3 treatment alone, and again hBD3 did not co-localise with EEA1 (Pearson coefficient r = 0.05; Fig 5Eviii). This implies that hBD3, alters the localisation of polyI:C allowing less polyI:C to access the early endosome. Although nucleic acids can induce type I Interferon by activation of TLR signalling [34] in the endosome, Interferon-β and IL-6 can also be produced by activation of cytoplasmic receptors.
To examine the consequences of the altered localisation of polyI:C by hBD3 and to determine the signalling pathways responsible for production of Interferon-β and CXCL10, we used cells from knockout mice specific for the two main pathways known to be involved. Firstly, we determined the Interferon-β response of BMDM exposed to polyI:C in the absence of the TLR3 adaptor TICAM1 and found that this was reduced in Ticam1-/- cells compared to wild type BMDM, indicating that the majority of the polyI:C effect on Interferon-β was not TICAM1 dependent. In the presence of hBD3, the polyI:C-induced Interferon-β response in Ticam1-/- BMDM was still significantly enhanced compared to polyI:C alone (Fig 6A), which suggests that hBD3 is not enhancing signalling through the TLR3/TICAM1 pathway. Conversely, in the absence of MAVS, where Interferon production is only through TLR3/TICAM1 signalling, a smaller amount of Interferon-β was induced by polyI:C. In the presence of hBD3 this Interferon-β induction was significantly inhibited (Fig 6A), suggesting that hBD3 inhibits TLR3/TICAM1 signalling. Treatment of BMDM from Ticam1(-/-)Mavs(-/-) double knockouts with polyI:C did not induce Interferon-β indicating that all the Interferon-β response to polyI:C in BMDM is dependent only on these two pathways (S4A Fig).
In contrast to Interferon-β, the induction of CXCL10 by polyI:C in wildtype BMDM was significantly reduced in the presence of hBD3. In Ticam1(-/-) BMDM the response to polyI:C was eliminated (Fig 6B) demonstrating that production of this cytokine is controlled primarily by TLR3/TICAM1 activation. Supporting this finding, polyI:C-induced CXCL10 in Mavs(-/-) BMDM (a TLR3-TICAM1 dependent response) was not significantly different to wildtype cells indicating that MAVS does not influence CXCL10 production and the production of CXCL10 in response to polyI:C in Mavs(-/-) cells was still significantly inhibited by hBD3 (Fig 6B).
Both RIGI and MDA5 are upstream of MAVS, so to dissect the effects of hBD3 on MAVS signalling we used the specific RIGI ligand, 5’ triphosphate double stranded RNA (5’ppp) and Mda5(-/-) mice [35]. Treatment of wildtype BMDM with 5’ppp resulted in a significant increase in Interferon-β, TNFα and CXCL10, and as expected with this ligand, these responses were dependent on the presence of MAVS (Fig 7A). The addition of hBD3 to the macrophages shortly after transfection of 5’ppp, resulted in a significant decrease in cytokine and Interferon-β production (Fig 7A), demonstrating that RIGI responses to 5’ppp are inhibited by the presence of hBD3. In contrast however, macrophages from Mda5(-/-) mice, revealed that Interferon-β induced by polyI:C was reduced compared to wildtype cells, indicating that MDA5 signalling was responsible for the majority of the polyI:C induced Interferon-β production (Fig 7B). This residual response which is likely to be TLR3-TICAM1 signalling, was not amplified in the presence of hBD3, indicating that MDA5 is required for the hBD3 enhancing effects on polyI:C-induced Interferon-β.
We show here that hBD3 enhances the production of various cytokines in response to polyI:C (TNF-α, IL6 in mouse cells and TNF-α, IL8 in human cells) and Interferon-β in both human and mouse cells. We demonstrate that this effect is dependent on the correct, disulfide stabilised structure of hBD3. Importantly, we show that this response is not specific to our synthetic hBD3 peptide, as transgenic animals expressing hBD3 from a genomic transgene, also demonstrate an increased type I Interferon response when injected with polyI:C. It was important to reproduce the results shown with our synthetic hBD3 peptide in this transgenic system to validate the augmenting effects of hBD3 as it has been demonstrated that synthetic peptides may be incorrectly folded giving misleading results [7]. The synthetic hBD3 peptide we use here, gives equivalent functional results to hBD3 produced in vivo.
When we dissect the pathways known to be activated by polyI:C, it is evident that although signalling through the RLR co-adaptor MAVS mediated pathway is up-regulated in the presence of hBD3, signalling through the endosomally located TLR3/TICAM1 pathway is suppressed. The inhibition of the TICAM1-mediated signalling pathway supports our findings with LPS, where we previously reported that hBD3-mediated inhibition of LPS signalling through TLR3/TICAM1 was lost in Ticam1 KO mice and could be inhibited by hBD3 cDNA in HEK293 cells [10].
In our experiments here, we use HMW (long) polyI:C which activates MDA5 with or without transfection and show that in BMDM, MDA5 is predominantly responsible for the Interferon-β response. Interestingly, it has been demonstrated previously that LMW and HMW polyI:C are recognised by different receptors with HMW polyI:C being recognised by MDA5 and LMW polyI:C by RIGI [24]. In addition large RNA structures generated by viral replication are believed to be important in effectively triggering MDA5 [36]. Recently Zou et al reported that forced delivery of HMW polyI:C to the cytoplasm with transfection was not necessary for RLR stimulation in GM-DCs and CD11bhiCD24lo DCs, although LMW polyI:C required transfection to interact with MDA5 [23]. We show here that similarly to the DCs, BMDM also take up HMW polyI:C effectively without transfection and activate MDA5. HBD3 exacerbates the signalling through MDA5. These researchers also report release of endosomal Cathepsin D and induction of necrosis by the activation of MDA5. We see no evidence of cell death (using an LDH assay, see S5 Fig) using polyI:C at 10μg/ml, but this is 5-fold less than that used by Zou et al [23].
Cationic lipids such as Lipofectamine are known to cause endosomal localisation [33]. Although hBD3 is a cationic peptide we do not observe similar outcomes when we compare the effects of hBD3 with the actions of lipofectamine on polyI:C stimulation of macrophages. In wild type cells stimulated with polyI:C, hBD3 increased Interferon-β and decreased CXCL10 production. In contrast, polyI:C and lipofectamine increased CXCL10 production and decreased Interferon-β. This effect is likely to be due to the change in localisation of polyI:C as a result of being in the presence of lipofectamine or hBD3 (see Fig 8). Our immunostaining demonstrates that hBD3 encouraged polyI:C to be more cytoplasmic compared to lipofectamine which causes increased endosomal localisation. Despite the likely electrostatic interaction of polyI:C and the highly charged hBD3 (+11) in the cell, our cellular uptake experiments using fluorescently labelled derivatives revealed that at 30min after addition to the cells hBD3 and polyI:C do not co-localise appreciably. It is possible that initially they may have interacted, allowing polyI:C to access the cell as hBD3 has been described as having cell penetrating properties [37]. In the presence of hBD3, polyI:C does not localise to the early endosome so presumably the cytoplasmic location of the ligand allows an increase in interaction with MDA5. It is possible that cationic hBD3 complexed with polyI:C, enables the ligand to rapidly escape the acidic endolysosome, perhaps in a similar way to pH-dependent fusogenic peptides that assist macromolecules to access the cytoplasm [33]. However we see polyI:C localised to the early endosome (by EEA1 positive immunostaining) in the presence of hBD3 implying that the structure of the early endosome is not disrupted by the presence of hBD3.
The main consequence of increased MDA5 signalling in response to polyI:C is increased IFN-β. This increase is additive when lipofectamine is also present (S4B Fig). However no increase is observed in the absence of MAVS which implies that the exacerbated response requires the RLR. It may be that lipofectamine complexes the polyI:C to create higher order structures that activate MDA5 more optimally when [36] hBD3 increases its cytoplasmic localisation.
MDA5 is important in relation to autoimmunity and mutations that inactivate or reduce expression of MDA5 have been shown to protect individuals from type I diabetes mellitus risk [38, 39]. In addition, a mutant form of MDA5 in mice that is active without viral infection induces a type I Interferon-dependent autoimmunity with similarities to lupus [25]. However Interferon-β has also been described as a protector against some types of inflammation such as dextran sodium sulphate induced colitis [40–42] and this protection can be observed in mice that express increased Interferon-β in response to dsRNA-producing intestinal bacteria [40–42].
Increased copy number of the cluster of β-defensins on human chromosome 8 is linked to increased incidence of the autoimmune disease psoriasis and effective treatment of psoriasis with UV irradiation is linked to suppression of type I Interferon and Th17 cells [43]. In addition the most effective treatments currently for psoriasis are monoclonal antibodies directed against IL-17 cytokine production or IL-12p40 (the cytokine subunit common to both IL-12 and IL-23) [18, 44]. It is thus potentially highly significant that we see strong elevation of IL12-p40 subunit in mice injected with both hBD3 and polyI:C. It is also possible that the other defensins on the CNV cluster may also demonstrate this effect and we have shown that hBD2 also heightens the response of mouse BMDM to polyI:C (S6 Fig).
It has recently been shown that human pDC produce Interferon-α in response to self or other DNA through TLR9 [9, 19]. We show here that macrophages increase the Interferon-β response to polyI:C in the presence of hBD3 through MDA5. Production of type I Interferons is normally the consequence of pattern recognition receptors binding virally produced nucleic acid pathogen associated molecular patterns (PAMP), such as double stranded RNA (dsRNA) produced during viral replication. Psoriasis has been reported to be exacerbated by the use of Interferon-α as therapy for Hepatitis C [45] and by Interferon-β therapy for multiple sclerosis [46]. Investigation of the psoriasis transcriptome has identified an increase in RIG-I like receptors (RLR), which also recognise viral PAMP leading to type 1 Interferon production [47]. During a pathogen infection, hBD3 expression increases [32, 48], and hBD3 has been shown to demonstrate potent anti-viral action in vitro [49]. Expression in pDC, monocytes and epithelial cells of the non-copy number variable defensin hBD1 has been shown to increase in response to virus exposure, while expression of the murine orthologue of DEFB103 (Defb14) increases in response to polyI:C [50, 51]. MDA5 is specialised for protecting mice against infection with various RNA viruses including picornaviruses (including Theiler’s and Mengo viruses and Encephalomyocarditis virus (EMCV)) as well as paramyxovirus and Norovirus [52, 53]. MDA5 knockout mice are highly susceptible to EMCV [35, 54]. During infection, rapid killing, detection and innate response are essential; therefore in this regard, high hBD3 copy number and potentiation of PRR may be beneficial. However an undesirable effect of increased copy number of the defensin cluster (and concomitant increase in expression of defensin peptides) may be over stimulation of PRRs leading to exuberant production of type I interferons. This double edged sword may provide protection against pathogens in the short term, but in the longer term contribute to the development of psoriasis in individuals with an increased copy number of the β-defensin cluster.
Animal studies were covered by Project License (PPL 60/4475), granted by the UK Home Office under the Animal Scientific Procedures Act 1986, and locally approved by the University of Edinburgh Ethical Review Committee. Human venous blood was collected with written patient consent from healthy volunteers according to Lothian Research Ethics Committee approvals ♯08/S1103/38.
Ultra pure Lipopolysaccharide (LPS) from E. coli 0111:B4, Lipoteichoic acid (LTA), Pam3CSK4, FSL-1, HKLM, polyI:C (HMW), FlTC-labelled polyI:C (HMW), R848, CpG and 5’ triphosphate double stranded RNA (5’ ppp-dsRNA) were purchased from InvivoGen (San Diego, USA), M-CSF, ELISA DuoSets and IFNβ antibodies were obtained from R&D Systems, (Abington, UK). Fluorescently labelled secondary antibodies were purchased from Jackson ImmunoResearch Laboratories (PA, USA). hBD3 (GIINTLQKYYCRVRGGRCAVLSCLPKEEQIGKCSTRGRKCCRRKK) was from Peptides International, and cys-ser hBD3 and cys-ser-TAMRA hBD3 were from Almac (Almac Group Ltd, Craigavon, UK).
The peptide was produced on a CEM Liberty1 microwave peptide synthesizer using standard Fmoc (fluorenylmethyloxcarbonyl chloride) chemistry. Amino acids were purchased from AAPPTec and were assembled on H-Rink amide ChemMatrix resin. Fmoc protecting groups were removed using 20% piperidine and 0.1 M hydroxybenzotriazole (HOBt) in dimethylformamide (DMF). Amino acids were coupled using 5 molar equivalents of diisopropylcarbodiimide (DIC) and 10 molar equivalents of HOBt in DMF. The N-terminal labelling of peptides with fluorescent dye was performed on resin-bound peptide using 4 equivalents of 5-(and-6)-carboxytetramethylrhodamine, succinimidyl ester (5(6)-TAMRA SE purchased from Biotium) and 6 equivalents of diisopropylethylamine (DIEA) in DMF, incubating for 2 hours. The peptide resin was then rinsed with DMF to remove excess fluorescent dye, washed with dichloromethane (DCM), and dried. Cleavage of the peptide from resin was performed in a trifluoroacetic acid (TFA)/ triisopropylsilane (TIS)/ 1, 2-ethanedithiol (EDT)/ phenol (90:4:4:2) mixture for 90 min. The resin was filtered and the filtrate was added to 90 mL of cold dry diethyl ether. The precipitate was collected by centrifugation and the diethyl ether was discarded. The peptide was purified on a C18 reverse phase HPLC column and the correct molecular weight was confirmed by ESI-MS. Oxidative folding was achieved in folding buffer (0.5–1.0 M guanidine hydrochloride (GuHCl), 0.1 M Tris, 1 mM glutathione (GSH), 0.1 mM oxidized glutathione (GSSG), pH 8.5) at a peptide concentration of 0.1 mg/mL and stirred for 48 hours. Folding was monitored by reverse phase HPLC, which revealed one major species that was used in subsequent experiments. Folding procedures were developed to give the correct HBD3 structure, as verified previously by nuclear magnetic resonance structure determination (Nix et al., 2013). The folded products were purified on a C18 reverse phase HPLC column and identified as fully oxidized peptides by ESI-MS. Quantitative concentrations were determined with amino acid analysis at the molecular structure facility at UC Davis.
The Mavs−/− (Cardif, Ips-1, or Visa) mutant line was generated by the Tschopp group in Lausanne. It is homozygous-viable null mutant in the C57B6J background. Ticam1 (Trif) -/- mice and MDA5-/- mice were used with the generous permission of Professor Shizuo Akira (Osaka University, Japan)[35, 55]. hBD3-Tg and DsRed-Tg were constructed by electroporation of ScaI linearised parental vector, which has a 1.5 Kb genomic fragment of the entire hBD3 gene DEFB103 including exons 1 and 2 and the intervening intron cloned into pTLC plasmid (a kind gift from Josh Brinkman, Danish Stem Cell Centre, DanStem, University of Copenhagen) using DEFB103 primers with 5' NheI and PacI sites for cloning, into ES cells. Cells that strongly expressed DsRed were selected by FACS and used to make DsRed-Tg mice. Transient cre treatment of these cells produced DsRed negative cells due to lox -mediated excision of the DsRed gene and allowed expression of hBD3 and EGFP (see S1 Fig). The ES cells were made into macrophages using the method of (Yeung et al 2015) which were strong expressors of EGFP. Clones before and after CRE treatment were injected into blastocysts at the University of Edinburgh MRC Evans Building, Transgenic Unit.
THP-1 cells were grown in RPMI with 10% fetal bovine serum (FBS) and differentiated into macrophages by the addition of 150nM PMA, 2 days before treatment. Mouse primary macrophages (BMDM) were generated from femur bone marrow grown for 8 days in DMEM with 10% fetal bovine serum and 20ng/ml M-CSF. Cells, seeded at 2 x 105 into 48 well plates, were grown without M-CSF for 24 hours prior to treatment. Replicate experiments were done with separate primary cell preparations from at least 3 mice for each experiment.
Mouse BMDM, seeded at 2 x 105 cells into 48 well plates were treated in serum free media with TLR or RIGI agonists at the concentrations indicated in the figures, in the presence or absence of hBD3 (5μg/ml). After an 18 hour incubation at 37°C, 5% CO2, TNF-α, IL-6, CXCL10 and Interferon-β were measured using mouse DuoSet ELISA (R&D Systems). Statistical significance was determined by an unpaired t-test using GraphPad software, with values expressed as mean +/- SEM and p < 0.05 considered significant.
Human venous blood was collected from healthy volunteers according to Lothian Research Ethics Committee approval, using sodium citrate anticoagulant (Phoenix Pharma, Gloucester, UK), and cells were separated by Dextran sedimentation, followed by discontinuous, isotonic Percoll gradient centrifugation as previously described [56]. PBMC were incubated at 4×106/mL in IMDM (PAA Laboratories, Somerset, UK) at 37°C, 5% CO2, for 1 h. Non-adherent cells were removed and adherent monocytes cultured for 6 days in IMDM with 10% autologous serum to generate monocyte-derived Mφ. PolyI:C was applied in concentrations indicated and cells harvested at 18 hours. RNA was isolated and human Interferon-β gene expression was measured using the Applied Biosystems Taqman Gene Expression Assay following the manufacturer’s instructions.
Cytospins of mouse BMDM were fixed in 4% PFA, washed, then blocked for 2 hours at room temperature (RT) in 10% donkey serum (Sigma, Poole, UK) in PBST (0.1% Tween in PBS). Slides were then incubated overnight at 4°C with hBD3 antibody (1:200) (DSHB, Iowa University, USA). After washing, slides were incubated with TxRd labelled anti-mouse antibody (1:400) for 2 hours at RT, further washed then mounted with Vectashield containing 1μg/ml DAPI. hBD3 immunostaining was visualised using a Zeiss Axioplan 2 microscope (Carl Zeiss UK Ltd., Welwyn Garden City, UK) equipped with Ludl filter wheel (Ludl Electronic Products Ltd, Hawthorne, NY, USA) and Chroma 83000 triple bandpass filter set (Chroma Technology Corp, Rockingham, VT, USA). In-house scripts written for IPLab (Scanalytics Corp, Fairfax VA, USA) were employed for image capture and image processing.
BMDM (2 x 104 cells/well) were cultured overnight on 8 well glass chamber slides (Nunc Inc, IL, USA) in DMEM with 10% FCS. Cells were treated with FITC-polyI:C (2.5μg/ml) in Optimem media (Life Technologies) with or without lipofectamine 2000 (Invitrogen, at 1:100 dilution, 10μl/ml) in the presence or absence of hBD3 or TAMRA-labelled HBD3 (0.5μg/ml). After 2, 10, 15 or 30 mins cells were washed in PBS and fixed in 4% PFA. For early endosome staining, cells were blocked with 10% donkey serum and incubated with anti-EEA11 antibody (Abcam, UK) for 1 hr at RT, then 30 min with donkey anti-rabbit Cy5. Cells were imaged using a 40x 1.3NA oil immersion objective on a Nikon A1R confocal microscope using Nikon Nis-Elements AR software for image acquisition (Nikon Instruments Europe, Netherlands). Image analysis was carried out in ImageJ (http://imagej.nih.gov/ij/). Pearsons coefficients were calculated using the JaCoP ImageJ plugin [57].
Male C57 Black/6 mice (6–8 weeks old) and hBD3 transgenic male mice (8 weeks) were injected intraperinoneally (i.p.) with polyI:C (100μg/mouse) in 200μl of physiological saline. Half of the C57 Black/6 mice also received an i.p injection of synthetic hBD3 (20μg/mouse). After 4 hr, mice were killed by cervical dislocation, exsanguinated and serum levels of TNFα and Interferon-β measured by ELISA.
Mouse BMDM plated at 1 x 106 cells on a 6-well plate were treated with FITC-pI:C (10μg/ml) in the presence of lipofectamine 2000 (at 1:100 dilution, 10μl/ml) or hBD3 (5μg/ml). For treatment with polyI:C in the presence of lipofectamine 2000, media was replaced with optimem before the addition of polyI:C complexed with lipofectamine 2000 (L-pI:C) and addition of hBD3 was delayed for 5 min to avoid direct interaction with L-pI:C complexes. After 18 hours cells were gently washed and gently removed from the dish into PBS containing 1% BSA. Fluorescence was measured with a BD FACSARIAII SORP (BD Biosciences, Oxford, UK), using a 640nm laser (670/14nm bandpass filter). Data analysis was done using FlowJo Version 7.5.5 (Treestar Inc, Olten, Switzerland). This experiment was carried out on 3 different preparations of BMDMs
|
10.1371/journal.ppat.1007104 | Tracking KLRC2 (NKG2C)+ memory-like NK cells in SIV+ and rhCMV+ rhesus macaques | Natural killer (NK) cells classically typify the nonspecific effector arm of the innate immune system, but have recently been shown to possess memory-like properties against multiple viral infections, most notably CMV. Expression of the activating receptor NKG2C is elevated on human NK cells in response to infection with CMV as well as HIV, and may delineate cells with memory and memory-like functions. A better understanding of how NKG2C+ NK cells specifically respond to these pathogens could be significantly advanced using nonhuman primate (NHP) models but, to date, it has not been possible to distinguish NKG2C from its inhibitory counterpart, NKG2A, in NHP because of unfaithful antibody cross-reactivity. Using novel RNA-based flow cytometry, we identify for the first time true memory NKG2C+ NK cells in NHP by gene expression (KLRC2), and show that these cells have elevated frequencies and diversify their functional repertoire specifically in response to rhCMV and SIV infections.
| Natural killer (NK) cells are a crucial component of the early innate immune response, and although NK cell responses have been thought be only non-specific, recent evidence suggests that NK cells are capable of expanding with some specificity, indicative of a memory-like adaptive response. The activating receptor NKG2C has been one cell surface protein associated with this memory-like NK cell expansion in the context of CMV and HIV infection in humans, yet very little is known about NKG2C+ NK cells in non-human primate (NHP) animal models. This is predominantly because there are no antibodies that can distinguish NKG2C from other NKG2 family molecules in NHP. Because vaccine and cure-related studies for HIV rely heavily on NHP models, this is a significant impediment towards understanding an NK cell population that may possibly improve responses to HIV. In this paper we present a solution, by adapting a technique whereby mRNA specific to NKG2C and NKG2A (KLRC2 and KLRC1, respectively) is fluorescently labeled while the cell is simultaneously stained using traditional flow cytometry, and provide a first-ever characterization of NKG2C+ NK cells in NHP. Further, we show that NKG2C+ NK cells expand in a memory-like fashion following rhCMV and SIV infections.
| Although NK cells have traditionally been thought to be innate immune cells that lack the antigen-specificity seen in the adaptive immune system, NK cells have very recently been reported to possess memory and memory-like functions [1–8]. Though this area of investigation is currently developing, subpopulations of NK cells that express NKG2C (CD159C) in humans or Ly49H and Ly49P in mice mobilize in response to CMV infection [9–13]. While this phenomenon has been described in human and murine studies, because of technical limitations it has not yet been possible to examine memory and memory-like NKG2C+ NK cells in NHP models. This is predominantly attributed to the high degree of homology in NHP between the extracellular domains of two NKG2 isoforms, activating NKG2C and inhibitory NKG2A –making the two indistinguishable via currently available antibodies and standard measurements [14, 15]. NHP models are crucial to multiple areas of medical research, including HIV and CMV infectious disease study and transplant biology [16–18] since the murine system does not always approximate human immunology. As such, the inability to study NKG2C+ memory NK cells in NHP models remains a major research deficit.
NKG2C and NKG2A both belong to the C-type lectin family of NK cell receptors. NKG2C recruits the adaptor protein DAP12, which has an ITAM (immunoreceptor tyrosine-based activation motif), and NKG2A has two ITIM (immunoreceptor tyrosine-based inhibitory motif) domains, which lead to recruitment of phosphatases, and downregulation of signaling [19, 20]. Because these two proteins act in opposition to each other, it is crucial to discriminate between cells that express either protein in order to more accurately determine what role these cells play during infection. As a result, we aimed to utilize RNA hybridization technology recently adapted for flow cytometry (PrimeFlow) to label the gene transcripts of rhesus macaque NKG2A and NKG2C (KLRC1 and KLRC2, respectively), taking advantage of several nucleotide differences between the two transcripts, in order to distinguish cells that transcribed these isoforms. This approach should allow simultaneous detection of surface and intracellular proteins as well as gene transcript levels with a single-cell resolution using polychromatic flow cytometry. In addition to differentiating between KLRC1+ and KLRC2+ NK cells, this technology should allow evaluation of NK cell population diversity, including memory cells, in the context of “normal” CMV infection, in chronic SIV infection, and in CMV-negative specific pathogen free (SPF) rhesus macaques. Understanding how KLRC1±KLRC2± NK cells function in the context of infection will help improve our basic understanding of NK cell biology, potentially inform preclinical HIV vaccine or cure studies relying on macaque models, and provide a significant technological advance to the study of memory NK cells in primates.
All animals were housed at the Tulane Primate Research Center (TNPRC) or Biomere (Worcester, MA). All study blood samplings were reviewed and approved by the Tulane University Institutional Animal Care and Use Committee or the Biomere Institutional Animal Care and Use Committee under protocol numbers 16–08 and 17–02. All animal housing and studies were carried out in accordance with recommendations detailed in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health with recommendations of the Weatherall report; “The use of non-human primates in research”. AAALAC numbers for TNPRC and Biomere– 00594 and 1152, respectively. Animals were fed standard monkey chow diet supplemented daily with fruit and vegetables and water ad libitum. Social enrichment was delivered and overseen by veterinary staff and overall animal health was monitored daily. Animals showing significant signs of weight loss, disease or distress were evaluated clinically and then provided dietary supplementation, analgesics and/or therapeutics as necessary. No animals were euthanized as part of this research.
Thirty Indian rhesus macaques were analyzed in this study: ten specific pathogen free (SPF) animals (rhCMV- and SIV-negative), twelve otherwise naïve animals that were naturally infected with rhCMV (rhCMV+), and eight chronically infected with SIVmac251 (all of which rhCMV+). SPF animals and age-matched non-SPF/rhCMV+ macaques were housed at the Tulane National Primate Research Center (TNPRC). SIV-infected macaques and additional rhCMV+ animals were housed at Biomere. All animals were colony housed until on study and then infected animals were housed under BSL2 conditions.
Whole blood was collected into EDTA-treated tubes. Peripheral blood mononuclear cells (PBMCs) were isolated by density-gradient centrifugation layered over 100% Ficoll. Cell aliquots were immediately analyzed or cryopreserved in 90% FBS, 10% DMSO (Sigma) and stored in liquid nitrogen vapor.
PBMCs were thawed and rested for 12h in R10 media at 37°C prior to surface and intracellular staining followed by RNA-Flow hybridization using the manufacturer’s recommended protocol (PrimeFlow, Affymetrix, Santa Clara, CA) with the antibodies detailed in the flow cytometry section below, and with rhesus macaque-specific KLRC1 and KLRC2 probesets. Rhesus-specific probesets were custom designed with the assistance of Affymetrix (Santa Clara, CA) specifically for this project to target rhesus KLRC1 and KLRC2. Target probes sequences for KLRC1 and KLRC2 are shown in S1 and S2 Figs, as are ‘blocking probe’ sequences used to prevent nonspecific binding. The blocking probes were designed in order to avoid amplifying/detecting undesired NKG2 homologues. Blocking probes do not have the ability to form branched DNA structures which hybridize to the label probe fluorophores as opposed to the target probes which are able to hybridize to label probe fluorophores. Both target probes and blocking probes were simultaneously added at the target probe hybridization step as per the manufacturer’s protocol. Probesets were labeled with Alexa-488 (KLRC1) and Alexa-647 (KLRC2) fluorophores by Affymetrix (Santa Clara, CA). All KLRC1 and KLRC2 gates were determined for each sample by comparing the samples stained with all antibodies and probesets with samples only stained with antibodies (no probeset control).
All antibodies used were purchased from BD Biosciences unless specified otherwise. For phenotypic panels antibodies against the following cell antigens were used: CD2 (RPA2.10), CD3 (SP34.2), CD337 (p30-15), CD14 (MϕP9), CD20 (L27), CD16 (3G8), CD56 (NCAM16.2), HLA-DR (G46-6), CD8α (SK1), KIR2D (NKVFS, Miltenyi [this antibody recognizes KIR3D in NHP as shown by Pomplun, N. et al. [21]]), CD159a/c (Z199, Beckman Coulter), CD366 (F38-2E2, Biolegend). Additionally, antibodies used for functional assays included TNF-α (MAb11), IFN-γ (B27), CD107a (H4A3). Flow cytometry data was acquired on a LSRII (BD Biosciences, La Jolla, CA) and analyzed with FlowJo software (version 10.2, Tree Star, Ashland, OR). t-SNE (t-distributed stochastic neighbor embedding) was carried out using the t-SNE feature in FlowJo using 1000 iterations and a perplexity of 20.
PBMCs from animals were prepared at 37°C and cells were cultured with Golgi Plug and Golgi Stop (BD Biosciences, concentrations as recommended by manufacturer), and PMA (3.3μg/mL, Sigma) and Ionomycin (5μg/mL, Sigma) or with unlabeled anti-CD16 (3G8, 20μg/mL) and cross-linked with F(ab’)2 (20μg/mL, Jackson Immunoreserach, West Grove, PA) for 14h in R10 media (RPMI + 10% FBS + 2% PenStrep (Gibco)).
Rhesus plasma was assessed for rhCMV-specific IgG by a previously reported rhCMV UCD52 whole virion ELISA [22]. After plates were coated with 4,400 PFU/mL of rhCMV UCD52 virus, the previously reported procedure was followed. The positivity threshold for detectable antibody levels was set to equal twice the OD of a rhCMV-seronegative plasma control at the starting dilution (1:30).
Statistical and graphing analyses were performed with GraphPad Prism 7.0 software (GraphPad Software, La Jolla, CA). Nonparametric Mann-Whitney U tests and Wilcoxon tests were used where indicated, and a p-value of p < 0.05 was considered to be statistically significant.
Total NK cells were identified among PBMC in rhesus macaques using traditional phenotypes optimized by our laboratory [5, 23–27]: CD3-CD14-CD20-NKG2A+ (Fig 1A). Unsurprisingly, the anti-NKG2A antibody was unable to distinguish between NKG2A and NKG2C, as has been previously shown by the Letvin lab whereby they showed that anti-human NKG2A antibodies were cross-reactive with four NKG2C alleles [14]. As a result we now classify bulk cells that are positive for this antibody as NKG2AC+ NK cells. Using RNA-Flow (see Methods) we next identified within NKG2AC+ NK cell populations those cells that expressed transcripts of the genes coding for NKG2A and NKG2C (KLRC1 and KLRC2, respectively), and accurately quantified the frequency of NK cells expressing one or both of these genes (Fig 1A–1C). Interestingly, absolute frequencies of both KLRC1+ and KLRC2+ NK cells (Fig 1B) were lower in SPF animals compared to either rhCMV+ or SIV-infected macaques—(KLRC1) 0.18%, 0.29%, and 0.68% of lymphocytes in SPF, rhCMV+, and SIV+ animals respectively; and (KLRC2) 0.31%, 2.55%, and 2.04% of lymphocytes in SPF, rhCMV+, and SIV+ animals respectively. These data demonstrate that while NK cells are less frequent in SPF animals in general, following rhCMV infection a less than 2-fold non-significant increase occurs in KLRC1+ NK cells, but the increase in KLRC2+ NK cells is 12-fold. This dramatic observation is congruent with other findings in human research, which show elevation of NKG2C+ NK cells specifically following CMV infection [9–11, 28]. Examining the frequency of KLRC1+ and KLRC2+ NK cells relative to the total NK cells population (Fig 1C) also revealed that in rhCMV+ and SIV-infected macaques there was an obvious reduction in KLRC1+ NK cells in lieu of KLRC2+ NK cells, but surprisingly, expression of both KLRC1 and KLRC2 remained high in NK cells from SPF animals.
To further clarify our findings, we re-optimized our technical approach to measure both KLRC1 and KLRC2 simultaneously. Using the gating strategy shown in Fig 2A we were able to clearly distinguish four distinct NK cell populations by expression of KLRC1 and KLRC2. Strikingly, this analysis revealed that in the absence of rhCMV infection, a KLRC1+KLRC2+ population is dominant (Fig 2B). In contrast, in rhCMV+ animals (rhCMV+ and SIV+ groups) the predominant population was single-positive KLRC1-KLRC2+. Consistently, KLRC1+KLRC2- and KLRC1-KLRC2- represented minority populations among all animal groups and could represent precursor or aberrant NK cells outside the normal NK cell repertoire. While the presence of the KLRC1-KLRC2- population was surprising, it must be noted that the Z199 clone that detects human NKG2A and rhesus macaque NKG2A and NKG2C is promiscuous and could be identifying minor NKG2 isoforms[14]. The extreme specificity and blocking probes used in the RNA-based flow cytometric technique make it highly unlikely to have non-specific signals (S1 and S2 Figs). The presence of a double-negative population is more likely resulting from some samples where mRNA levels being below the threshold of detection of this assay. Regardless, these findings point to the overall importance of this study which are now able to confirm true KLRC2+ (NKG2C) NK cells in macaques which have only been incompletely described previously. Also, consistent with observations in humans we also find that both KLRC1 and KLRC2 are expressed on minor populations of T cells (S3 Fig). Finally, we can determine that the observation that rhCMV+ and SIV+ animals have higher relative and absolute frequencies of KLRC1-KLRC2+ compared to KLRC1+KLRC2+ NK cells is likely analogous to the memory and memory-like functions observed in human CMV infection [10, 12, 13], whereby prior to CMV exposure both inhibitory NKG2A and activating NKG2C are expressed, but NKG2A is downregulated following infection. Further supporting the notion that CMV specifically expands NKG2C+ NK cells, we found a significant positive correlation between increasing KLRC1-KLRC2 (NKG2C)+ NK cells and rhCMV-binding IgG, as a surrogate indicator of virus replication (Fig 3D). Concurrently there was a significant negative correlation between frequencies of KLRC1+KLRC2± NK cells and increasing rhCMV-specific IgG (Fig 3A and 3B). There was, however, no association between rhCMV-specific IgG and KLRC1-KLRC2- NK cells (Fig 3C). Interestingly, no correlation was found between SIV viral loads (median 3.00 x 106 virus copies/ml; range 5.66 x 104 to 3.30 x 107) and any of the NK cell subpopulations. Collectively, these data suggest that perturbations of KLRC1±KLRC2± NK cells is primarily driven by rhCMV status.
We next wanted to confirm phenotypically that the KLRC1 and KLRC2 definitions were indeed identifying NK cell subpopulations that are analogous to their human counterparts in which NKG2C+ NK cells are more activated and differentiated. Indeed FcRγIII receptor CD16 was higher in rhCMV+ and SIV-infected animals compared to SPF and was consistently higher on KLRC2+ NK cells. (Fig 2C, S1 Table). CD56 is typically expressed on most circulating NK cells in humans, but is expressed on only a small frequency of cytokine-producing or less differentiated NK cells in macaques [29]. Consistent with this notion, we found CD56 expression was higher in general on homeostatic KLRC1+KLRC2+ NK cells, but was poorly expressed on KLRC1-KLRC2+ NK cells expanded by rhCMV infection seen in the CMV+ and SIV+ groups. While this trend was not present in SPF animals we noted that expression of CD56 was generally higher in SPF samples as compared to the CMV+ and SIV+ groups, though it was not statistically significant. In general KIR expression is increased as NK cells differentiate and it was hypothesized that rhCMV or SIV infection are increasing activation and differentiation. Indeed expression of the common macaque KIR, KIR3D, was lower in all SPF animals and had the highest expression on KLRC1-KLRC2+ NK cells. CD2, which has been shown to synergize with NKG2C to promote adaptive NK cell functions [30], was also found to be increased on KLRC2+ cells in rhCMV+ macaques. Unfortunately, cross-reactive antibodies against the CD57 carbohydrate epitope, also associated with memory NK cell phenotypes, do not currently exist for monkeys and thus could not also be evaluated here. Nonetheless, these findings collectively suggest this population is generally more activated and differentiated and has a phenotype consistent with adaptive functions.
Since our analyses suggested rhCMV infection may be driving KLRC2+ NK cell expansion, we next used t-SNE to evaluate NK cell subpopulation clustering and diversity. NK cells from SPF animals clustered into two major groups–corresponding with KLRC1+KLRC2+ and KLRC1-KLRC2+ populations. In contrast, NK cells from rhCMV+ and SIV+ animals clustered into far more minor and distinct groups (Fig 4A). The phenotypic characteristics of these groups were also highly variable depending on infection status and subpopulation (Fig 4B). These data suggest that there is a greater diversity of NK cell subpopulations following infection with rhCMV or SIV as compared with the uninfected SPF group. These findings are in strong agreement with previous analyses by showing that human NK cell diversity increases following infection with HIV and other pathogens [31–33].
Next we wanted to examine whether there were any functional differences among each of the KLRC1+ and KLRC2+ populations. Mitogen stimulation revealed that all NK cell subsets from all animal groups were capable of surrogate cytotoxic (CD107a) and cytokine-based (IFN-γ, TNF-α) responses (Fig 5). Unfortunately, the number of positive events for the KLRC1+KLRC2- population were too few for us to reliably report from the functional assay. Nevertheless, the remaining quadrant populations were functional, with the KLRC1+KLRC2+ and KLRC1-KLRC2+ populations demonstrating the most robust responses. Interestingly, upon mitogen stimulation we observed that NK cells from SPF animals produced proportionately greater cytokines, which could be indicative of a more immature status as expected given the lack of virus exposure. Following stimulation through CD16 that could mimic ADCC, again all NK cell subpopulations were functionally competent. Perhaps even more obvious in this assay, NK cells from SPF animals favored cytokine production, whereas those from rhCMV+ animals were more adept for CD107a upregulation as a surrogate indicator of cytotoxicity (S4 Fig). These findings further suggest that CMV infection is necessary to activate and prime cytotoxic functions, particularly those dependent on antibodies and corroborates findings in humans mediated by NKG2C+ γ-chain deficient memory-like NK cells [13]. Indeed, the memory-like programming observed for rhCMV could be epigenetic in nature if not memory per se. Collectively, all NK subpopulations from SIV-infected animals were functionally responsive to mitogen, but were poorly responsive to CD16 cross-linking. These findings are well in-line with previous observations of NK cell dysfunction in HIV and SIV infections. In this paper we present a cross-sectional analysis of several infected and uninfected animals. Further studies need to be carried out in order to examine the kinetics of infection and how SIV or CMV may play a role in shaping the NK cell repertoire. Importantly, we also show that with this technique we are able to also examine and identify KLRC1 and KLRC2+ NK cells from peripheral lymphoid tissues such as the spleen and primary sites of infection such as the colon (S5 Fig). This will allow us to examine the role that these NK cell populations play in the earliest stages following infection in the relevant tissues.
In conclusion, we report that it is now possible to specifically identify NKG2C+ and NKG2A+ macaque NK cells using their respective transcripts, KLRC2 and KLRC1, as proxy. Further, we show for the first time that rhCMV infection results in increased NK cell diversity and a specific increase in NKG2C+ NK cells. Altogether these findings strengthen the argument for NKG2C+ memory and memory-like NK cells arising in response to CMV and lentivirus infections and provide a tangible NHP model in which to study them.
|
10.1371/journal.pcbi.1001050 | Measuring the Evolutionary Rewiring of Biological Networks | We have accumulated a large amount of biological network data and expect even more to come. Soon, we anticipate being able to compare many different biological networks as we commonly do for molecular sequences. It has long been believed that many of these networks change, or “rewire”, at different rates. It is therefore important to develop a framework to quantify the differences between networks in a unified fashion. We developed such a formalism based on analogy to simple models of sequence evolution, and used it to conduct a systematic study of network rewiring on all the currently available biological networks. We found that, similar to sequences, biological networks show a decreased rate of change at large time divergences, because of saturation in potential substitutions. However, different types of biological networks consistently rewire at different rates. Using comparative genomics and proteomics data, we found a consistent ordering of the rewiring rates: transcription regulatory, phosphorylation regulatory, genetic interaction, miRNA regulatory, protein interaction, and metabolic pathway network, from fast to slow. This ordering was found in all comparisons we did of matched networks between organisms. To gain further intuition on network rewiring, we compared our observed rewirings with those obtained from simulation. We also investigated how readily our formalism could be mapped to other network contexts; in particular, we showed how it could be applied to analyze changes in a range of “commonplace” networks such as family trees, co-authorships and linux-kernel function dependencies.
| Biological networks represent various types of molecular organizations in a cell. During evolution, molecules have been shown to change at varying rates. Therefore, it is important to investigate the evolution of biological networks in terms of network rewiring. Understanding how biological networks evolve could eventually help explain the general mechanism of cellular system. In the past decade, a large amount of high-throughput experiments have helped to unravel the different types of networks in a number of species. Recent studies have provided evolutionary rate calculations on individual networks and observed different rewiring rates between them. We have chosen a systematic approach to compare rewiring rate differences among the common types of biological networks utilizing experimental data across species. Our analysis shows that regulatory networks generally evolve faster than non-regulatory collaborative networks. Our analysis also highlights future applications of the approach to address other interesting biological questions.
| With the advent of large-scale genomic and proteomic technologies in discovering interacting and regulatory relationships in cells, many types of biological networks, though incomplete, have been constructed in various eukaryotic species [1]–[19]. The kinds of networks currently include, but are not limited to, protein interaction, genetic interaction, transcription factor-target regulatory, miRNA-target regulatory, kinase-substrate phosphorylation, and metabolic pathway. Biological networks have been used to explain differences between closely related species that share high sequence similarities [1], [2], [7]. For example, human and chimpanzee genomic sequences are found to have only 1.23% differences in SNPs and 3% in indels [20]. However, the subtle sequence divergence is hardly sufficient to explain phenotypical, behavioral and social differences between the two species. As a result, biological networks (organizations of molecules) are proposed to play a central role in speciation complementary to individual molecules [1], [2], [7]. However, it is still largely unknown how fast biological networks evolve.
Biological network research has followed the path of sequence research to some degree. In the past three decades, biological sequence research has experienced three stages: initial sequencing data generation, pairwise alignment and evolutionary rate analysis. Simple models such as the Jukes-Cantor model [21] describe evolutionary sequence divergence in terms of time. In fact, various biological sequences evolve at different rates depending upon their functional importance [22], [23]. Genomic sequence analyses in various species have helped us to learn levels of conservation among genomic regions and genes [24]–[26]. Similarly, proteomic sequence and structure analyses show that protein regions have varied evolutionary constraints [27], [28]. Analogous to sequence analysis, the development of biological network research has three similar stages: network construction by large-scale experiments and computational predictions [1]–[19], pairwise network comparison to find conserved edges as interologs or regulogs [29], [30] and building general network alignment tools [31], [32], and finally investigating levels of conservation and evolutionary change on biological networks.
One of the advantages of network study is that we can make analogies to draw intuition. For example, in commonplace social contexts, we readily observe that some “network” relationships change faster than others. Personal acquaintance networks may change in days, friendship networks and co-worker networks in months or years, while family networks change over decades. This intuition of network stability differences could be quantified and compared by the rewiring rate that reflects the nature of network relationships. Similarly, in cellular systems biological networks may rewire at various rates during evolution.
Increasingly we have seen many approaches to compare biological networks across organisms, uncovering interesting relationships of network evolution and the functional implications [7], [33]–[39]. Due to current limitations of network construction technologies and the large evolutionary distance between the species compared, the overlap between current network datasets is small. Nevertheless, the estimation of the rewiring rate in protein interaction networks is possible [33]. Various methods were used in different studies inconsistent for direct comparison, with each focused on one of the biological network types. Also, most of the studies were species specific that did not compare species with large evolutionary divergence.
Given that previous studies have set the stage, now is an opportune time to quantify network rewiring in all these comparisons in a unified way. In the past three years, more data has become available for a greater number of species covering many types of biological networks [1], [2], [4], [5], [7]. The comprehensive set of network data allows systematic comparison of rewiring rates of biological networks and drawing more robust conclusions by using a set of species pairs.
We show here the rewiring rates of several types of biological networks in eukaryotes. The approach used is consistent across network types and robust to network data quality. We observed that the rewiring rate is characteristic of the type of edge (relationship between node entities) in both biological and commonplace networks. This analysis gives an initial picture of biological network rewiring and provides intuition and useful tools for the future when more network data becomes available.
To calculate the rewiring rate of biological networks, we first established node orthology between two species, and then defined edge orthology as a conserved relationship between orthologous entities across different species, which is a generalization of “interologs” in protein interaction network and “regulogs” in TF regulatory network [29], [30]. One species network is considered reference, and three sets of nodes are identified. Common nodes (CNs) are nodes present in both networks, loss nodes (LNs) only in reference network and gain nodes (GNs) only in the other compared network. Four types of rewired edges are then identified and counted including gain or loss edges between CNs, loss edges involving LNs, and gain edges involving GNs (see Figure 1). The rewiring rate was measured by the total number of rewired edges (R) between two networks normalized by the combined network size, the total number of possible edges if two networks were both “complete” (C), and divergence time (T). Total number of rewired edges (R) counts all non-conserved edges (interologs, regulogs or other type of “logs”) in two networks. The total number of possible edges (C) has five components: total possible edges of complete networks consisting of only common nodes (CNs), nodes that are only present in one of the two networks (GNs or LNs), and total possible edges between the two (between CNs and LNs, or CNs and GNs) (see Figure S1, see Materials and Methods). The measure is in essence percentage edge change of network in a given time period. We have collected data for each type of network for different species (see Table S1), and calculated rates for different time divergence species pairs (see Figure 1).
For all types of biological networks, we observed faster rewiring rates for smaller divergence species pairs and slower rewiring rates for larger divergence species pairs, with a strong negative linear relationship between rewiring rate (per edge per Mys) and divergence time (Mys) in Log-Log scale (see Figure 2, Table S2). It was thus inappropriate to use the rewiring rate calculated from a specific species pair as a general measure for a network type. Using species pairs with different divergence times could result in large differences. However, different species pairs with similar divergence times tended to have close rewiring rates. This indicated that our rewiring rate measure was dependent upon divergence time but not on species.
We then asked the question whether the observed negative linear relationship in Log-Log scale between rate and divergence time in networks is parallel to what is seen in nucleotide sequence evolution. For sequence evolution, we use the equation from the Jukes-Cantor model, where P is the percentage of sequence change and T is divergence time [21]. Though it is a simple model with only one parameter (α), Jukes-Cantor model captures the core relationship between P and T, and is sufficient in this case for comparing sequences with networks. P/T is the approximation of the instantaneous sequence evolutionary rate (dP/dT) and can be used for direct comparison with rewiring rate of networks. A negative linear relationship was observed in Log-Log scale between P/T and T (see Figure 2), and was especially strong at large divergence times.
Further, we used simulated networks to determine whether the observed relationship is specific to real biological networks. A simulation-based network rewiring model was built based on four parameters, corresponding to node changes, edge changes, and preferential attachment to rewiring networks while maintaining scale-free topology (see Materials and Methods). As a simulation of evolutionary divergence, two branches of networks were compared after the same number of rewiring steps and rewiring rates calculated (see Figure S2). The rewiring rate calculated from the simulation model also shows a negative linear relationship in Log-Log plot with number of rewiring steps (see Figure 2).
The analysis above indicated that the negative linear relationship between the rewiring rate and time in real networks could be universal and reflect underlying principles in evolution. This intuitively corresponds to the saturation of percentage change. For nucleotide sequences, as divergence becomes larger, the percentage of sequence change saturates at 0.75 according to the Jukes-Cantor model. New nucleotide changes happen on top of previous changes, which have little effect on percentage difference. Our analysis showed that the same is true for networks.
We used the fitted rates from linear models for each type at 800 Mys divergence, roughly half the time of eukaryotic history (see Table 1). The “banding” of networks on the plot into characteristic groups with order of magnitude rate differences between them indicates the robustness of the rewiring rate calculation and the actual rate difference between networks.
In fact, the above described rewiring rate is an “average” rate rather than “instantaneous” rate for networks. As the Jukes-Cantor model shows for sequences, evolutionary rate (α) could only be approximately measured using instantaneous rate (dP/dT) between closely related species (dT is small), where α is proportional to dP/dT. When the divergence gets large, the approximation of instantaneous rate with the average rate is poor and the relationship between α and dP/dT becomes non-linear. The logic is directly applicable to our case for networks.
Ideally, instantaneous rewiring rate should be measured using networks between closely related species. However, little network data are available for close species, which inhibits the calculation of instantaneous rewiring rates. The disadvantage of using the average rates described above is that at large evolutionary distance, network rewiring approaches saturation and is hard to compare. And the limited number of species network comparisons does not allow accurate estimations of instantaneous rates by the linear model at less than 10Mys divergence (see Table S2).
Another idea of comparing rewiring of biological networks is to use networks for a given divergence of the same species pairs. Since networks are of the same divergence, we use the percentage of edge changes among total possible changes, which is R/C, to measure the extent of rewiring (see Table 2). This method circumvents the disadvantages of average rewiring rate and limited species comparisons of networks, while it maintains the ability to distinguish the extent of network rewiring. For each of the 11 species comparisons listed in Table 2, biological networks are ordered according to their percentage of rewiring. We then count the number of cases where one type of biological network is observed to rewire more or less than another (see Table 3). Thus for each comparison between species (at a given level of divergence), we get an ordering of network rewiring (e.g. transcription regulatory>phosphorylation regulatory>protein interaction>metabolic pathway). We found that the ordering is consistent amongst all the 11 comparisons in this study. This result further supports the differences found in network rewiring using averaged rates.
The formalism of network rewiring was also applicable to non-biological networks to get some intuition for fast or slow rewiring processes (see Table 4). Three different representative commonplace networks with very different divergences were constructed, including co-authorship networks, family trees and Linux kernel design networks (see Figure S3). The three types of non-biological networks showed differential rewiring rates in the order of magnitudes (see Table 4). Consistent with our intuition, for example, family trees have less rewiring than co-authorship networks. Contrary to popular opinion of frequent computer software updates, Linux kernel design network in fact evolves approximately one order of magnitude slower than a typical family tree (more family samples needed for statistically significant arguments). It is clear that rewiring rate could help us understand the nature of edge relationship in networks, thus can be used for direct comparisons among all kinds of biological and social networks.
Rewiring of biological networks consist of two sources: edge change between conserved nodes, and edge change from node gain and loss. We observed that a large fraction and in many cases the majority of network rewiring is attributed to the gain and loss of nodes (see Table S3). In fact, gene content turnover of two species contributes to the gain and loss of nodes in networks. Some studies have suggested differential gene content turnover of gene families, such as transcription factors and metabolic enzymes, in completely sequenced genomes [40]–[42]. Therefore, it is important to assess the impact of gene family evolution on the extent of their respective network rewiring.
In order to examine whether the turnover of a specific set of genes, such as kinases and TFs, have impact on their corresponding network rewiring, we examined the gene content turnover of 3 GO categories using 3 species pairs (see Table 5). The 3 GO categories (transcription factor activity, kinase activity, and metabolic process) are selected to be compared with TF-target regulatory network, kinase-substrate phosphorylation network, and metabolic enzyme network, respectively. For the 3 categories of proteins, we did not observe a clear pattern in which some categories had faster turnover than others. This suggests that differences in network rewiring across networks may not come from the gene content turnover of corresponding GO category proteins. The rewiring of networks should mostly reflect the characteristic of biological relationships rather than specific GO category molecules themselves.
It is also interesting to notice that even if the fungi S. cerevisiae and K. lactis have the largest divergence of 150 Mys among three species pairs, the gene content turnover is much less than the other two pairs. This slow gene content turnover with a large species divergence further supports the role of network rewiring during evolution.
Cellular molecules, as nodes in biological networks, are under differentiated selection pressure, and therefore evolve at different rates. Genomic analyses from model organisms have shown the spectrum of sequence conservation among types of genomic annotations, in which protein coding exon sequences are the most conserved, intron sequences are the least conserved, and regulatory cis/trans elements are somewhere in between [43]. Proteins as the products of DNA coding sequences are generally thought to be under great constraint. Another special product from DNA sequences is ribosomal RNA, which is considered the most conserved locus in the genome [44].
We asked whether the edge rewiring rates in biological networks were in the range of node changes. Since there is no analogous concept of “total possible edges between nodes” in sequence comparisons, a naïve sequence/network identity-based method was used to measure the percentage change between two sequences/networks for consistency (see Materials and Methods). Here, only edge changes in networks are counted to compare with nucleotide change in sequences. Sequence identity is calculated as the percentage of the number of unchanged nucleotides or amino acids in global alignment per length of the alignment. Similarly, network identity is calculated as the percentage of the number of unchanged edges out of total number of edges in two networks. Then, the rate can be calculated as (1- percentage identity)/(divergence time) for both sequence and network. This equates one edge change with one nucleotide or amino acid change. We realized this might not be the best, but a default to start with.
Using this definition, we observed that biological networks evolve in a range comparable to that of protein sequences in both species cases (see Figure 3). Transcription factor-target regulatory networks, the fastest rewiring biological networks, were comparable to the top 0.1% and 4% of the fastest evolving protein sequences in Homo sapiens and Sacchromyces cerevisiae, respectively. The slowest rewiring metabolic pathway network was comparable to the bottom 23% and 36% of the slowest evolving protein sequences. The density distribution of protein coding DNA sequence rates had a similar peak position but a smaller standard deviation than protein sequence rates, because an amino acid change does not necessarily result from changes of all its three codon positions. Therefore the evolutionary rate distinction between protein coding sequences and biological networks became more significant: with 0.5% and 4% of sequences slower than metabolic pathway networks in human and yeast, respectively, and 0% and 4% of sequences faster than transcription factor-target regulatory networks. The 18S rRNA sequences evolved slower than all biological networks analyzed here: approximately 60% rate of the slowest rewiring metabolic pathway network in human and 1% of the rate in yeast.
Since rewiring rates are capable of distinguishing different network types, we attempted to use rewiring rates to study different subtypes of edges within protein interaction networks. Relating protein 3-D structures to protein interaction networks helped us to distinguish simultaneously possible (permanent) interactions from mutually exclusive (transient) interactions [45]. The difference between the two types of interactions is whether an interaction between two proteins has competition from a third potential interacting protein for the same interacting site. It has long been hypothesized that protein pairs of permanent interactions tend to co-evolve during evolution [46]. The co-evolutionary effect could help to maintain the stability of permanent interactions.
Structural interaction networks (SINs) for both human and yeast were constructed using updated and coherent datasets. Permanent and transient interactions were identified through interacting site regions in proteins and number of interacting partners for each site. Conservation of permanent and transient interactions was measured by their presence in another reference species network (see Table 6). Significant conservation distinction was observed for permanent and transient interactions in both yeast (p-value = 0.001) and human networks (p-value = 0.05) using Fisher's Exact Test. Stronger conservation of permanent protein interactions indicated that the interacting sites within two proteins were more constrained to maintain the interaction via co-evolution of interacting sites.
The results above showed that the rewiring rate of network edges reflects the biological nature of edge types. It is also plausible that proteins with different characteristics might have different rewiring rates than their network partners. Here, we used protein interaction networks to investigate how protein paralogs behave during evolution in terms of changing their interacting partners. We collected all paralog pairs present in human and yeast interaction networks and calculated the rewiring rate difference between each pair (see Materials and Methods). The distribution of the rate difference was then compared with a background distribution calculated for all protein pairs in the networks (see Figure 4).
In both human and yeast networks, the paralog pairs had rate difference distribution shifted to zero compared to background (Wilcoxon test p-value<e−15 in yeast, p-value = 0.004 in human). The result suggested that paralog pairs tend to have a smaller rewiring rate difference, demonstrating a closer evolutionary rate of network change. In fact, as paralogs emerge from the event of gene duplication in ancestral species, they share sequence similarities [47]. Here, we showed that paralogs also shared network similarities as the network rewiring rates of paralogs were similar. After the gene duplication events which lead to their formation in ancestral species, paralogs are likely to have similar constraint on sequences and network partners due to their shorter evolutionary history than random protein pairs.
King and Wilson proposed [48] and Bourman et al. [1] then demonstrated that fast changing regulatory relationships in transcription factor-target networks could account for the species differences, which could hardly be explained by the highly conserved protein and DNA coding sequences. Following that study, small- and large-scale evidence has been presented to support the view that after the divergence of two species, fast change in regulatory relationships may have a critical role in speciation [2], [7]. As we have shown above, transcription factor-target regulatory networks and kinase-substrate phosphorylation networks are two major types of regulatory networks that have the fastest evolutionary changing rates among networks and protein sequences, confirming the importance of regulation in species evolution.
Unlike sequence data that one is essentially sure of every base, network data either generated from experiments or computational predictions are currently subject to high number of false positives and false negatives. Because many distinct experimental approaches are used to generate network data, different biological networks may have varied systematic bias during their construction. It is inevitable that our results might be subject to change when new network data become available.
For each type of biological networks, we used consistent data source and method to build networks for species, which ensures the uniform definition of edges and facilitates comparison between species.
Instead of trying to build high quality networks for all biological networks in multiple species, which is difficult due to lack of gold-standard positives and negatives, we applied a general method to assess the influence of false positives and negatives to rewiring rate calculation for all biological networks. Beltrao et al. have used a sampling-based sensitivity analysis to assess the robustness of rewiring rate relative to the amount of protein interaction data used [33]. Here, we applied a similar method to six representative types of biological networks used in this study. The effects of false negatives and false positives are simulated by random sampling. That is, we randomly add and remove a fraction of edges of the two compared real networks, forming simulated “corrected” networks, and then calculate rewiring rates. A series of disruption fractions of random edges are used to simulate false positive and negative rates from low to high (see Figure S5, see Materials and Methods).
Rewiring rates of most of the biological networks are robust to network size change and disruption, especially when the disruption fraction is lower than 50%. However, the rates of metabolic pathway networks have shown clear deviations at large disruption levels. The observed one order of magnitude difference between metabolic pathway networks and protein interaction networks (10−5 for protein interaction network, 10−6 for metabolic pathway) disappears at approximately 70% disruption level. We conclude from these results that the network rewiring rate is only slightly affected by network size, and is especially robust at sampling levels above 50%. The results of this study should still hold when new network data arrives.
We also investigated the potential size effect of fungi TF-target regulatory networks used in our study. These networks were constructed using binding sites from ChIP-chip experiments of one or two TFs, which results in relatively small networks. Besides the simulated disruption described previously on these small networks, edges were added to the S. cerevisiae network from another ChIP-chip study between the existence nodes to generate a larger network [49]. The same disruption analysis was performed on the larger network. Rewiring rates calculated from the larger network decreased about half order of magnitude than from the original small network (see Figure S5). This is largely due to the increase of total possible edge changes in our calculation. As a result, the current subnetwork of TF-target regulatory network might lead to a bias of faster rewiring rate.
A comprehensive simulation analysis was further performed to assess the effects of both network size and network quality (see Materials and Methods). Two simulated scale-free networks were constructed with some common edges, and sub-samples of both networks were taken for comparison. Random rewiring of both sub-network were performed to mimic false positives and negatives. Percentage of edge change (R/C) was calculated for each sub-sampling fraction. As the size of the compared sub-networks decreases, percentage of rewiring increases (see Table S4). The upward bias of percentage of rewiring is approximately one order of magnitude corresponding to 1% sub-sampling fraction. Because the fungi TF regulatory network used in this study is approximately 20–100 times smaller than the complete networks estimated by the number of edges and the number of TFs [49]. We thus estimated that the true rate of fungi TF regulatory network could be half to one order of magnitude slower than we calculated. Considering the above estimation of network size effect on rewiring measurement, fungi TF regulatory network should still rewire faster than or in a similar pace as kinase phosphorylation network, and much faster than other types of biological networks (see Table 1).
miRNA regulatory networks were constructed using a consistent miRNA target prediction method [50]. In the current stage of miRNA research, most miRNAs are found or predicted using sequence conservation, and regulatory relationship is predicted mainly by searching for complementary sequence in 3′ UTRs [9]–[11]. Therefore, the turnover of miRNAs is small with lack of species-specific miRNAs and their corresponding targets. For example, a total of 459 conserved miRNAs are present in the networks comparing human and mouse. However, only 18 and 9 miRNAs are human-specific and mouse-specific, respectively. The mere gene content turnover of only 6% for miRNAs is much less than 67% and 74% for TFs and kinases (see Table 5). This ascertainment bias could result in under-estimation of rewiring rates.
To estimate the effect of novel miRNAs to our rewiring measurements, we randomly added a series numbers of hypothetical novel miRNAs to actual human and mouse miRNA regulatory networks. The targets of those hypothetical miRNAs are also randomly selected with degree distribution maintained. Rewiring rates calculated from these simulations showed that discovering potential species-specific miRNAs could result in an increase of rewiring rate (see Table S5). With the advance of miRNA research from novel miRNA discovery to better target prediction methods, it is possible that the current rewiring rates of miRNA regulatory networks will be adjusted higher.
For all types of biological networks and simulated networks we observe a negative linear relationship between rewiring rate and divergence time (see Figure 2). Generally speaking, the average rewiring rate calculated comparing distant species networks tends to be smaller than the instantaneous rate comparing close species networks. For networks from two distant species, overlap of their nodes becomes smaller due to loss of conservation. As a result, the total number of possible edges C increases and rewiring rate decreases correspondingly. In conclusion, a larger difference between node sets of two distant species networks might be the main reason for this bias.
The major effect of node gain and loss on rewiring rate was further confirmed by a sensitivity analysis based on network rewiring simulation model (see Materials and Methods). Each of four independent parameters in our model was tested for its relative importance to model output—rewiring rate. Not surprisingly, we found that some parameters are more significant to the model than others. Removal of node has the strongest effect (negative linear) on rewiring rate, because rewired edges associated with a node are removed along with the node, which decreases the total number of rewired edges. Adding node also has some effect (positive linear) on rewiring rate, because of the increased number of total rewired edges associated with the node. Nevertheless, removing and adding edges have only small effects on rewiring rate (see Figure S4). It is reasonable that removing and adding nodes has a major influence on rewiring rate as it affects all edges associated with nodes rather than individual edges.
It is also possible that there are “cores” for each type of networks that slow down the rewiring process when it approaches the cores. The cores are partial networks that are the most constrained and conserved during evolution, possibly reflecting their functional importance. Therefore, network types with a smaller ratio of rewiring rate changes and divergence time (flat lines) might have larger cores, because of greater resistance to rewire the cores; while network types with a larger ratio (steep lines) might have smaller cores (see Figure 2).
Biological networks are characterized by their functional relationships: protein binding, expression regulation, phosphorylation, etc. We introduce another way to categorize biological networks into collaborative and regulatory networks by the reversibility of edges to help understand rewiring rate distinction among network types. Collaborative networks are the biological networks with reversible edges—either the edges are undirected or directed but reversible. By reversibility we mean that a reversed edge is biologically possible between a pair of nodes. Regulatory networks have irreversible edges: a reversed edge may not be biologically possible. By this definition, transcription factor-target regulatory networks, miRNA-target regulatory networks, and kinase-substrate phosphorylation networks fall into the regulatory network group; and protein interaction networks, genetic interaction networks, and metabolic networks fall into the collaborative network group.
Our network rewiring analysis shows that in general, regulatory networks tend to rewire faster than collaborative networks (see Table 1). Two of the regulatory networks, transcription factor-target regulatory networks and kinase-substrate phosphorylation networks, are the fastest rewiring biological networks in this study. Transcriptional regulation of gene expression by transcription factors is carried out by transcription factor binding to the transcription start site commonly upstream of a gene. Recognition of a binding site is often specific to a sequence pattern buried in the site [51]. Post-translational modification of protein substrate by kinases also involves recognition of sequence patterns in substrate's phosphorylation site [52]. Sequence pattern matching as a major factor in establishing regulatory relationships could be an important reason of fast rewiring. A single nucleotide/amino acid change in the target's binding-recognition sites, could lead to a “digital” recognition site change. Besides, a number of studies have showed that both transposable element insertion and genomic rearrangement led to considerable indel changes at transcription factor binding sites [53]–[58]. The digital and indel changes in binding-recognition sites greatly contribute to the large turnover of transcription factor-target regulatory network.
Collaborative networks show slower rewiring rates than regulatory networks. Contrary to “digital” or “indel” changes in regulatory networks, changes tend to be “structurally continuous” in collaborative networks. Here, we generally refer to the globular interactions which are the majority in physical interaction networks. On the other hand, the general collaborative physical interaction network in this study still includes interactions mediated by kinases and domains such as SH3 which are in fact regulatory relationships. In fact, protein functions gradually change as sequence changes, and most proteins do not change their functions radically with their sequences conserved. As a natural implication of the sequence-function paradigm, it is not surprising that collaborative protein networks rewire as protein sequences evolve. In this study we include two representations of metabolic networks. Metabolic enzyme networks are constructed using enzymes as nodes and edges connect two nodes if the product of one serves as the substrate of the other. The rewiring rates of metabolic enzyme networks are similar to other collaborative networks (see Table 2). On the other hand, metabolic pathway networks that are constructed using chemical compounds as nodes and reactions as edges rewire the slowest. For example, the biosynthesis metabolic pathway of acetyl-CoA from pyruvate is identical in human and yeast, but the corresponding metabolic enzyme network rewires (see Figure S6). In fact, metabolic reactions process chemical compounds into energy and nutrition, and are mostly essential for living. Our results suggest that the essentiality is partly reflected in the slower rewiring rate of metabolic pathway networks than that of other types of biological networks and protein sequences. Based on these results, we think that enzymes for reactions are less constrained to change while the underlying reactions remain highly conserved.
We now know that there are two layers of cellular evolution, individual molecules and organizations of molecules. Therefore, it is our ultimate goal to understand how individual molecule changes affect cells and their organization and collaboration.
Some factors may also influence and shape the landscape of biological networks (see Figure 5). It has been shown that external environment can influence the conservation of regulatory relationship and network motifs in prokaryotic transcription factor-target networks [59], [60]. Relationships tend to be conserved in organisms living in similar environmental niches, despite large evolutionary distance. Whole-genome duplication events rapidly reorganized transcription regulatory networks through the survived duplicates and their functional divergence afterwards [61]–[65]. And the regulatory networks, in a feedback way, could affect the survival of duplicated genes [66].
This study attempts to systematically investigate the evolutionary rate of all known types of biological networks in terms of rewiring. According to our results, it is possible that small changes of molecular sequences lead to large network re-organizations and this augmentation effect makes small molecular changes more detectable by natural selection. This is especially true for regulatory networks with the greatest augmentation effects caused by minor changes of regulators. If the above assumptions are true, network rewiring should be an essential tool to understand the differences between closely related species such as human and chimpanzee, because their molecular sequences are nearly identical. More importantly, intra-species network rewiring variations will help at an individual level beyond SNPs and structural variations.
In the future, we foresee additional calculations and analyses that could be performed when accurate and more complete network data becomes available for more species. Analogous to sequence analysis, we can build species trees comparing biological networks and infer branch lengths using rewiring rates. From this study, we know that types of biological networks and molecular sequences evolve at different rates, but it is still unclear whether network rewiring “speeds up” in some species and “slows down” in others. We can use benchmark rates and develop comparative ratios to measure this. This is actually quite similar to using dN/dS ratio (non-synonymous changes versus synonymous changes) to measure selection pressure on coding sequences. Building the tree is important to understanding biological system evolution compared to traditional molecular evolution.
Network hubs and bottlenecks are of general interest in biological research due to their topological importance. Both hub and bottleneck proteins in human and yeast protein interaction networks tend to rewire their edges faster than non-hub non-bottleneck proteins (see Figure S7). One reason for this is that hubs with large degrees tend to have more rewired edges, and therefore faster rewiring rates. Further detailed analysis is needed to understand the rewiring rates of bottleneck proteins.
It is also interesting to look for rewiring “hotspots” and “coldspots” within biological networks. Subnetworks and motifs that are enriched in fast or slow rewiring edges may have biological function implications. Immune response, transport and localization associated genes in human protein interaction networks have been found to change interacting partners relatively quickly [33]. The analysis could also be applied to other types of biological networks.
Further network rewiring analysis will possibly investigate factors affecting network rewiring (see Figure 5). These efforts will greatly increase our understanding of cellular system evolution, intra-species variation, and speciation.
For different types of biological networks, we gathered data from multiple sources. Binary protein physical interaction networks and genetic interaction networks were extracted from BioGRID database v2.0.55 (http://thebiogrid.org/) for 5 species: H. sapiens, C. elegans, D. melanogaster, S. pombe and S. cerevisiae [67]. Metabolic pathway networks of compound reactions were obtained from KEGG database (http://www.genome.jp/kegg/) for 16 species: H. sapiens, M. mulatta, M. musculus, C. elegans, C. briggsae, D. melanogaster, D. pseudoobscura, S. pombe, D. hansenii, C. albicans, K. lactis, C. glabrata, S. bayanus, S. mikatae, S. paradoxus and S. cerevisiae [68]. Metabolic enzyme networks were constructed from the pathway networks for 7 species: H. sapiens, M. mulatta, M. musculus, C. elegans, D. melanogaster, D. hansenii, and S. cerevisiae, by establishing directed edges from upstream reaction enzymes to downstream reaction enzymes. miRNA-target regulatory networks were constructed from miRBase (http://www.mirbase.org/) predictions with edges pointing from miRNAs to target genes in 5 species: H. sapiens, M. musculus, D. rerio, C. elegans and D. melanogaster [50]. Transcription factor-target regulatory networks were extracted from various sources: S. cerevisiae, C. elegans and D. melanogaster networks from large-scale ChIP-Chip and ChIP-Seq experiments [3]–[5], C. albicans, K. lactis, S. bayanus, S. mikatae networks from recent small-scale experiments [1], [2]. Kinase-substrate phosphorylation network for S. cerevisiae was obtained from large-scale protein chip experiments [6]. Phosphorylation networks of yeast species S. cerevisiae, C. albicans and S. pombe were constructed in two steps. We first obtained phosphorylation sites identified by MassSpec [7], and also obtained kinase binding specificity data from kinase binding specificity experiments [69]; then used MOTIPS analysis pipeline to identify responsible kinases for each phosphorylation site by matching position weight matrices (PWMs) [70]. Structural Interaction Networks (SINs) for H. sapiens and S. cerevisiae were constructed in a similar way as the first version of yeast SIN [43], using protein domain interaction data from iPfam database Release 20.0 (http://ipfam.sanger.ac.uk/) [71].
For social co-authorship network, we parsed the co-author lists of 2009 Nobel Prize Winner Thomas A. Steitz's 2009 and 2006 publications from PubMed (http://www.ncbi.nlm.nih.gov/pubmed/) [69], and constructed co-authorship networks for Dr. Steitz. For social family tree network, we obtained data from a typical family with its trees in 1983 and 2009 (see Figure S3). Edges in family trees stand for either marriage or child/parent relationship. Linux kernel design networks are obtained for 3 versions, v2.6.4, v2.6.15 and v2.6.27. From v2.6.4 to v2.6.15 and from v2.6.15 to v2.6.24, the time separations are around 2 years and 2.5 years, respectively [72]. One edge in Linux kernel design networks represents one function calling or using another function.
Protein sequences and protein coding DNA sequences for H. sapiens, M. musculus and S. cerevisiae were downloaded from BioMart database (http://www.biomart.org/) [73], and from SGD (http://www.yeastgenome.org/) for S. mikatae. 18S ribosome RNA sequences for all 4 species were extracted from Entrez database (http://www.ncbi.nlm.nih.gov/Entrez/) [74]. Orthologous sequences in H. sapiens-M. musculus and S. cerevisiae-S. mikatae pairs were then aligned using MUSCLE software v4.0 (http://www.drive5.com/muscle/) [75] for calculations of sequence identity.
Sequence orthology for non-fungi species pairs used in this study was downloaded from InParanoid database v7.0 (http://inparanoid.sbc.su.se/cgi-bin/index.cgi) [76]. Orthology for fungi species pairs was obtained from Fungal Orthogroups Repository v1.1 (http://www.broadinstitute.org/regev/orthogroups/) [77]. Paralog pairs in H. sapiens and S. cerevisiae were extracted from HomoloGene database Release 64 (http://www.ncbi.nlm.nih.gov/homologene) [74].
We used a consistent method to calculate rewiring rates comparing two networks for all network types. First, orthology relationships between nodes from the same network type in two species were established. Second, three sets of nodes were distinguished. Common Node (CN) set includes nodes having orthologous counterparts present in both networks. Loss Node (LN) set includes nodes present in the reference network but absent of orthologous counterparts in the compared network. And Gain Node (GN) set includes nodes present in the compared network but not having orthologous counterparts present in the reference network. Third, we counted the total number of rewired edges (R) between two networks. Rewired edges between two networks were defined as the union of edges between pairs of CNs that only present in one network and all edges involving LNs and GNs. Fourth, we counted the total number of possible edges (C) in the two networks. This was basically the number of non-redundant edges if two networks are both fully connected. Finally, the following equation was used to calculate the rewiring rate for a pair of networks:The time divergence is either estimated evolutionary divergence time (in Mys) between two species in biological networks or passed period of time (in years, and then coerced to Mys) in commonplace networks (see Table S1). Thus, the rewiring rate was measured as the number of rewired edges per edge per Mys. It can be interpreted as the averaged fraction of rewired edges among all possible edges in a period of one million years.
However, total number of possible edges was calculated differently among network types. Calculation for collaborative networks, including social networks, is simpler because their edges are reversible (see Figure S1):Note that here we did not allow self interactions and only allowed one edge between two nodes. For metabolic networks that allow two reciprocal edges between two nodes (for directional reactions), we just multiplied the above calculated result by 2. For regulatory networks involving irreversible edges, we further separated nodes into regulators (Regs) and targets (Tars) and only allowed edges from Regs to Tars, but not from Tars back to Regs. In addition, regulators in transcription factor-target regulatory network and kinase-substrate phosphorylation network could themselves be targets of other regulators, but not in miRNA-target regulatory network. Considering all these factors (see Figure S1),and
Rewiring rates and their corresponding estimated divergence times were plotted on Log-Log scale and then fitted with linear regression model. Using species pairs with divergence time of t Mys, the rewiring rates, r, was then regressed for each type of biological networks (see Table S2).
The rewiring rate calculation described above was not directly comparable to sequence evolution rate calculation, as there is no equivalent to the ‘total number of possible edges’ as in networks. Therefore, we used identity-based evolutionary rate measures instead to compare networks and sequences as:The evolutionary rate calculated based on identity was:
Rewiring rates for all individual nodes were calculated for H. sapiens and S. cerevisiae protein interaction networks by comparing them to D. melanogaster and S. pombe networks, respectively. Number of rewired edges for each node was counted as the number of gained or lost edges involving this node. This number was then divided by network size and by divergence time to get rewiring rate for a node. Network size is difference for CNs, GNs and LNs. For CNs, network size is the sum of the number of CNs, GNs and LNs from the two networks; for GNs, network size is the sum of CNs and GNs; and for LNs, network size is the sum of CNs and LNs. Rewiring rate difference was then calculated for all node pairs including all paralog pairs.
The model had four parameters: probabilities of adding a node (adding one edge with that node using preferentially attachment), removing a node (randomly for all existing nodes and all edges with that node), adding an edge (using preferentially attachment) and removing an edge (randomly for all existing edges). Preferential attachment mechanism maintains the scale-free topology of networks. To begin with, a small scale-free network was used as a seed to the model. For each rewiring step, nodes and edges were added/removed according to the probability parameters, and the resulting network was recorded for the next step.
For the relationship analysis of rewiring rate and rewiring steps, two independent rewiring branches were simulated with each 1000 steps (see Figure S2). The networks from the two branches were compared after every 50 steps and rewiring rate was calculated.
For parameter sensitivity analysis, 200 parameter-set samples were generated, with the four probability parameters randomly generated from a uniform distribution on the interval [0,1]. The same seed network was used for all 200 simulations using the 200 random parameter-sets. All simulations were stopped after 100 steps and rewiring rate was calculated corresponding to each of the 200 parameter-sets.
Two simulated scale-free networks were built with some common edges for comparison. The pair of networks were sub-sampled of their edges to a series of fractions, from 95% to 1%. To assess the amount of false positives and false negatives in network data to rewiring rate calculation, we further perturbed the compared network pair (either real biological networks or simulated networks) by randomly adding and removing edges on both networks. Edges were added using preferential attachment. A series of perturbation percentages were used to simulate levels of false positive and negative rates.
|
10.1371/journal.ppat.1003446 | Bruton's Tyrosine Kinase (BTK) and Vav1 Contribute to Dectin1-Dependent Phagocytosis of Candida albicans in Macrophages | Phagocytosis of the opportunistic fungal pathogen Candida albicans by cells of the innate immune system is vital to prevent infection. Dectin-1 is the major phagocytic receptor involved in anti-fungal immunity. We identify two new interacting proteins of Dectin-1 in macrophages, Bruton's Tyrosine Kinase (BTK) and Vav1. BTK and Vav1 are recruited to phagocytic cups containing C. albicans yeasts or hyphae but are absent from mature phagosomes. BTK and Vav1 localize to cuff regions surrounding the hyphae, while Dectin-1 lines the full length of the phagosome. BTK and Vav1 colocalize with the lipid PI(3,4,5)P3 and F-actin at the phagocytic cup, but not with diacylglycerol (DAG) which marks more mature phagosomal membranes. Using a selective BTK inhibitor, we show that BTK contributes to DAG synthesis at the phagocytic cup and the subsequent recruitment of PKCε. BTK- or Vav1-deficient peritoneal macrophages display a defect in both zymosan and C. albicans phagocytosis. Bone marrow-derived macrophages that lack BTK or Vav1 show reduced uptake of C. albicans, comparable to Dectin1-deficient cells. BTK- or Vav1-deficient mice are more susceptible to systemic C. albicans infection than wild type mice. This work identifies an important role for BTK and Vav1 in immune responses against C. albicans.
| The opportunistic yeast Candida albicans is a commensal organism of the human digestive tract, but also the most common cause of human fungal infections. Phagocytosis, the process by which innate immune cells engulf pathogens, is vital to prevent C. albicans infections. The major phagocytic receptor involved in anti-fungal immunity is Dectin-1. We identify two new interacting proteins of Dectin-1 in macrophages: Bruton's Tyrosine Kinase (BTK) and Vav1. In the course of phagocytosis, different phosphoinositides (PIs) are formed in the phagosomal membrane to allow the recruitment of proteins equipped with specialized lipid-interaction domains. We show that BTK and Vav1 colocalize with the lipid PI(3,4,5)P3 at the phagocytic cup, but not with diacylglycerol (DAG), which marks more mature phagosomal membranes. Inhibition of BTK affects the production of DAG and the recruitment of DAG-interacting proteins. BTK and Vav1 are essential for C. albicans immune responses, as BTK- or Vav1-deficient macrophages show reduced uptake of C. albicans and BTK- or Vav1-deficient deficient mice are more susceptible to systemic C. albicans infection. This work identifies an important role for BTK and Vav1 in immune responses against C. albicans.
| Innate immune cells eliminate pathogens by phagocytosis, a process of internalization followed by degradation of the pathogen. Germline-encoded pattern recognition receptors (PRRs) recognize pathogen-associated molecular patterns (PAMPs) on bacteria, viruses, yeast and other microorganisms. Recognition of a PAMP by its receptor initiates a coordinated sequence of events that includes the recruitment of ancillary proteins and the formation of various second messengers at -or close to- the site of initial contact with the pathogen.
Macrophages and neutrophils are the first line of defense against Candida albicans, a common cause of human fungal infections [1]. C. albicans is an opportunistic commensal yeast that is part of the normal gut microbiota [2]. Innate immune cells must therefore tolerate commensal C. albicans, yet adequately deal with its pathogenic counterpart. The major PRR involved in anti-fungal immunity is Dectin-1, a C-type lectin present on neutrophils, macrophages and dendritic cells that recognizes fungal β-glucan. Dectin-1 contains an extracellular C-type lectin domain and an intracellular ITAM-like domain essential for downstream signaling [3]. Upon activation of Dectin-1, phosphorylation of its ITAM-like domain leads to the recruitment of spleen tyrosine kinase (Syk) [4]. The role of Syk in Dectin-1-mediated phagocytosis is cell type-dependent: Syk is essential for phagocytosis in dendritic cells, but not in macrophages [3], [4]. Other proteins that interact with Dectin-1 are PKCδ [5], the tetraspanin CD37 [6], Galectin-3 [7] and TLR2 [8]. Ectopic expression of Dectin-1 in fibroblasts or kidney cells confers phagocytic capacity to these cells [3], [7]. Dectin-1 is thus a bona fide phagocytic receptor, but the detailed mechanisms that underlie Dectin1-mediated phagocytosis are not known.
Actin drives phagocytosis: formation of the phagocytic cup depends on the formation of F-actin, and closure of the phagosome requires the reversal of actin polymerization [9]. In the course of FcγR-mediated phagocytosis -the best-understood model of phagocytosis-, several phosphoinositides (PI) are formed in the phagosomal membrane, which serve as docking stations for proteins with PI-specific interaction domains. Phosphatidylinositol 4,5-bisphosphate (PI(4,5)P2) is ubiquitously present in the plasma membrane and is transiently enriched in phagocytic cups [10]. As the cup forms, phosphatidylinositol 3-kinase (PI3K) converts PI(4,5)P2 to phosphatidylinositol 3,4,5-trisphosphate (PI(3,4,5)P3) which in turn can be converted to PI(3,4)P2 by the SH2-containing inositol 5′-phosphatase (SHIP). Proteins with a Pleckstrin Homology (PH) domain can bind to PI(4,5)P2, PI(3,4,5)P3 or PI(3,4)P2, allowing their recruitment to the maturing phagosome. In addition, phospholipase C (PLCγ) acts on PI(4,5)P2 to yield the second messengers IP3 and diacylglycerol (DAG). IP3 triggers an increase in cytoplasmic Ca2+ concentration, while DAG serves as a docking site for proteins with a C1 domain. The quantity, timing and localization of PI(3,4,5)P3, PI(3,4)P2 and DAG formation vary, depending on the phagocytic receptor and the identity of the particle engaged. Formation of PI(3,4,5)P3/PI(3,4)P2 and localization of actin to the phagocytic cup are also known to occur during phagocytosis of C. albicans by macrophages [11], [12] but downstream pathways engaged by the PI during Dectin-1 phagocytosis remain to be studied in detail.
Here we describe two new interactors of Dectin-1: Bruton's Tyrosine Kinase (BTK) and the guanine nucleotide exchange factor Vav1. We provide evidence that these proteins bind to PI(3,4,5)P3-rich membranes and that BTK is involved in the production of DAG during C. albicans phagocytosis. BTK and Vav1-deficient macrophages show reduced rates of phagocytosis and BTK and Vav1-deficient mice succumb more readily to C. albicans systemic infections than wild type mice.
To facilitate imaging of phagocytosis, we applied two new imaging tools. First, we used a Candida strain that expresses a variant of blue fluorescent protein (Candida-BFP). Second, to study β-glucan exposure on C. albicans yeasts and hyphae, we site-specifically fluorescently labeled the extracellular carbohydrate recognition domain of Dectin-1 (Dectin1-CRD-Alexa647) using the bacterial enzyme sortase [7]. Candida-BFP was incubated in DMEM with 10% serum for 15, 30, 90 and 180 min followed by staining with Dectin1-CRD-Alexa647 (Figure 1A). At 15 and 30 min we observed moderately stained C. albicans yeasts with increased staining of bud scars (white arrows), congruent with increased β-glucan exposure at these sites [13]. At 90 and 180 min, formation of hyphae was extensive with strong, homogeneous β-glucan exposure. Under these growth conditions, when compared to C. albicans yeast, C. albicans hyphae thus expose higher levels of β-glucan, which is expected to affect signaling through Dectin-1.
To identify proteins that interact with Dectin-1 during phagocytosis of live C. albicans, we performed immuno-isolation experiments using the RAW-Dectin1 macrophage cell line (Esteban et al., 2011). The macrophages were co-incubated with live C. albicans for one hour, a time point that marks early hyphal formation and increased β-glucan exposure. Cells were lysed and Dectin-1, together with its interacting partners, was immunoprecipitated using anti-HA antibodies. Proteins present in these samples were analyzed and identified by SDS-PAGE, followed by LC-MS-MS. From the list of proteins, we selected two that were retrieved in complex with Dectin-1 in the samples with C. albicans, but that were absent from the control sample. These proteins were Bruton's Tyrosine Kinase (BTK) and the guanine nucleotide exchange factor Vav1.
To investigate the expression of BTK, Vav1 and other proteins already known to be involved in Dectin1-mediated phagocytosis, we incubated RAW-Dectin1 macrophages with live C. albicans. Phagocytosis continues throughout the period of coincubation, but C. albicans morphology changes from yeast form to hyphal form over time. Total cell lysates from the different time points were analyzed by immunoblot. Spleen tyrosine kinase (Syk), a known interactor of Dectin-1, was present at constant levels at all-time points (Figure 1B). Phosphorylation of Syk increased at 15 min and then waned. BTK and Vav1 were likewise present at constant levels. PLCγ2, a key enzyme in phagocytosis, was present at constant levels throughout the time course; its phosphorylation (at Y1217) was most pronounced around 90 and 120 min.
Next we confirmed the interaction between Dectin-1 and BTK, and between Dectin-1 and Vav1. Immunoprecipitation with polyclonal anti-BTK antibody, followed by immunoblotting with anti-HA, showed an increased interaction between BTK and Dectin-1 starting at 15 min, which gradually increased (Figure 1C). The interaction between Vav1 and Dectin-1 is strongest at the later time points (Figure 1C). Quantification of the Dectin-1/BTK and Dectin-1/Vav1 interactions using ImageJ software showed that the level of the Dectin-1/BTK complex peaks at 60–90 min, while the strongest Dectin-1/Vav1 interaction was observed at 180 min (Figure 1D). We conclude that BTK and Vav1 are expressed at constant levels, and that their interactions with Dectin-1 are strongest at the later time points when C. albicans has formed hyphae.
To study the subcellular localization of BTK, Vav1 and Syk, we constructed stable RAW macrophage cell lines that express these proteins as N- or C-terminal mCherry fusions in the RAW-Dectin1 background. In unstimulated cells, BTK-mCherry and mCherry-Vav1 were cytosolic, whereas mCherry-Syk localized to both the cytosol and nucleus (Figure 2, top). Next the distribution of the mCherry-tagged proteins was investigated after coincubation with Candida-BFP for 30, 90 or 180 min. At 30 min, when C. albicans yeast-form cells are present, BTK-mCherry, mCherry-Vav1 and mCherry-Syk showed a clear localization to the Candida-BFP-containing phagocytic cup (Figure 2A, arrows). After 90 and 180 min, when C. albicans had formed extensive hyphae, BTK-mCherry and mCherry-Vav1 showed enrichment in a cuff region of the phagocytic cup when engulfing Candida-BFP hyphae (Figure 2A, arrows). mCherry-Syk was more evenly distributed along the phagosomal membrane, with additional foci of enrichment outside the cuff region. There was no enrichment of BTK-mCherry or mCherry-Vav1 in membranes of closed, more mature, phagosomes. N- or C-terminal mCherry fusions with vimentin, expressed as controls in the same RAW macrophage cell line, did not show recruitment to the phagocytic cup (data not shown). 3D reconstructions of RAW macrophages in the process of ingesting C. albicans hyphae showed that the cuff regions of BTK/Vav1/Syk recruitment are cylindrical sleeves surrounding the phagosomal membrane (Figure 2B). Recruitment of BTK-mCherry, mCherry-Vav1 and mCherry-Syk to the phagocytic cup was quantified by comparing the fluorescence intensity in the cup/phagosomal membrane to that of the cytosol. Recruitment of mCherry-Syk was strongest, followed by mCherry-Vav1 and BTK-mCherry (Figure 2C).
Is recruitment of BTK and Vav1 to the phagocytic cup dependent on Dectin-1? We incubated the RAW mCherry cell lines with β-glucan-coated beads or with zymosan, the phagocytosis of which is Dectin1-dependent [3]. All three fusion proteins localized to phagosomes containing β-glucan-coated beads or zymosan. Recruitment of BTK, Vav1 and Syk therefore indeed relies on Dectin-1 (Figure S1A and B). To address the difference in geometry of yeast versus hyphal particles, we incubated the RAW mCherry cell lines with UV-killed C. albicans yeasts or hyphae. While the shape of the phagocytic cup differs, recruitment of BTK, Vav1 and Syk occurs in all cases (Figure S1C and D). Regardless of the geometry of the ingested particle, BTK and Vav1 are recruited to the phagocytic cup but not to the mature phagosome during Dectin1-mediated phagocytosis (Figure 2D). Immunofluorescence was performed to investigate the localization of Dectin1 during phagocytosis of C. albicans hyphae. Dectin1 was enriched in the cuff region of the phagocytic cup to which BTK/Vav1/Syk were recruited, but also showed areas of enrichment outside the cuff region (Figure 3). Phagosomes that contain completely internalized C. albicans yeasts or hyphae showed very little Dectin1 staining.
Different PIs are present in the phagosomal membrane when the phagocytic cup forms, as well as during phagosomal maturation. To visualize membrane PI composition, PI-binding protein domains fused with fluorescent proteins have been used as imaging tools (biosensors or probes) [14]. We transfected the RAW-Dectin1 cell line with biosensors for PI(4,5)P2 (PH-PLCδ-GFP), PI(3,4,5)P3 (PH-BTK-GFP), PI(3,4,5)P3/PI(3,4)P2 (PH-Akt-RFP) and DAG (C1-PKCδ-GFP) to study their formation during C. albicans phagocytosis.
PI(4,5)P2 was present in the plasma membrane at rest, but multiple regions of enrichment were observed at 30 and 90 min at sites where macrophages contacted Candida-BFP yeast or hyphae (Figure S2). Membranes of sealed phagosomes no longer showed PI(4,5)P2 enrichment, consistent with the reported localization of PI(4,5)P2 during FcγR-medicated phagocytosis [10]. PI(3,4,5)P3 as visualized by the BTK-PH domain showed enrichment in some, but not all, of the PI(4,5)P2-rich regions (Figure S2, arrows).
Next we investigated the localization of PI(3,4,5)P3/PI(3,4)P2 and DAG. High levels of both PIs were present in cups or phagosomes containing C. albicans. PI(3,4,5)P3/PI(3,4)P2 and DAG colocalized in some early phagosomes, but phagosomes containing only PI(3,4,5)P3/PI(3,4)P2 or only DAG were also present (Figure 4A, white and green arrows). Recruitment of PH-Akt-RFP (PI(3,4,5)P3/PI(3,4)P2) and C1-PKCδ-GFP (DAG) or both to Candida-containing cups and phagosomes was quantified at the 30-min time point (Figure 4B). More than 20% of cups/phagosomes were PI(3,4,5)P3/PI(3,4)P2-positive and a comparable percentage was positive for both PI(3,4,5)P3/PI(3,4)P2 and DAG. At this stage 4% of cups/phagosomes were positive for DAG only. PI(3,4,5)P3/PI(3,4)P2 were more prominent in phagocytic cups, while DAG was more abundant in sealed phagosomes. These results indicate that during Dectin1-mediated phagocytosis of β-glucan-exposing C. albicans yeast or hyphae, PI(3,4,5)P3/PI(3,4)P2 are formed early during initiation of phagocytosis and that DAG-rich membranes/phagosomes appear at a more advanced stage (Figure 4C).
The presence of different PIs allows the sequential docking of a specialized set of effector proteins to the membrane during phagosomal maturation. DAG is a potent second messenger, activating members of the Protein Kinase C (PKC) family. We transfected the RAW-Dectin1 cell line with GFP-tagged PKCα, PKCβ1, PKCδ, PKCε or PKCζ to determine their localization during C. albicans phagocytosis. They all localized to the C. albicans-containing phagocytic cup (Figure 4D), albeit to varying degrees. The recruitment of PKCδ and PKCε was most pronounced, while the recruitment of PKCα, PKCβ and PKCζ was moderate (Figure 4E).
BTK and Vav1 both contain a Pleckstrin Homology (PH) domain that can bind to PI(3,4,5)P3. Vav1 also contains a putative DAG-binding C1 domain, incapable of binding DAG owing to the presence of hydrophilic and non-charged residues in key binding positions [15]. The Syk polypeptide does not contain predicted PH or C1 domains and presumably localizes to the phagocytic cup through interaction of its tandem SH2 domains with the phosphorylated ITAM-like motif of Dectin-1. We investigated the possible colocalization of BTK, Vav1 and Syk with PI(3,4,5)P3 and/or DAG during phagocytosis of C. albicans. The BTK-mCherry, mCherry-Vav1 and mCherry-Syk RAW cell lines were transfected with the biosensor construct PH-BTK-GFP to visualize formation of PI(3,4,5)P3 and then incubated with Candida-BFP. The BTK-mCherry and mCherry-Vav1 fusion proteins colocalized with PI(3,4,5)P3 at the phagocytic cup, suggesting binding (Figure 5A,B), while recruitment of mCherry-Syk was more diffuse, confirming PI(3,4,5)P3-independent recruitment of Syk (Figure 5C, arrows). Next, we examined colocalization of BTK, Vav1 and Syk with DAG using the C1-PLCδ-GFP biosensor. BTK-mCherry showed some colocalization with DAG at 30 min. After 90 min, BTK-mCherry and DAG clearly localized to different regions of the phagocytic cup (Figure 5D, arrows). mCherry-Vav1 and mCherry-Syk showed a similar distribution, and neither colocalized with DAG (Figure 5E,F). BTK and Vav1 thus bind to PI(3,4,5)P3 and not to DAG during phagocytosis of C. albicans, consistent with the presence of PH domains in both proteins and a non-DAG binding C1 domain in Vav1. The colocalization of BTK and Vav1 with PI(3,4,5)P3 suggests a role for these proteins at an early stage of phagocytosis, as PI(3,4,5)P3 marks cups and immature phagosomes (see above).
Rearrangement of the actin cytoskeleton drives phagocytosis, enabling engulfment of the fungal particle by the macrophage. We investigated cytoskeletal changes during Candida phagocytosis by electron microscopy using fixation with tannic acid, a method that preserves actin structures [16]. RAW-Dectin1 macrophages were incubated with C. albicans for 30 min. Areas of decreased staining intensity were observed surrounding the C. albicans-containing phagocytic cups, indicative of actin polymerization (Figure 6A). These cuffs had a smooth appearance, were of uniform thickness across the sections examined and were distinct from the cytosol, which was more granular in appearance. We also examined actin polymerization with the biosensor LifeAct, which reports on the distribution of filamentous (F-) actin, in combination with biosensors that bind to PI(3,4,5)P3 and DAG. F-actin formation was detectable at the phagocytic cup of Candida-BFP yeast and hyphae at 30 and 90 min. There is a clear separation of DAG-rich regions of the phagosome from the F-actin rich membranes (Figure 6B). This suggests regional membrane specializations, with different functionalities and different peripheral proteins associated with them. F-actin showed perfect colocalization with PI(3,4,5)P3 at the Candida-BFP containing phagocytic cup (Figure 6C). Also, BTK-mCherry, mCherry-Vav1 and mCherry-Syk colocalize with F-actin after 30 and 90 min of C. albicans phagocytosis (Figure 6D). We conclude that PI(3,4,5)P3-rich areas are formed in the course of Dectin1-mediated phagocytosis of C. albicans and that BTK and Vav1 are recruited to these areas, with ensuing formation of F-actin.
Having established interactions of BTK and Vav1 with Dectin-1 in the course of phagocytosis and their recruitment to PI(3,4,5)P3-rich membranes, we next investigated the functional importance of these proteins in phagocytosis of C. albicans. We synthesized the highly selective irreversible BTK inhibitor PCI-32765 [17]. The IC50 of this BTK inhibitor for BTK, Tec kinase and Syk is 0.5, 78 and >10,000 nM, respectively, corresponding to a BTK selectivity of 156 fold (Tec) and >10,000 fold (Syk) [17]. In B cells, PCI-32765 irreversibly inhibited autophosphorylation of BTK (IC50: 11 nM), phosphorylation of BTK's physiological substrate PLCγ (IC50: 29 nM), and phosphorylation of downstream kinase ERK (IC50: 13 nM) [17]. RAW-Dectin1 macrophages were pre-incubated with different concentrations of BTK inhibitor (starting from 50 nM), followed by coincubation with Candida-BFP for 1 hour and staining with fluorescently labeled Concanavalin A to distinguish accessible (extracellular) particles from internalized Candida-BPF particles. Preincubation of RAW-Dectin1 macrophages with 50 nM PCI-32765 reduced uptake of C. albicans by 30% and increasing inhibitor concentrations progressively blocked phagocytosis (Figure 7A). These results indicate an important role for BTK during C. albicans phagocytosis.
PLCγ, which converts PI(4,5)P2 into DAG, is a key enzyme during phagocytosis by innate immune cells and can be regulated by BTK [18], [19]. We hypothesized that during phagocytosis of C. albicans, early localization of BTK to the phagocytic cup activates PLCγ to increase synthesis of DAG and so enables recruitment of PKC members to the phagosomal membrane. To address this possibility, RAW-Dectin1 macrophages were incubated with BTK inhibitor or the PLC inhibitor U73112, followed by coincubation with C. albicans for 1 hour. Total DAG levels in these macrophages were measured by lipid extraction, DAG kinase assays and quantification of the product, phosphatidic acid, by thin layer chromatography. Addition of the PLC inhibitor U73112 reduced total DAG levels by 50% while the BTK inhibitor did not significantly affect total DAG levels (Figure 7B and C). However, in samples without inhibitor we observed a small increase in DAG levels in the presence of C. albicans, possibly due to increased production of DAG at the phagocytic cup during phagocytosis. To examine the effect of the BTK inhibitor on local production of DAG at the phagocytic cup, we performed confocal microscopy with the C1-PLCδ-GFP biosensor and the PKCε-GFP construct in the absence and presence of the BTK inhibitor. Addition of the inhibitor strongly reduced recruitment of the C1-PLCδ-GFP biosensor and the PKCε-GFP construct to the phagocytic cup (Figure 7D and E). These results underscore the importance of investigating local changes in lipid composition as opposed to changes in total levels. BTK is thus involved in the production of DAG and the subsequent recruitment of PKC family proteins at the phagocytic cup (Figure 7F). The interactions of BTK and Vav1 with Dectin-1, their recruitment to PI(3,4,5)P3-rich membranes, the possible contribution of Vav1/BTK to actin rearrangements and the role of BTK in the production of DAG are summarized in Figure 7G.
To further investigate the contributions of BTK and Vav1 to phagocytosis we determined the phagocytic capacity of peritoneal macrophages and bone marrow-derived macrophages (BMDMs) from wild type mice and from dectin-1, btk and vav1 knockout mice. Phagocytic indices were determined after incubation of peritoneal macrophages with zymosan for 30 min or with C. albicans for 30 min or 1 hour. While Dectin1-deficient peritoneal macrophages did not phagocytose zymosan, the btk and vav1 deficient cells showed an intermediate phenotype (Figure 8A). Uptake of C. albicans by btk and vav1 knockout peritoneal macrophages at the 30-min time point was also significantly reduced compared to wild type cells (p<0.05). In addition, the btk−/− and vav1−/− BMDMs displayed a reduction in phagocytosis similar to that seen for the dectin1−/− BMDMs (p<0.05) (Figure 8A). BTK and Vav1 are thus important contributors to Dectin1-mediated uptake of C. albicans by BMDMs.
In vivo immune responses of wild type and dectin1−/−, btk−/− and vav1−/− mice to C. albicans were tested in a systemic candidiasis model. Tail vein injection of the four groups of mice with C. albicans showed that dectin1−/− mice were most susceptible to C. albicans infections, while the majority of the wild type animals survived the systemic infection. Btk−/− and vav1−/− animals displayed an intermediate phenotype, both during systemic infection with 5×104 colony forming units (CFU) (Figure 8B) and with 1×105 colony forming units (CFU) of C. albicans (data not shown). We found no statistical difference in C. albicans loads in kidneys harvested from mice about to succumb to infection (Figure 8C). In addition, histological analysis of the kidneys showed similar fungal loads (Figure 8D) and comparable immune cell invasion of the tissues in the different groups (Figure 8E). Elevated chemokine and cytokine levels in the kidney represent early responses to C. albicans infection and correlate with virulence [20]. We determined secretion of the proinflammatory cytokines TNFα and IL-6. Wild type, dectin1−/−, btk−/− and vav1−/− peritoneal macrophages secreted both TNFα and IL-6 (Figure 8F,G) in response to exposure to C. albicans. Btk−/− and vav1−/− macrophages generally secreted more TNFα and IL-6 than wild type and dectin1−/− macrophages (the difference between wild type and btk−/− TNFα secretion reached statistical significance). Next we investigated TNFα and IL-6 levels in mouse kidney during systemic C. albicans infection. Kidneys were harvested 11 days after tail vein injection with 5×104 CFU. Although levels of TNFα and IL-6 were slightly higher in the dectin1−/−, btk−/− and vav1−/− mice than in wild type, these differences did not reach statistical significance (Figure 8H,I). We conclude that disease progression in response to C. albicans systemic infection is accelerated in dectin1−/−, btk−/− and vav1−/− animals compared to wild type animals.
BTK and Vav1 are best known for their role in adaptive immunity: BCR signaling in B cells (BTK) and B and T cell development as well as activation of mature lymphocytes (Vav). Vav family members orchestrate cytoskeletal rearrangements, with Vav1 being expressed in the hematopoietic system in particular [21]. BTK and Vav also participate in innate immune reactions. BTK contributes to Fc-mediated phagocytosis [22] and was previously implicated in Dectin1-dependent pathways [23], [24], while the Vav protein family participates in Dectin-1/Mac-1 signaling in neutrophils [25]. We identified BTK and Vav1 as novel interaction partners of the β-glucan receptor Dectin-1 and confirmed their importance during C. albicans phagocytosis and immune responses during systemic infection with C. albicans. Dectin-1/BTK and Dectin-1/Vav1 complexes form during phagocytosis of live C. albicans, particularly during ingestion of C. albicans hyphae (Figure 1 and 2). BTK and Vav1 are found at membranes enriched for PI(3,4,5)P3 and colocalize with markers for F-actin (Figures 5 and 6) where BTK is involved in production of DAG at the phagocytic cup (Figure 7). Macrophages deficient in BTK or Vav1 display reduced phagocytosis and BTK- or Vav1-deficient mice succumb more readily to systemic C. albicans infections than do wild type animals (Figure 8).
BTK and Vav1 can now be added to the list of Dectin1-interacting proteins, which includes Syk [4], PKCδ [5], the tetraspanin CD37 [6], Galectin-3 [7] and TLR2 [8]. The multiple interactions of Dectin-1 with its partners reflect the complexity of Dectin-1 signaling and the (sub)complexes in which it participates. Under non-phagocytic conditions, the receptor remains at the plasma membrane, but engagement by a β-glucan ligand initiates phagocytosis and signaling from the nascent phagosome. With multiple Dectin1-interacting proteins identified, the interesting possibility of tripartite or multicomponent signalosomes arises. Dectin-1 multicomponent signalosomes exist, as complexes were found that encompassed PKCδ, Syk and Dectin-1 [5]. We tested the possibility of a Dectin1/BTK/Vav1 tripartite complex, but could not detect the three proteins in a single complex (data not shown). However, Vav1 could be a target of BTK phosphorylation, as the SH3 domain of BTK interacts with Vav1 in B cells [26]. The signaling events that occur after engagement of Dectin-1 and the relationships between BTK, Vav1 and the known Dectin-1 mediator Syk remain to be clarified. Although interaction and signaling data cannot be extrapolated to different cell types, relevant information can be extracted from the literature. Syk and Vav interact in yeast-two-hybrid experiments and in B and T cells, where Syk directly phosphorylates Vav [27].
BTK and Vav1 localize to the Candida-containing phagocytic cup, both when C. albicans yeasts and hyphae are internalized (Figure 2). During phagocytosis of hyphae, BTK-mCherry and mCherry-Vav1 show strong recruitment to a “cuff” region where engulfment of the hyphae is ongoing. C. albicans hyphae grown under these conditions expose high levels of β-glucan (Figure 1). The strong RAW-Dectin1 engagement in the cuff regions and the ensuing recruitment of BTK and Vav1 could therefore be a result of increased exposure of β-glucan on the C. albicans hyphae under these conditions. The exposure of β-glucan on C. albicans yeast and hyphae is a matter of ongoing debate.
Our data show that hyphae generated by growth in DMEM media with serum at 37°C for 90–180 min display high levels of β-glucan (Figure 1A), which is also the case during disseminated infection [28]. However, it was also reported that Dectin1 does not bind to C. albicans hyphae generated by overnight growth at 37°C in serum-free RPMI media [29], producing hyphae that appear morphologically distinct. Different growth conditions may well produce differences in β-glucan exposure and yield distinct hyphal morphologies.
Phosphoinositides (PIs) formed in the phagosomal membrane serve as docking stations for proteins with the appropriate binding domains. During phagocytosis of C. albicans, PI(4,5)P2 is enriched at sites of contact, followed by production of PI(3,4,5)P3 at the phagocytic cup and disappearance of PI(4,5)P2 and DAG-enrichment as the phagosomes seal (Figures S2 and Figure 3). BTK and Vav1 colocalize with PI(3,4,5)P3 (Figure 5), which suggests a role for these proteins at an early stage of phagocytosis. F-actin also colocalizes with BTK and Vav1 in the PI(3,4,5)P3-rich areas, which fits the known role of Vav1 in actin cytoskeleton rearrangement and suggests a possible contribution of BTK to F-actin formation (Figure 6). The observation that Vav1 colocalizes with PI(3,4,5)P3 but not with DAG supports the notion that the C1 domain of Vav1 is not a functional DAG-binding domain [15]. During CR3- and FcγR-mediated phagocytosis, actin tail formation follows local production of PI(3,4,5)P3 [30]. BTK also localizes to actin-rich cups during FcγR-mediated phagocytosis [22]. Our data and those in the literature emphasize similarities between Dectin1- and FcγR-mediated phagocytosis.
The enzyme PLCγ converts PI(4,5)P2 to DAG at the phagocytic cup and the PLC inhibitor U73112 blocked C. albicans phagocytosis by the RAW-Dectin1 macrophages (data not shown). PLCγ is also essential for FcγR-mediated phagocytosis [10]. DAG-rich membranes recruit proteins of the PKC family that have a DAG-interacting C1 domain [31], [32]. All PKC isoforms examined localized to the Candida-containing phagocytic cup, with the Ca2+-independent family members PKCδ and PKCε displaying the strongest recruitment (Figure 3). Conventional PKCs require increased intracellular Ca2+ levels for activation and binding to DAG [33] but DAG-independent recruitment of the different PKC isoforms, for example via protein-protein interactions, remains possible as well. PKCδ and PKCε are involved in early steps of phagocytosis: PKCδ interacts with Dectin-1 and Syk and is required for phagocytosis of zymosan [5] whereas PKCε enhances FcγR-mediated phagocytosis [34]. Conventional PKCs also contribute to other processes, such as the generation of the respiratory burst, but are dispensable for FcγR-mediated internalization [35]. Further characterization of the contributions of individual PKC members to FcγR- and Dectin1-mediated phagocytosis is necessary.
BTK and Vav1 are important for phagocytosis of C. albicans, as btk−/− and vav1−/− peritoneal macrophages and BMDMs displayed reduced phagocytosis (Figure 8). BTK- and Vav1-deficient mice are also more susceptible to systemic infections with C. albicans. Dectin1-, BTK- and Vav1-deficient mice succumb earlier to infections than do wild type mice, but fungal burdens, immune cell invasion and cytokine levels in the kidney are comparable (Figure 8). While the phenotype of the BTK- and Vav1-deficient mice might be due to reduced phagocytosis by macrophages, the mice used here are complete knockouts. We therefore must remain vigilant to the possibility that defects in macrophage or innate immune cells may not be solely responsible for the results reported here. Btk−/− mice have B cell defects [36], [37] while vav1−/− mice have reduced numbers of T and B cells [38], [39]. Although B and T cells are not thought to play a major role in immune responses during systemic C. albicans infections, a (minor) contribution cannot be excluded. In addition to their role in phagocytosis, BTK and Vav1 might contribute to other innate processes related to C. albicans immune responses, such as production of reactive oxygen species or cytokines. The complexities of these interconnections clearly require further study.
BTK and PLCγ are functionally connected, as knockdown of BTK resulted in reduced PLCγ phosphorylation in response to stimulation of the TREM-1/DAP12 pathway in a lymphoma cell line [19]. In our RAW-Dectin1 cell line, pharmacological inhibition of BTK resulted in decreased phagocytosis, reduced DAG levels and compromised recruitment of PKCε to the phagocytic cup (Figure 7). We therefore hypothesize that in this setting BTK is responsible for activation of the DAG-producing enzyme PLCγ. In FcγR-mediated phagocytosis, PLCγ phosphorylation and recruitment to the phagocytic cup are Syk-dependent [10], [40]. It remains to be established if Syk or BTK or both are responsible for PLCγ phosphorylation and/or activation during Dectin1-mediated phagocytosis. BTK was previously shown to be involved in Dectin1-dependent arachidonate release by macrophages in response to incubation with zymosan or particulate β-glucan [23], [24]. Phosphorylation of BTK on tyrosine 223 and phosphorylations of PLCγ2 were induced by incubation with zymosan or particulate β-glucan. However, incubation with the BTK inhibitor LFM-A13 did not reduce phosphorylation of PLCγ2 during incubation with zymosan [24].
The role for Vav proteins remains incompletely understood. Our data add to previous observations that Vav proteins are instrumental in phagocytosis, and participate in a cell type-dependent manner. vav1/vav2/vav3 triple-knockout macrophages ingest IgG-opsonized erythrocytes normally [41] but vav1−/− and vav3−/− neutrophils are deficient in FcγR-mediated phagocytosis [42]. With regards to fungal particles, in microglia phosphorylation of Vav1 is induced by particulate β-glucan and this phosphorylation is affected by a Src family kinase inhibitor as well as a Syk inhibitor. Vav1 knockdown in a microglial cell line resulted in reduced uptake of β-glucan particles [43]. vav1/vav3 double knockout neutrophils show reduced binding to zymosan, and vav1/vav3 mice have increased susceptibility to C. albicans infection [25] accompanied by reduced PLCγ phosphorylation. While in our hands vav1 peritoneal macrophages have defects in both C. albicans and zymosan uptake (Figure 8A), Li et al. reported that thioglycollate-induced peritoneal macrophages do not display reduced phagocytosis of zymosan [25]. Whether these differences are due to macrophage activation status remains to be established. In addition, the precise role of Vav1 and a possible link to PLCγ during phagocytosis by macrophages remains incompletely understood.
Differences in cell wall composition between C. albicans and a non-pathogenic yeast like Saccharomyces cerevisiae might influence recruitment of downstream factors. A systematic assessment of the involvement of BTK, PLCγ, DAG and PKC family members during Dectin1-mediated phagocytosis of C. albicans, S. cerevisiae and other fungi should help clarify their roles.
Animals used in this study were housed at the Whitehead Institute for Biomedical Research, which is certified by the United States Office of Laboratory Animal Welfare (OLAW) under the guidance of the Public Health Service (PHS) Policy on Humane Care and Use of Laboratory Animals. Whitehead Institute's Animal Welfare Assurance was approved 11/3/2009 (IACUC, A3125-01) All studies were carried out in accordance with procedures approved by the Massachusetts Institute of Technology Committee on Animal Care (Ploegh lab, CAC# 1011-123-14).
C. albicans strain SC5314 was cultured in YPD + Uri (2% bactopeptone, 1% yeast extract, 2% glucose and 80 µg/ml uridine) at 30°C. To generate a blue fluorescent protein (BFP)-expressing C. albicans strain, the GFP sequence of the pENO1-yEGFP3-NAT plasmid [28] was replaced with the TagBFP sequence (Evrogen) with codon usage adapted for C. albicans. C. albicans SC5314 was transformed with the pENO1-TagBFP-NAT plasmid and selected with 200 µg/ml nourseothricin (Werner Bioagents, Jena, Germany) resulting in the Candida-BFP strain. The RAW-Dectin1-LPETG-3×HA cell line (RAW-Dectin1) [7] was used for most phagocytosis experiments. Cells were grown in DMEM medium with 10% inactivated Fetal Bovine Serum (IFS) at 37°C and 5% CO2. For the production of retrovirus, HEK293T cells (ATCC) were transfected using TransIT transfection reagent (Mirus) and virus-containing supernatant was harvested after 24 hours.
The imaging constructs used in this study are listed in Table 1. A Dectin1-CRD-LPETG-His bacterial expression vector was cloned and expressed as described for Dectin1-CRD [13]. Mammalian expression vectors were constructed for N- or C-terminal mCherry fusions in vectors based on the retroviral plasmid pMSCVpuro (Clontech) (pMSCVpuro-mCherry-N and pMSCVpuro-mCherry-C). The BTK, Vav1 and Syk open reading frames were cloned from mouse spleen cDNA into the tagging vectors, resulting in N- or C-terminal fusion of all three genes (mCherry-BTK, BTK-mCherry, mCherry-Vav1, Vav1-mCherry, mCherry-Syk and Syk-mCherry). The vectors were used to create stable cell lines in the RAW-Dectin1 background by retroviral transduction and selection with puromycine. The resulting six cell lines were tested for expression of the mCherry-fused proteins by immunoblotting and microscopy and the BTK-mCherry, mCherry-Vav1 and mCherry-Syk cell lines were selected for further experiments. For expression of the biosensor constructs in the RAW macrophage cell lines, cells were transfected using Fugene (Roche).
To identify proteins that interact with Dectin-1 during phagocytosis, RAW-Dectin1 macrophages were incubated with live C. albicans at MOI 5 for 1 hour or left unchallenged. Cells were harvested by scraping into ice-cold PBS and lysed in NP40 buffer (25 mM Tris pH 7.4, 150 mM NaCl, 5 mM MgCl2 with 0.5% NP40 and protease inhibitors). Epitope-tagged Dectin-1 was immunoprecipitated from the total lysates with anti-HA beads (Roche). Eluates were run on a SDS-PAGE gradient gel and silver stained to visualize proteins. Each lane was cut into regions according to molecular weight, which were then reduced, alkylated and subjected to trypsin digestion. The resulting peptides were extracted, concentrated in vacuo, and analyzed by reverse-phase chromatography and tandem mass spectrometry. The resulting CID spectra were searched against a species-specific database generated from NCBI's non-redundant database using SEQUEST. For the generation of anti-BTK and anti-Vav1 antisera, BTK and Vav1 were cloned from mouse spleen cDNA into bacterial expression vector pET28a with an N-terminal His tag (pET28a-BTK and pET28a-Vav1). Vectors pET28a-BTK and pET28a-Vav1 were used to transform Rosetta cells and transformants were induced with IPTG for protein expression. His-Vav1 was isolated from the soluble fraction and His-BTK was isolated from the insoluble fraction in 8 M urea. Both proteins were purified using NiNTA beads and BTK was refolded by stepwise dialysis to eliminate urea, followed by FPLC purification of the peak containing the monovalent BTK. Purified Vav1 and BTK were injected in rabbits and serum was harvested and used for immunoblot and immunoprecipitation experiments (anti-BTK and anti-Vav1). Other antibodies used were: anti-Syk (Cell Signaling), anti-phospho-Syk (Tyr525/526; Cell Signaling), anti-PLCγ2 (Cell Signaling), anti-phospho-PLCγ2 (Tyr1217; Cell Signaling), anti-p97 [44], anti-HA-HRP (Roche). For immunoblotting, protein extracts were separated on 8% or 12% SDS-polyacrylamide gels and transferred to a nitrocellulose membrane using a semi-dry system. For BTK and Vav1 immunoprecipitation experiments, cells were lysed in NP40 buffer followed by immunoprecipitation with 2 µl of anti-BTK or anti-Vav1 antisera and 30 µl of Protein-A beads (Repligen). Beads were washed and eluates were analyzed using SDS-PAGE gels.
Confocal images were collected in the W.M. Keck Facility for Biological Imaging using a PerkinElmer Live Cell Imaging spinning disk confocal system and Volocity software. The PerkinElmer Live Cell Imaging spinning disk confocal system was mounted on a Zeiss Axiovert 200M with a 100× 1.4NA Plan-Apochromat objective. Excitation light was generated by gas and solid state lasers (Argon laser for 488 nm, Krypton laser for 568 nm, solid state laser for 405 nm and 647 nm) and passed through an AOTF for wavelength selection and laser power control. A quadruple bandpass filter separated the excitation and emission light inside the CSU-22 confocal scanhead (Yokogawa) and a filter wheel (Prior Scientific) provided selection of emission filters (TagBFP & RFP: dual-band 445/60 and 615/70 nm; GFP: 527/55 nm). Volocity image acquisition software was used to capture images from a Hamamatsu Orca-ER cooled-CCD camera and to control all the equipment. For 3D reconstructions of phagocytic cells, Z planes were acquired at 0.15 µM distance and Volocity image acquisition software was used to create the XYZ views. Electron microscopy sections were examined using a FEI Tecnai Spirit at 80 KV. Routine morphology was performed by trimming and fixing the tissue in 2.5% gluteraldehyde, 3% paraformaldehyde with 5% sucrose in 0.1 M sodium cacodylate buffer (pH 7.4) and 0.2% tannic acid. Samples were post fixed in 1% osmium in veronal-acetate buffer. The tissue was stained in block overnight with 0.5% uranyl acetate in veronal-acetate buffer (pH 6.0), then dehydrated and embedded in em812 resin. Sections were cut on a Leica Ultracut UCT microtome with a Diatome diamond knife at a thickness setting of 50 nm, stained with uranyl acetate, and lead citrate. The sections were examined using a FEI Tecnai Spirit at 80 KV.
For general confocal microscopy, Candida-BFP was added to RAW-Dectin1 macrophages at an MOI of 10, fixed with 4% PFA in PBS and mounted on slides in 50% glycerol. β-1,3-glucan conjugated beads were a kind gift of Jatin Vyas and prepared as described [45]. Zymosan A (Sigma) was labeled with Alexa647 carboxylic acid, succinimidyl ester (Invitrogen) by incubation in 0.1 M Na2CO3 at room temperature. Candida-BFP was UV-killed by exposure to 100.000 µJ/cm2 in a UV-crosslinker for four rounds. Recruitment of fluorescent proteins to the phagocytic cup or phagosome was quantified using ImageJ software according to the method of Flannagan and Grinstein [46]. Phagocytic indices of RAW-Dectin1 cells, BMDMs or peritoneal macrophages were determined by incubation with Candida-BFP or zymosan-Alexa647 at MOI 10 for 30 minutes or 1 hour. Cells were fixed in 4% PFA and stained with Concanavalin A-FITC (Sigma) to distinguish unengulfed yeasts. Inhibitors were added to the media at the indicated concentrations for 1 hour prior to incubation with Candida-BFP. Images were aquired by confocal microscopy and the number of intracellular Candida-BFP per macrophage was determined by counting 75-200 cells per experiment. Dectin1-CRD-LPETG was incubated with Staphylococcus aureus sortase A enzyme and GGG-Alexa647 probe resulting in Dectin1-CRD-LPETGGG-Alexa647 that was used for staining of Candida-BFP yeasts and hyphae. For immunofluorescence, cells were grown on coverslips and fixed in 4% PFA in PBS, washed with PBS and incubated in 50 mM NH4CL in PBS for 10 min. Next, cells were incubated in Binding Buffer (0.1% Saponin, 0.2% BSA in PBS) for 30 min followed by incubation in Binding Buffer with anti-HA-Alexa488 (Invitrogen) antibody for 60 min, several washes with PBS and mounting for spinning disk confocal microscopy.
4-Amino-3-(4-phenoxyphenyl)-1H-pyrazolo[3,4-d]pyrimidine (1) was prepared from 4-phenoxybenzoic acid and malonitrile as described (International Patent Publication No. WO 01/019829 and [17]. Alkylation of pyrazole (1) with 3-methanesulfonyl N-Boc hydroxypiperidine (2) followed by removal of the Boc-protecting group and acylation with acryloyl chloride gave the racemate of PCI-32756 in three steps (Figure S3). All chemicals were of commercial sources and were used as received. DriSolv anhydrous CH2Cl2, DriSolv anhydrous MeOH, DriSolv anhydrous DMF were purchased from EMD Chemicals. Redistilled, anhydrous N,N′- diisopropylethylamine (DiPEA), trifluoroacetic acid (TFA), triisopropylsilane (TIS) N-methylpyrrolidone (NMP) was obtained from Sigma-Aldrich. LC-ESI-MS analysis was performed using a Micromass LCT mass spectrometer (Micromass MS Technologies, USA) and a Paradigm MG4 HPLC system equipped with a HTC PAL autosampler (Michrom BioResources, USA) and a Waters Symmetry 5 µm C8 column (2.1×50 mm, MeCN∶H2O (0.1% formic acid) gradient mobile phase, 150 µL/min). HPLC purifications were achieved using an Agilent 1100 Series HPLC system equipped with a Waters Delta Pak 15 µm, 100 Å C18 column (7.8×300 mm) using A: H2O, B: MeCN and C: 1% aqueous trifluoroacetic acid as mobile phase (3 mL/min). (R/S)-1-Boc-3-Hydroxypiperidine (1.05 g, 5 mmol) was dissolved in CH2Cl2 (25 mL) and subsequently triethylamine (1.39 mL, 10 mmol) and methanesulfonyl chloride (0.394 mL, 5.1 mmol) were added. After stirring overnight, the reaction was concentrated under reduced pressure, redissolved in ethyl acetate, washed with water and brine, dried over MgSO4 and concentrated in vacuo. The crude mesylate was dissolved in anhydrous DMF (20 mL). Pyrazole 1 (1.02 g, 3.33 mmol) and potassium carbonate (0.92 g, 6.66 mmol) were added and the reaction was stirred until TLC analysis showed complete conversion. The reaction was diluted with water and extracted with CH2Cl2. The organic layer was dried over MgSO4, concentrated in vacuo. Purification over silica gel chromatography (CH2Cl2→MeOH/CH2Cl2) gave intermediate 3. Intermediate 3 was dissolved in dioxane (20 mL) and freshly prepared hydrochloric acid (35 mmol) was added. The solution was stirred for 1 h, concentrated in vauo. The crude amine (1 g, 2.58 mmol) was redissolved in CH2Cl2 (10 mL). To this was added Et3N (1.8 mL, 12.9 mmol) and acryloyl chloride (0.22 mL, 2.71 mmol). After 5 h, the reaction was quenched by the addition of water. The solution was extracted and the organic layer was dried over MgSO4 and concentrated in vacuo. The crude product was purified by reverse phase HPLC (28→43%B in 20 min, 3 mL/min) affording the title compound (46.3 mg, 0.105 mmol) as a white solid. LC/MS: Rt 8.52 min; linear gradient 5→80% B in 10 min; ESI/MS: m/z = 441.2 [M+H]+. 1H NMR (400 MHz, CD3OD) δ ppm 8.39 (s, 1H), 7.69-7.66 (m, 2H), 7.44-7.39 (m, 2H), 7.21-7.08 (m, 5H), 6.82 (dd, J = 16.8, 10.8 Hz, 0.6H), 6.66 (dd, J = 16.8, 10.8 Hz, 0.4H), 6.17 (d, J = 16.8 Hz, 0.6H), 6.15 (d, J = 16.8 Hz, 0.4H) 5.77 (d, J = 10.8 Hz, 0.6H), 5.65 (d, J = 10.8 Hz, 0.4H), 4.95-4.93 (m, 1H), 4.56 (d, J = 12.4 Hz, 0.6H), 4.24 (d, J = 11.2 Hz, 1H), 4.06 (d, J = 14.0 Hz, 0.6H), 3.89 (dd, J = 12.8, 8.8 Hz, 0.4H), 3.58 (dd, J = 12.4, 9.6 Hz, 0.6H), 3.42-3.36 (m, 1H), 2.45-2.34 (m, 1H), 2.28-2.23 (m, 1H), 2.15-2.05 (m, 1H), 1.80-1.68 (m, 1H). 2.72 (t, J = 7.6 Hz, 2H), 2.15 (dt, J = 7.2, 2.8 Hz, 2H), 1.96 (d, J = 2.8 Hz, 1H), 1.82 (quin., J = 7.6 Hz, 2H), 1.52 (quin., J = 7.2 Hz, 2H).
RAW-Dectin1 macrophages were coincubated with Candida albicans in 6-well dishes for 1 hour, washed with phosphate-buffered saline and subjected to lipid extraction [47]. The chloroform/methanol phase was dried under N2 and DAG kinase assays were performed as described [48]. See supplementary materials for further details. The dried lipids were dissolved in 40 µl of solubilizing buffer (7.5% octyl-ß-D-glucoside and 5 mM cardiolipin in 1 mM diethylenetriaminepenta acetic acid (DETAPAC, pH 7.0) by vigorously vortexing for 20 sec and incubating at RT for 10 min. Then, 100 µl of 2× reaction buffer (100 mM imidazole HC1, pH 6.6, 100 mM NaCl, 25 mM MgCl2, and 2 mM EGTA), 4 µl of 100 mM freshly prepared DTT and 20 µl of E. coli DAG kinase (Sigma-Aldrich)) were added while keeping the samples on ice. The reaction was initiated by addition of 3 µCi [γ33P]-ATP prepared by dilution in 20 µl of 1 mM DETAPAC, pH 6.6. After vortexing briefly, the reaction was incubated at 25°C for 30 min. Lipids were extracted as described above and the reaction products were analyzed by TLC, which was developed in acetone followed by CHCl3/MeOH/acetic acid (65/15/5 [vol/vol/vol]) solution. Radiolabelled lipids were detected by exposure to imaging screens (BAS-MS; FujiFilm), scanned on a BAS-2500 (FujiFilm)), and quantified with Quantity One software.
Animals were housed at the Whitehead Institute for Biomedical Research and maintained according to protocols approved by the Massachusetts Institute of Technology Committee on Animal Care. C57BL/6 mice were purchased from Jackson Labs, dectin1−/− [49], btk−/− [36] and vav1−/− mice [39] were kind gifts from Stu Levitz, Whasif Khan and Victor Tybulewicz, respectively. Bone marrow-derived macrophages (BMDMs) were differentiated from mouse bone marrow by growth in DMEM (high glucose; Gibco) with 10% HI-FBS (Hyclone) and 5% M-CSF-containing culture supernatant from L929 cells. Experiments were performed after 7 days of differentiation. Peritoneal macrophages were harvested by peritoneal lavage with upto 10 ml PBS. Cells were seeded for experiments in DMEM (high glucose; Gibco) with 10% HI-FBS (Hyclone) and used for experiments the next day. For cytokine analysis, macrophages were incubated with C. albicans and supernatants were harvested after 16 hours. Cytokine concentrations were determined by Discovery assay cytokine array by Eve Technologies. For the mouse model of systemic candidiasis, 5×104 or 1×105 CFU of C. albicans SC5314-derived strain Candida-BFP was administered intraveneously to age-matched C57BL/6 wild type, dectin-1, btk and vav1 −/− mice in a final volume of 200 µl in PBS. Mice were weighed and monitored daily and euthanized when >20% of initial body weight was lost. Kidneys were harvested for enumeration of fungal burden, histology or cytokine analysis. To determine fungal burden, kidneys were homogenized and lysates were plated on YPD plates with antibiotics. For histology, tissues were fixed in 4% formalin in PBS, embedded and sectioned in paraffin and slides were stained with Hematoxylin and Eosin (H&E) or Gomori Methenamine Silver (GMS). For cytokine analysis, kidneys were homogenized in 1 ml PBS followed by cytokine analysis using a Th1/Th2/Th17 mouse cytometric bead array (BD Biosciences) and LSR Flow Cytometer according to the manufacturer's directions.
For statistical analysis unpaired t tests were used with 95% confidence interval. All graphs show means and standard deviations of three independent experiments.
Supplemental information includes three figures.
|
10.1371/journal.pbio.1000497 | Protein Evolution by Molecular Tinkering: Diversification of the Nuclear Receptor Superfamily from a Ligand-Dependent Ancestor | Understanding how protein structures and functions have diversified is a central goal in molecular evolution. Surveys of very divergent proteins from model organisms, however, are often insufficient to determine the features of ancestral proteins and to reveal the evolutionary events that yielded extant diversity. Here we combine genomic, biochemical, functional, structural, and phylogenetic analyses to reconstruct the early evolution of nuclear receptors (NRs), a diverse superfamily of transcriptional regulators that play key roles in animal development, physiology, and reproduction. By inferring the structure and functions of the ancestral NR, we show—contrary to current belief—that NRs evolved from a ligand-activated ancestral receptor that existed near the base of the Metazoa, with fatty acids as possible ancestral ligands. Evolutionary tinkering with this ancestral structure generated the extraordinary diversity of modern receptors: sensitivity to different ligands evolved because of subtle modifications of the internal cavity, and ligand-independent activation evolved repeatedly because of various mutations that stabilized the active conformation in the absence of ligand. Our findings illustrate how a mechanistic dissection of protein evolution in a phylogenetic context can reveal the deep homology that links apparently “novel” molecular functions to a common ancestral form.
| Many protein families are so diverse that it is hard to determine their ancestral functions and to understand how their derived functions evolved. The existence of so many different functions within protein families often creates the impression that complex, novel functions must have evolved repeatedly and independently. Nuclear receptors (NRs) are a large family of related proteins that regulate key biological processes in animals by binding to specific DNA sequences and triggering expression of nearby target genes. Many NRs are activated by a specific hormone or other small molecule, but some do not require a ligand, and still others are incapable of activating gene expression and so act primarily as repressors of transcription. To understand how the functional diversity of NRs evolved, we reconstructed the structural and functional characteristics of the ancient protein from which the entire family evolved, using genomic, biochemical, functional, and structural analyses in a phylogenetic framework. We show, contrary to current belief, that the ancestral NR was a ligand-activated transcriptional activator that existed in the earliest period of animal evolution. Our analysis reveals how the extraordinary functional diversity of modern receptors was generated by subtle tinkering with this ancestral template—slightly reshaping the ligand cavity, stabilizing the protein's active conformation so it no longer required a ligand, or disabling the protein's capacity to activate transcription without affecting its other properties. We predict that, when sufficient data are gathered to allow detailed evolutionary reconstructions in other protein families, it will become apparent that most protein functional diversity evolved by tinkering with ancient functions; invoking the evolution of wholesale “novelty” will seldom be necessary.
| By sequencing genomes of taxa occupying key positions in the metazoan tree of life, it has become possible to infer when important animal gene families originated and proliferated [1]–[3]. Sequence data alone, however, cannot yield insight into the functions and structures of ancient proteins or the processes by which their descendants evolved. Further, many gene families have diversified so extensively that comparisons of extant proteins from model organisms are insufficient to reveal which functions are ancestral and which are derived. In principle, it should be possible to reconstruct the history of a protein family by phylogenetically analyzing the underlying structural mechanisms that produce functional diversity among densely sampled members of the family. Such a strategy would be analogous to detailed studies of the evolution of animal development, which have revealed the deep homology of diverse morphologies in distant lineages and the mechanisms by which they evolved from common ancestral forms [4].
The members of the superfamily of nuclear receptor (NR) transcription factors, for example, are regulated in diverse ways—by ligands, postranslational modifications, and association with other proteins or DNA—depending on the cellular context [5]. NRs have a modular domain structure, including a highly conserved DNA-binding domain (DBD) and a moderately conserved ligand-binding domain (LBD)—which in most receptors contains a ligand-regulated transcriptional activation function—along with extremely variable hinge and N-terminal domains. There is considerable diversity in the functions of NR LBDs, which can be roughly classified into three major groups. In one class, the LBD's transcriptional function can be activated by a specific hydrophobic ligand, such as a steroid, retinoid, or fatty acid; the ligand binds in a deep internal cavity, remodeling and stabilizing the LBD's conformation to generate a new binding surface for coactivator proteins, which increase transcription of nearby genes [5]. The second class of NRs are ligand-independent transcription factors, often called “constitutive” receptors, the LBDs of which can adopt the active conformation and activate gene expression in the absence of a ligand or other modifications. Some members of this class lack the internal cavity and are not known to bind any ligands, whereas others do bind hydrophobic molecules, which up- or down-regulate their baseline activity [6]–[9]. In the third class of NRs, the LBD lacks the capacity to interact with coactivators, so these receptors function primarily as transcriptional repressors that occupy NR response elements or dimerize with and thereby silence other NRs [10]–[12].
It is widely believed that the NR superfamily evolved from a ligand-independent transcriptional activator, with binding of different ligands gained independently in numerous NR lineages [13],[14]. The alternate view—that NRs evolved from a liganded ancestor, with ligand-dependence lost in the lineages leading to the ligand-independent receptors—has received little attention. These two hypotheses exemplify opposite views on the general nature of molecular evolution and the origin of complex functions. The hypothesis that the ancestral NR was ligand-independent implies that a complex molecular function—allosteric regulation of transcription by binding a ligand—evolved de novo many independent times, requiring evolution to repeatedly create novelty and complexity [15],[16]. In contrast, the hypothesis of a ligand-activated ancestor implies that evolution produced new functions primarily by subtle tinkering with a conserved ancestral mechanism [4],[17], which allowed receptors to accommodate new molecular partners or lose dependence on those partners because of mutations that modified or degraded existing functions.
Several limitations have impeded rigorous inference about the ancestral NR's characteristics and the diversification of the superfamily. First, the root of the gene family phylogeny has been ambiguous, leaving unknown the location of the ancestor relative to its descendants. Second, the topology of the NR phylogeny has been uncertain, because of limited sequence sampling and/or use of outdated phylogenetic methods. Third, the functions of NRs in taxa branching near the root of the metazoan phylogeny have not been characterized. Finally, whether distantly related NRs with similar functions share homologous or convergent underlying mechanisms has not been determined. Recently acquired information—including genome sequences from basal metazoans and extensive data on NR structures and functions—along with improved algorithms for phylogenetic analysis of large datasets, now allow these barriers to be overcome. Here we report on biochemical, functional, structural, and phylogenetic analyses of the NR superfamily, which allow us to reconstruct the characteristics of the ancestral nuclear receptor and understand how the functional diversity of NR LBDs evolved.
The root of the NR phylogeny has been unknown because of uncertainty about the relative ages of the various NR family members. NRs appear to be a metazoan innovation, because they are absent from the genomes of choanoflagellates, fungi, plants, and prokaryotes. Until recently, however, all fully sequenced animal genomes have come from protostomes and deuterostomes, both of which contain virtually all the major NR subfamilies [18]; these data indicate only that most NR diversity was already established by the time of the protostome-deuterostome ancestor. To determine the most ancient NR lineages, we identified NRs in the newly sequenced genome of the demosponge Amphimedon queenslandica—a representative of the Porifera, the most anciently branching metazoan phylum based on whole-genome phylogenies [19]. We found that the A. queenslandica genome contains two NRs, which we refer to as AqNR1 and AqNR2. We amplified transcripts of each by polymerase chain reaction, verified their sequences, and analyzed their developmental expression using in situ hybridization. AqNR2 is expressed ubiquitously, whereas AqNR1 is expressed in a range of cells that contact the external environment and possess apico-basal polarity (Figure S1). We also identified NRs in genomes from two other recently sequenced early-branching lineages, the placozoan Trichoplax adhaerens and the cnidarian Nematostella vectensis, which contain 4 and 17 NRs, respectively (see also [20]). These results point to very limited NR diversification before the origin of the Eumetazoa and indicate that basal metazoan species have the potential to shed light on early NR evolution.
To determine the phylogeny of the NR superfamily, we used model-based phylogenetics to analyze a taxonomically diverse database of 275 NR protein sequences (Figure 1, Table S1). The alignment includes the DBDs and LBDs of the complete NR complements in 11 sequenced genomes from eight broadly sampled animal phyla, plus NRs from 30 other species strategically chosen to maximize phylogenetic accuracy and minimize redundant signal [21]. Unlike previous studies, which used sparser sequence sampling and/or less powerful methods [14],[18],[22], phylogenetic analysis of this alignment using maximum likelihood yielded a well-resolved phylogeny (Figures 1, S2) with strong support for the placement of the basal metazoan sequences and for the relationships among most major NR families. (A few aspects of the topology, however, had weak support, such as whether the SF-1 class has a monophyletic or paraphyletic relationship to the group containing the steroid hormone receptors.) We also conducted Bayesian Markov Chain Monte Carlo (BMCMC) methods using a slightly smaller 174-sequence dataset, assembled by removing sequences at the ends of very long branches and multiple orthologs within the same phylum. This analysis recovered a nearly-identical phylogeny to the maximum likelihood analysis (Figures 1, S3).
The phylogeny is unlikely to be an artifact of the presence of rapidly evolving sites or taxa. When the 40% of sites with the fastest evolutionary rates were removed from the analysis, only the placement of the RXR group was affected (Figure S4). Further, maximum likelihood analysis of the reduced 174-sequence dataset—from which the longest terminal branches were removed—yielded the same phylogenetic relationships as the unreduced analysis (Figure S5).
The phylogeny can be rooted at a single most parsimonious location between AqNR1 and AqNR2, allowing the ancestral NR to be located on a specific branch of the phylogeny. This rooting provides a coherent history of NR expansion by gene duplication with few subsequent losses (Figure 2). All alternative rootings that place other NR lineages in a basal position require many additional duplications and losses (Figure 2A). For example, placing the clade of ligand-independent estrogen related receptors (ERRs) as the outgroup requires two additional duplications and 12 additional losses compared to the optimal root; placing the ligand-independent NR4 class as the outgroup requires two additional duplications and 15 additional losses.
The phylogeny indicates that AqNR2 is orthologous to the fatty acid-binding HNF4 family and that AqNR1 is the unduplicated ortholog of all other NRs. This phylogeny indicates that the last common ancestor of all Metazoa contained two NRs—one ortholog of HNF4 and one of AqNR1, which subsequently gave rise to all other NR classes (Figure 2). After the divergence of demosponges from other metazoans but before the split of Cnidaria from the Bilateria, nine more duplications gave rise to most of the major recognized NR types, except for those in the named classes NR1 and NR4, which proliferated during the interval between the cnidarian-bilaterian ancestor and the protostome-deuterostome ancestor (Figure 2). Many NR subfamilies diversified further within the vertebrates.
Support for the placement of AqNR2 with the HNF4s and of AqNR1 as sister to all other NRs is strong, with posterior probabilities of 1.0 and 0.98, chi-square confidence values of 1.0 and 1.0, and approximate likelihood ratios of 471 and 112, respectively (Figures 1, S2–S3). The next-best rearrangement of the relationships between the sponge receptors and the rest of the NRs has a likelihood several orders of magnitude lower than that of the ML tree. This alternate phylogeny (Figure S6) would place AqNR1 and AqNR2 as sister paralogs specific to the sponge lineage. It would imply that the ancestral metazoan contained a single NR; duplication of this ancestral gene in the sponges would have yielded AqNR1 and AqNR2, and the first duplication that separated the HNF4 group from the rest of the NR superfamily would have occurred in the Eumetazoa after they diverged from demosponges. The rest of the superfamily's history would remain unchanged.
Hypotheses concerning the functions of the ancestral proteins are strongly affected by the functions of proteins that branch off the family phylogeny near its root. To understand how the functions of the NR LBD evolved, we experimentally characterized the capacity of AqNR1 and AqNR2 LBDs to bind and be regulated by ligands. In a reporter gene assay, AqNR1 had very weak intrinsic activity—less than 2-fold activation—when incubated with serum from which hydrophobic small molecules were stripped using dextran-charcoal. When treated with complete serum, however, AqNR1 transcription increased by 30-fold, suggesting that the receptor is activated by a hydrophobic ligand that is present in mammalian serum, such as a fatty acid or steroid (Figure 3A).
To characterize AqNR1's potential ligand, we expressed and purified AqNR1-LBD in bacteria, extracted bound hydrophobic molecules, and used mass spectrometry to identify the bound compounds. Like mammalian HNF4 [23], AqNR1-LBD bound an array of bacterial free fatty acids (FAs) with tail lengths ranging from 16 to 19 carbons, with preference for 18∶0 and 18∶1 fatty acids (Figures 3B, S7). When AqNR1 was incubated with complete mammalian serum, palmitic acid was the dominant FA bound, along with lower proportions of 18∶0 and 18∶1 FAs (Figure 3B). We then confirmed and quantified FA binding by the purified AqNR1-LBD using an enzymatic assay. As predicted, we found that AqNR1-LBD binds both E. coli and mammalian FAs; receptor occupancy by FAs approximately doubles when the protein is treated with complete serum but does not increase upon treatment with stripped serum (Figure 3C).
To more directly test the hypothesis that AqNR1 binds and is activated by FAs, we characterized the functional effects of mutations in the predicted AqNR1 ligand pocket. We first predicted the structure of AqNR1-LBD using a homology model based on the X-ray crystal structure of mammalian HNF4, the NR with the highest sequence similarity to AqNR1. The predicted structure (Figure 4A) indicates that AqNR1 is likely to have a large ligand pocket (835 Å3) with ample space to accommodate FAs. As in the crystal structure of mammalian HNF4s, the FA in AqNR1 is predicted to be coordinated by a hydrogen bond from a conserved arginine (Arg226 in rat HNF4α, Arg492 in AqNR1) to the FA's carboxylate oxygen; further, packing interactions between hydrophobic amino acids that line the pocket and the ligand's tail are also conserved.
We then used directed mutagenesis and functional assays to experimentally test the hypothesis that AqNR1 binds and is activated by FAs in a manner conserved with mammalian HNF4. As predicted, when the basic Arg492 was mutated to alanine, reporter activation by AqNR1 in the presence of complete serum was abolished (Figure 4B), and FA binding by the purified protein was dramatically reduced (Figure 3C). Replacement of Arg492 by several other amino acids, each of which would remove the hydrogen bond to the FA's carboxyl oxygen, also abolished reporter activation by AqNR1 (Figure 4B). Although rat HNF4α is a weaker activator in this cell line than AqNR1, mutations at this site in HNF4α also reduced activation, consistent with a common structural mode of ligand-binding (Figure 4B). Rat HNF4α, which is thought to be activated by fatty acids produced endogenously in liver cells [24], is not further activated by complete serum, indicating that its specific ligand is different from that of AqNR1 (Figure 4B).
We also mutated several hydrophobic residues in AqNR1 that are predicted to contact the FA's tail; as expected, activation by complete serum was dramatically reduced (Figure 4B). One bulky mutation in the predicted pocket, I444W, conferred strong activity on AqNR1 in the presence of stripped or full serum, implying that this mutation stabilizes the active conformation without ligand or allows binding of an unknown ligand that is present in the cultured cells or medium. Taken together, these data indicate that AqNR1's transcriptional activity is affected by binding of a hydrophobic ligand, that the ligand may be a FA, and that key aspects of the AqNR1's structure-function relations are largely conserved with those of mammalian HNF4. Identification of the specific natural ligand for AqNR1—like that of the ligand for mammalian HNF4 [25]—requires further research, as does determination of whether that ligand has an endogenous or exogenous source.
Purified AqNR2 also bound fatty acids (Figure 3B). It did not, however, activate transcription in the mammalian reporter assay but acted as a very strong repressor of basal transcription, irrespective of the type of serum used (Figure 3A). These results indicate that AqNR2 can repress transcription and, like its ortholog HNF4 and its paralog AqNR1, bind FAs. We cannot rule out the possibility that AqNR2 may have the capacity to activate transcription in the presence of some yet unknown ligand.
A robust rooted protein family phylogeny, functional data on basally branching receptors, and recently gathered information on the functions and structures of NRs from model and non-model organisms allow us to infer the characteristics of the ancestral NR. Although the sequences of the NR superfamily are too divergent to allow unambiguous reconstruction of the ancestral NR LBD at the amino acid level, there is substantial phylogenetic signal in the structural and functional features of NR LBDs. We coded these features as discrete phylogenetic characters and reconstructed the best-supported ancestral states using phylogenetic methods (Figure 5A). The ancestral NR (AncNR) is decisively reconstructed as having had the capacity to activate transcription, bind a ligand, and be activated by that ligand. The vast majority of extant NRs, including those in the basal lineages, have these characteristics. The handful of exceptions—ligand-independent activators and pure repressors—are scattered across the tree and are in most cases nested deep within groups of liganded activators, indicating that these states are almost certainly derived. The fact that some ligand-independent receptors bind ligands, which can up- or down-regulate their baseline activity [6]–[8], further supports the reconstruction of the ancestor as having possessed these features. No ligand-independent activators are present in the basally branching NR clades.
When the evolution of these functional characters is traced on the NR phylogeny, the hypothesis of a ligand-binding and ligand-activated AncNR is by far the most parsimonious reconstruction. This scenario explains the characteristics of the entire NR superfamily with only five losses of dependence on ligand. Three of these losses were accompanied by a loss of ligand-binding; in the other two instances (the ERRs and constitutive androstane receptor, CAR), receptors evolved “constitutive” transcriptional activity but retained the ancestral capacity to bind ligands, which regulate that baseline activity (Figure 5A). In contrast, the alternative hypothesis of a ligand-independent AncNR would require both ligand-binding and dependence on the ligand for activation to have been gained 12 independent times, plus a subsequent reacquisition of ligand-independent activity in one lineage (Figure S8). Reconstruction of these characters on the alternate phylogeny that places AqNR1 and AqNR2 as sponge-specific duplicates causes no change in the support for an ancestral liganded-activated receptor vis-à-vis the “constitutive ancestor” hypotheses (Figure S9).
It is also clear that AncNR had the capacity to activate transcription rather than acting as a pure repressor. An ancestral activator requires five losses of activity in the lineages leading to the inactive repressor NRs (Figure 5A), whereas 11 independent gains of transcriptional activity would be required if AncNR were transcriptionally inactive.
A key element of assessing homology is to determine whether shared structures and mechanisms underlie apparently similar features in different lineages. To further test the hypothesis that ligand-binding and activation are homologous functions derived from the ancestral NR—and that ligand-independent activation was repeatedly derived—we analyzed the underlying structural mechanisms for these functions in a phylogenetic framework. We coded as discrete phylogenetic characters the relevant structural features of NR LBDs and phylogenetically reconstructed the best supported ancestral state for each (Figure 5A).
AncNR is decisively reconstructed as having had the shared features of extant ligand-activated receptors that underlie ligand binding and activation. Specifically, there is strong support for the ancestor having possessed (1) the classic NR fold in the active conformation consisting of three layers of helices in highly conserved positions; (2) an open ligand pocket with volume of at least 300 Å3, bordered by helices H3, H4-5, H7, H10, and H12; and (3) a surface for binding coactivator proteins, composed of residues in the ligand-stabilized helices H3, H4-5, and H12, with a conserved coactivator-recognition motif in the latter [5],[26],[27]. These features and a similar location of the ligand within a highly conserved LBD structure are shared by even the most distantly related ligand-activated NRs (Figure 5B); indeed, even the ligand-independent receptors retain some or all of these features.
The identical structural basis for ligand-activation throughout the superfamily provides strong evidence that this function is derived from the common NR ancestor. It is plausible that the ancestral ligand was an FA, because several of the most basal lineages bind FAs. Further, the key hydrogen bond between the FA's carboxyl-group oxygen and the Arg side chain on helix 5 is conserved in several basal lineages, including HNF4s, RXRs, and AqNR1. The ligand that historically activated AncNR could have been a ubiquitous endogenous molecule that served as a receptor cofactor, a hormone-like regulatory compound endogenously produced under specific conditions, or an exogenous nutrient or other substance, such as fatty acids produced by bacteria or other species.
In contrast, the structural elements that appear to confer ligand-independence differ dramatically among the ligand-independent activators (Figure 6). In Nurr1/DHR38, the mollusk estrogen receptor, and the vertebrate ERRs, the pockets are filled with multiple bulky hydrophobic side chains that mimic the presence of ligand [7],[8],[28],[29], but the sites and states involved in the three receptor classes are all different, with a single convergent exception in two of the three receptors (Figure 6A). In Drosophila Ftz-F1, the H6/H7 region adopts an unprecedented loop conformation that turns inward and fills the cavity (Figure 6A). In CAR, residues in helix 12 (H12) form unique hydrogen bonds to H4-5, and a novel helix, absent from other NRs, packs against H12, stabilizing the active conformation (Figure 6B) [30]–[32]. Finally, in the crystal structure of mouse LRH-1, the active conformation is stabilized without ligand due to a unique salt bridge between residues in H7 and H10, which replaces a similar bridge between the ligand and H10 in orthologs of the same protein of humans and other species, and in SF-1, the closest paralogous NR (Figure 6C) [33].
These radical differences in putative underlying mechanisms indicate that ligand-independent activity is a convergent character with independent evolutionary origins rather than a homologous feature inherited from the common NR ancestor. The hypothesis of a ligand-dependent AncNR explains the structure-function relations of the vast majority of present-day receptors as due to descent from an ancestor that possessed those same features. In contrast, the hypothesis of an unliganded AncNR can explain the structure-function relations of only a single NR as due to descent from the ancestral NR; it requires the ancestral basis for ligand-independent activity to have been independently lost and replaced with different underlying mechanisms in all other lineages of ligand-independent receptors and the shared mechanisms for ligand-dependent activation to have been gained independently in the many lineages of liganded receptors.
Our findings indicate that NR LBDs evolved their functional diversity by tinkering with a ligand-dependent transcriptional activator. Ligand-regulated NRs are thermodynamically tuned so that in the appropriate contexts the active conformation is favored in the presence of activating ligand but not its absence. The most common functional shift during NR evolution was modification of ligand specificity due to subtle changes in the shape and surface properties of the ancestral ligand pocket. Both historical and contemporary studies indicate that such shifts in ligand preference can evolve through a relatively small number of mutations that subtly alter the ligand cavity (e.g., [34]–[36]).
In a few lineages, ligand-independent activity evolved by mutations that stabilized the active conformation in the absence of ligand; in two such cases, the cavity remained open, yielding a receptor whose baseline activity can be antagonized or super-activated by ligands [6],[8]. Laboratory and clinical data contain many examples of ligand-independent activity evolving via single point mutations that add sufficient stability to the active conformation in the absence of ligand (Table S2). Historical studies also document the evolution of constitutive activity with a very simple genetic basis [33]. Such transitions tip the thermodynamic balance so that the formerly switchable LBD becomes stuck in the “on” position, irrespective of ligand. In contrast, evolving a ligand-dependent receptor from a ligand-independent ancestor would require mutations that (1) generate a ligand pocket of the appropriate size and shape to accommodate some ligand and (2) destabilize the active conformation just enough to abolish ligand-independent activity but not so much that the capacity is lost to activate transcription when ligand is present. We observed no such transitions on the NR phylogeny, and we are aware of only one NR mutation that accomplishes this end in the laboratory; that example reflects a return to the ancestral amino acid state in a receptor that binds ligand but also possesses ligand-independent activity [37].
Finally, in a few other lineages, inactive repressor NRs evolved by degradation of the activation function without loss of other functions, such as DNA binding, dimerization, or corepressor binding (see [11]). Indeed, most inactive NRs have simply lost the co-activator interaction motif in H12 but retain the classic LBD secondary and tertiary structure, and some even retain an open pocket [12]. Inactive repressor NRs have been shown to evolve from ligand-activated precursors via simple genetic mechanisms that disable ligand or coactivator binding but leave intact other functions of the receptors, such as DNA binding, dimerization, and corepressor binding [11].
Most gene families, like the NRs, have some common conserved core function—some catalytic activity, for example, or the capacity to interact with DNA. Functional diversity within such families is conferred by members' binding to and carrying out that function on different partners. Our observations in the NRs underscore the capacity of evolution to produce dramatic functional diversity by tinkering with a common ancestral template over long periods of time. The varied and subtle nature of these tinkering events is revealed only when densely sampled structural and functional data are analyzed in a phylogenetic context. We predict that, when sufficient data are gathered to allow detailed evolutionary reconstructions, it will become apparent that most protein superfamilies diversified by subtle modification and partial degradation of ancient, deeply homologous functions. Invoking the evolution of wholesale “novelty” will seldom be necessary.
Nuclear receptor protein sequences were obtained by mining the genomes of Amphimedon queenslandica, Trichoplax adhaerens, Nematostella vectensis, Lottia gigantea, Capitella capitata, and Branchiostoma floridae (Table S1). The assembled genomes and developmental expressed sequence tags were screened using tBlastn with LBD and DBD amino acid sequences from each known NR family. Further analysis using PFAM domain analysis (PF00104 and PF00105) [38] and a hidden Markov model-based method (PTHR11865) [39] confirmed the presence of only two NRs in the A. queenslandica genome, which has been sequenced at approximately 9-fold coverage [19]. In some of the other genomes, gene model sequences were modified to resolve gaps in the sequence by performing a local assembly with gene traces or to correct the predicted protein sequence based on alignment with other conserved domain sequences. Complete NR complements from the curated whole-genome databases of H. sapiens, D. melanogaster, C. intestinalis, F. rubripes, and S. purpuratus were also included. Additional nuclear receptors were identified by using the SMART domain-based sequence annotation resource [40] to search the UniPROTKB/TrEMBL database based on the amino acid sequence of the ERR of Marisa cornuarietis. Receptors for which only partial sequence was available (missing >20% of the DBD or LBD) and those entirely lacking a DBD domain (e.g., human DAX1 and SHP) or LBD domain (e.g., D. melanogaster Knirps and Knrl) were excluded from the analysis.
A total of 275 nuclear receptor sequences were aligned. Full-length sequences containing the DBD, highly variable hinge region, and LBD were aligned using Multiple Sequence Alignment by Log-Expectation (MUSCLE) v. 3.6 [41] in order to identify the boundaries of the conserved regions. After removal of the variable (non-alignable) hinge region, sequence blocks corresponding to the DBD and LBD were then aligned separately using MUSCLE. The DBD and LBD alignments were checked manually to remove lineage-specific indels, and the LBD alignment was checked to ensure correct alignment of the conserved AF-2 core sequence (φφ*κφφ motif; φ, hydrophobic; *, any residue; κ, charged). Amino acids C-terminal to this AF-2 core sequence could not be reliably aligned among all nuclear receptors and the LBD alignment was therefore truncated after the AF-2 core sequence. The DBD and LBD alignments were then concatenated in MacClade 4 (Sinauer Associates, Inc., MA, USA) for subsequent phylogenetic analyses. We then used APDB software [42] to characterize the quality of our alignment with reference to the 26 NR LBDs in the alignment for which X-ray crystallographic structures are available; the average iRMSD (the root mean square difference of the intramolecular distances between aligned pairs of alpha-carbons) over the entire LBD alignment was 0.82 angstroms, well under the resolution of the structures themselves, indicating that the alignment has high structural plausibility.
Phylogenetic analyses were performed using maximum likelihood in PhyML v. 2.4.5 [43] and Bayesian analysis using MrBayes v. 3.1 [44]. The Jones-Taylor-Thornton model with a four-category discrete gamma distribution of among-site rate variation (ASRV) and a proportion of invariant sites was used. For ML, all model parameters were optimized by maximum likelihood. Support was evaluated by obtaining the approximate likelihood ratio for each node—the estimated ratio of the likelihood of the best tree with the split to the best tree without the split [45]—as well as the chi-square confidence metric, which approximates 1−p, where p is the probability that an approximate likelihood ratio as great or greater than that observed at a resolved node would occur if the null hypothesis of an unresolved node is true [45]. To identify the next best alternative tree for the basal split between the AqNR1 and AqNR2-containing groups, we used Phyml to optimize the branch lengths and model parameters on each of the two possible rearrangements of the ML tree around this internal branch and then report their likelihoods. To determine the effect of fast-evolving sites on the inference of phylogeny, we used PAML software to identify sites in the top two octiles of the gamma distribution (Table S3) and repeated the analysis with those 113 sites removed. To facilitate adequate sampling of tree space in Bayesian analysis [46], we used a 174-sequence taxon-trimmed MUSCLE-aligned dataset including nuclear receptors from the following taxa representative of the major metazoan lineages: Acropora millepora, Nematostella vectensis, and Tripedalia cystophora (cnidarians); Amphimedon queenslandica and Suberites domuncula (poriferans); Branchiostoma floridae (cephalochordate); Capitella capitata and Lottia gigantea (lophotrochozoans); Ciona intestinalis (urochordate); Drosophila melanogaster (ecdysozoan); Homo sapiens (vertebrate); Saccoglossus kowalevskii (hemichordate); Strongylocentrotus purpuratus (echindoderm); and Trichoplax adhaerens (placozoan). Terminal branches of length ≥0.76 in the PhyML analysis (except for AqNR2) were removed. Four heated chains were run for 8 million generations with temperature 0.3; the cold chain was sampled every 100 generations. Priors were uniform on topologies, uniform (0, 5) on branch lengths, and uniform (0.1, 10) on the alpha shape parameter. The first 6,694,000 generations were discarded as burn-in, because at this point in the chain the standard deviation of posterior probabilities over all splits was <0.01 and the two chains had converged as evaluated using the “compare” option of AWTY software [47]. We also repeated ML analysis on this reduced 174-sequence dataset and found no change in the relationships among NR types (Figure S5).
The phylogeny shows that AqNR1 is the ortholog of the previously identified but misnamed “RXR” gene identified in the sponge Suberites domuncula [48] and is identical to the “HNF4” gene previously reported in A. queenslandica [49]. To determine the minimum number of gene duplications and losses, we used SDI software [50]. The 275-sequence ML phylogeny was reduced by collapsing sets of orthologous NRs within major taxa—Porifera, Placozoa, Cnidaria, Protostomia, Deuterostomia—into single clades. To avoid spurious inference of duplication/loss due to phylogenetic error, nodes with likelihood-ratio support <10 that conflicted with the accepted taxonomic phylogeny (Porifera, (Placozoa, (Cnidaria, (Protostomia, Deuterostomia)))) were treated as unresolved and reordered to be congruent with the taxonomic phylogeny. SDI software was then used to reconcile this gene family tree with the taxonomic tree and identify the root with the lowest possible mapping cost (duplications plus losses). The mapping cost was also calculated for all possible roots of the gene family phylogeny, except for rootings on branches after the Cnidaria/Bilateria divergence, which have higher mapping costs and were judged to be implausible. Reconstructions of ancestral structural and functional states were performed manually using Fitch parsimony.
Demosponge Amphimedon queenslandica were collected from Heron Island Reef, Great Barrier Reef, and total RNA was isolated from larvae using RNeasy Mini kit (Qiagen, Valencia, CA). The coding regions of AqNR1 and AqNR2 were obtained using BD SMART RACE cDNA Amplification kit (Clontech, Mountain View, CA), and the full reading frames were amplified by RT-PCR, cloned into pGEM-T EASY vector (Promega, Madison, WI), and verified by sequencing. In situ hybridization analysis of RNA expression was conducted as previously described [49].
AqNR1, AqNR2, and rat HNF4α receptor LBDs, including the hinge region and carboxy-terminal extension, were amplified by high-fidelity PCR using Phusion DNA polymerase (New England Biolabs, Ipswich, MA) and cloned into a GAL4-DBD-pSG5 expression vector (gift of D. Furlow). AqNR1-LBD (gi ACA04755) consisted of amino acids 263636, and AqNR2-LBD (GU811658) included amino acids 118–852. Rat HNF-4α template was a gift from Frances Sladek; the LBD used consisted of amino acids 116–465 (NP_071516). Site-directed mutagenesis was performed using QuickChange II (Stratagene, La Jolla, CA) and verified by sequencing.
Chinese hamster ovary (CHO-K1) cells were grown in a 96-well plate and transfected with1 ng of receptor plasmid, 100 ng of a UAS-driven firefly luciferase reporter (pFRluc), and 0.1 ng of the constitutive phRLtk Renilla luciferase reporter plasmid, using Lipofectamine and Plus Reagent in OPTIMEM (Invitrogen, Carlsbad, CA). After 4 h, transfection medium was replaced with phenol-red-free αMEM supplemented with 10% dextran-charcoal-stripped fetal bovine serum (Hyclone, Logan, UT). Cells were allowed to recover and express protein for 48 h, and then assayed by luminometry using the Dual-Glo luciferase system (Promega, Madison, WI). Firefly luciferase activity was normalized by Renilla luciferase activity.
AqNR1 LBD (residues 415–636), AqNR1 mutant proteins, AqNR2 LBD (residues 616–852), and rHNF4α (residues 133–382) were expressed as N-terminal hexahistidine maltose binding protein fusions with a TEV cleavable linker in pLIC-MBP (a gift from J. Sondek) and grown in E. coli BL21(DE3) pLysS cells using standard methods. Protein was purified using affinity chromatography using standard methods. Following TEV cleavage, the resulting 6xHis-tagged MBP was removed using an additional nickel affinity column and the AqNR1 or AqNR2 was polished via gel filtration. Pure AqNR1 or AqNR2 was dialyzed against 150 mM ammonium acetate (pH 7.4) prior to lipid extraction.
Organic solvent extraction was performed on purified LBDs from bacteria to facilitate detailed characterization of bound ligands in the absence of protein. Before extraction, 0.1 mg of C13 labeled palimitic acid was added as an internal standard. Lipid from approximately 4 mg of wild-type or mutant forms of AqNR1 LBD, AqNR2, or rHNF-4 LBD were extracted with a 2∶1 chloroform/methanol (v/v) solution and then analyzed by negative ion ESI/MS. All extractions were performed in duplicate. Mass spectra were acquired on a LTQ FT Hybrid Mass spectrometer (Thermo Finnigan LTQ-FTMS, Somerset, NJ) equipped with an electrospray source. Typically, 10 µL of the aforementioned lipid solution was diluted into 10 µL water/acetonitrile (2∶1 v/v) and subjected to ESI/MS in the negative ion mode. In addition to the fatty acids shown in Figure 2B, an additional unidentified substance at ∼421 m/z was also bound when AqNR1 was incubated with either complete or stripped serum. All samples were run in triplicate. Data acquisition and analysis were performed using the instrument's xcalibur software.
Purified wild-type or mutant hexahistidine maltose binding protein fused AqNR1 was incubated with undiluted complete (Invitrogen - 26010) or cyclodextran/charcoal stripped (HyClone -SH30068.03, Waltham, MA) serum at a ratio of 20 mg protein to 5 ml undiluted serum. The protein/serum mixture was incubated overnight at 4°C followed by re-purification over a nickel affinity column. Protein purity was assessed by SDS-PAGE and fractions containing pure wt or mutant AqNR1 were pooled. Bound lipids were then quantified using the free fatty acid quantification kit from BioVison Inc. (Mountain View, CA). 0.5 mg of each purified LBD was subject to chloroform/detergent extraction to isolate the long chain free fatty acids. Extracted fatty acids were enzymatically converted to their CoA derivatives and oxidized, allowing quantitation in a colorimetric assay (λ = 570) relative to a standard curve generated using palmitic acid.
Efforts to determine the crystal structure of AqNR1-LBD were unsuccessful, so its structure was predicted by homology modeling and energy minimization. The AqNR1 LBD amino acid sequence was aligned to and threaded on human HNF-4α (PDB 1M7W) and then energy minimized with palmitic acid—the most abundant experimentally bound ligand—using the Homology module in InsightII (Accelrys, Inc., San Diego, CA).
To calculate ligand pocket volumes of receptors with X-ray crystal structures, we used VOIDOO [51] in probe-occupied mode. We assigned the centroid of the bound ligand or a manually defined point as a starting locus for cavity searches. Cavity volumes were calculated using 10 random orientations of the protein using 10 different “van der Waals growth factors” ranging from 1.1–1.3. Mean and mode cavity volumes with standard deviation are listed in Table S4.
To calculate ligand pocket volumes of receptors whose structures have not yet been determined by x-ray crystallography, we inferred homology models. Specifically, we created homology models of the LBDs of AqNR1 (gi 167859601, residues 404–534), annelid ER (186908731, residues 231–479), Branchiostoma SR (170178459, residues 298–532), and Branchiostoma ER (170178461, residues 250–504). In each case, we used as templates crystal structures of several NR LBDs with a variety of cavity volumes, including human ERα with estradiol (PDB 1ERE:A, cavity volume 447 Å3), human ERR3 apo form (1KV6:A, cavity volume 262 Å3), human ERR1 apo form (3D24:A, cavity volume 42 Å3), and AqNR1 as modeled on template HNF4A with DAO (1MV7:A, cavity volume 680 Å3). We generated 10 models for every protein with Modeller 9.7 using the default parameters [52]. Models were visually inspected for artifacts (e.g., knotting) and further assessed using RamPage in CCP4i software; only models with 95% of residues in the preferred region and <1% of residues in the outlier region of the Ramachandran map were accepted. We then used Voidoo software as described above to calculate cavity volumes, which are listed in Table S5.
|
10.1371/journal.pcbi.1003526 | Neuronal Spike Timing Adaptation Described with a Fractional Leaky Integrate-and-Fire Model | The voltage trace of neuronal activities can follow multiple timescale dynamics that arise from correlated membrane conductances. Such processes can result in power-law behavior in which the membrane voltage cannot be characterized with a single time constant. The emergent effect of these membrane correlations is a non-Markovian process that can be modeled with a fractional derivative. A fractional derivative is a non-local process in which the value of the variable is determined by integrating a temporal weighted voltage trace, also called the memory trace. Here we developed and analyzed a fractional leaky integrate-and-fire model in which the exponent of the fractional derivative can vary from 0 to 1, with 1 representing the normal derivative. As the exponent of the fractional derivative decreases, the weights of the voltage trace increase. Thus, the value of the voltage is increasingly correlated with the trajectory of the voltage in the past. By varying only the fractional exponent, our model can reproduce upward and downward spike adaptations found experimentally in neocortical pyramidal cells and tectal neurons in vitro. The model also produces spikes with longer first-spike latency and high inter-spike variability with power-law distribution. We further analyze spike adaptation and the responses to noisy and oscillatory input. The fractional model generates reliable spike patterns in response to noisy input. Overall, the spiking activity of the fractional leaky integrate-and-fire model deviates from the spiking activity of the Markovian model and reflects the temporal accumulated intrinsic membrane dynamics that affect the response of the neuron to external stimulation.
| Spike adaptation is a property of most neurons. When spike time adaptation occurs over multiple time scales, the dynamics can be described by a power-law. We study the computational properties of a leaky integrate-and-fire model with power-law adaptation. Instead of explicitly modeling the adaptation process by the contribution of slowly changing conductances, we use a fractional temporal derivative framework. The exponent of the fractional derivative represents the degree of adaptation of the membrane voltage, where 1 is the normal leaky integrator while values less than 1 produce increasing correlations in the voltage trace. The temporal correlation is interpreted as a memory trace that depends on the value of the fractional derivative. We identify the memory trace in the fractional model as the sum of the instantaneous differentiation weighted by a function that depends on the fractional exponent, and it provides non-local information to the incoming stimulus. The spiking dynamics of the fractional leaky integrate-and-fire model show memory dependence that can result in downward or upward spike adaptation. Our model provides a framework for understanding how long-range membrane voltage correlations affect spiking dynamics and information integration in neurons.
| The leaky integrator properties of a neuron are determined by the membrane resistance and capacitance which define a single time constant for the membrane voltage dynamics [1]–[3]. However, the voltage trace of real neurons can follow multiple timescale dynamics [4]–[6] that arise from the interaction of multiple active membrane conductances [7]–[11]. Such processes can result in power-law behavior in which the membrane voltage cannot be characterized with a single time constant [5], [12]–[15]. Since power-law dynamics can span all the scales of interest of neuronal behavior [16]–[20], it is necessary to develop a framework to study such processes and their effect on the electrical and computational capacities of neurons.
In the classical leaky integrate-and-fire model the temporal evolution of the voltage is local [21], [22]. The value of the voltage at a given time depends only on the value of the voltage in the immediate previous time step. Such a process is called Markovian. However, coupling of active conductances does not allow the value of the voltage to be memoryless [11], [17], [23]–[26]. Instead, long time correlations affect the membrane voltage for hundreds of milliseconds. The emergent effect of these membrane correlations is a non-Markovian process that can be modeled with a fractional derivative [27]–[31]. A fractional derivative represents a non-local process [32]–[34] in which the value of the variable is determined by integrating a temporal weighted voltage trace, also called the memory trace. Although fractional derivatives and integrals are almost as old as traditional calculus [32], [35], [36], they have not been widely used due to limited computer power. In the fractional integrate-and-fire model the exponent of the fractional derivative goes from 0 to 1, with 1 representing the normal derivative. As the exponent of the fractional derivative decreases, the weights of the voltage trace increase. Thus, the value of the voltage is increasingly correlated with the trajectory of the voltage in the past.
We developed and analyzed a fractional leaky integrate-and-fire model. The only parameters of the model are the conductance, capacitance, and the fractional exponent. By varying the fractional exponent our model reproduces the upward and downward spike-frequency adaptations found experimentally in pyramidal neurons [37], [38], tectal neurons [39] and fast-spiking cells of layer IV in the rat barrel cortex [40]. Furthermore, the model replicates not only the adapting firing rate but also the long first-spike latency seen in pyramidal neurons in layer V [38]. The model also produces spikes with longer first-spike latency and high inter-spike variability with power-law distribution, which cannot be reproduced by the classical integrate-and-fire model. We further analyze spike adaptation and the responses to noisy and oscillatory input. Overall, the spiking activity of the fractional integrate-and-fire model deviates from the spiking activity of the Markovian model and reflects the temporal accumulated intrinsic membrane dynamics that affect the response of the neuron to external stimulation.
The objective of this project was to develop a fractional leaky integrate-and-fire model of neuronal activity to study spiking adaptation. For this purpose we developed a fractional differential model of the leaky integrator combined with a regular spiking generation mechanism. Complex multiple timescale neuronal systems can be studied using fractional or power-law dynamics; examples range from ion channel gating properties, to diffusion of intracellular signals in Purkinje and pyramidal cells, synaptic strength and firing rate adaptation [14], [19], [27], [37], [41]–[44].
We define the fractional leaky integrate-and-fire model as(1)along with the fire-and-reset condition(2)where is the membrane potential, and is the order (exponent) of the fractional derivative, with . In the case of , the fractional model is the same as the classical leaky integrate-and-fire model. When the membrane potential reaches a threshold (), a spike is generated and is reset to for a refractory period . The passive membrane time constant is . Parameter values are given in Table 1 (see Methods). For , the fractional derivative of the voltage ( in Eq. 1) can be defined with the Caputo [45] fractional derivative(3)
By numerically integrating the above fractional derivative (Eq. 3) using the L1 scheme [46], we approximate the fractional derivative of order , where ,(4)where , and is the value of time such that . For all simulations, we use the time step ms. By combining the right sides of Eqs. 1 and 4, and solving for at time () that depends on all past values of (from to ), we obtain(5)where we define the Markov term weighted by the gamma function as(6)and the voltage-memory trace as(7)
The voltage-memory trace (Eq. 7) can be further divided into the differentiation of past voltage () (Eq. 8) weighted by a function that depends on (Eq. 9):(8)(9)where is the time counter for past events and the positive integer corresponds to the value of time ( = ) at which the voltage is integrated (Eq. 5).
The voltage-memory trace contains information of all the previous voltage activity of the neuron. Clearly, this is a computationally intensive problem due to the expanding matrix over time. We integrate this equation using our recently developed Fractional Integration Toolbox [47]. Since the fractional integration in Eq. 5 needs at least two inputs, is first integrated using the classical leaky integrate-and-fire model.
In this section we show the spiking properties of the fractional leaky integrate-and-fire model and compare our results to experimental data, mainly from cortical pyramidal neurons.
The voltage memory provides non-local dynamics that affects the spiking activity of the cell. The voltage-memory trace decays over time and is dependent on the value of . For the process is identical to the classical leaky integrator, while for values of the past trajectory activity increasingly contributes to the present value of the voltage. Since the weight is always positive, the sign of the voltage-memory trace depends on . With positive applied current the fractional model generates action potentials (Fig. 7A), with positive until the cell fires a spike and is reset (Fig. 7B). However, when there is a spike and voltage is reset, becomes negative. After the voltage escapes from the refractory period, the voltage-memory trace is positive until the next spike (Fig. 7C). As opposed to the classical leaky integrate-and-fire model, the memory trace accumulates over multiple spiking events, changes its dynamics, and thus it affects the ISI (Fig. 7D).
The weight of the voltage-memory trace is determined by the fractional order . The weight is 0 for and it is always positive for . Fig. 7E shows the results of a simulation with standard parameters for 100 time steps (). The x-axis corresponds to the 1–100 temporal weights at t = N. The y-axis corresponds to the value of each weight . The increase in weight can be interpreted as the influence of the past state on the future state of the voltage. In a classical leaky integrator any past value is forgotten as a function of the time constant. In the fractional leaky integrator all the past values could continue to influence the future behavior of the system, particularly, for low values of . Fig. 7F illustrates this point by showing the dynamics of the weight of the initial condition () as a function of . The other important term that comes from the fractional derivative is the fractional coefficient which is a weight factor for the Markov process (Eq. 6). When becomes smaller, this function grows rapidly and it makes the effect of the input current on the voltage dynamics stronger (Fig. 7G). It is the combination of the weighted Markov process and the opposite effect of the memory trace that contribute to the long term spiking dynamics of the fractional model.
Most of our results suggest that the value of has to be low in order to reproduce the spike timing adaptation observed experimentally. Although the fractional model has a continuous dependence on the power-law dynamics cause the effects to be nonlinear. For close to 1 the effects of the Markov term weighted by the gamma function dominate the dynamics. It is only when decreases that the voltage-memory trace can slow down the time evolution of the voltage. This is illustrated by plotting the value of the Markov term versus the memory trace for simulations in which we apply a step current (Fig. 8). When the memory trace is zero and the voltage only moves along the Markov term axis. As the value of decreases the voltage trajectory is deflected, taking longer to depolarize. The power-law dependency results that when the value of in that the memory trace dominates in the initial moments of the depolarization and slows down the dynamics (Fig. 8).
When the input current is constant and , the sub-threshold voltage dynamics (Eq. 1) have an analytic solution which is the same as the solution of the classical integrate-and-fire model [73]. Similarly, for constant input current Langlands et al. [74] derived the analytic solution of the sub-threshold voltage for from the fractional cable equation. In that model the integration of the memory trace is restarted after every spike, thus wiping out the memory trace (see Methods). Such a system is capable of reproducing the delay to first-spike to constant input but then produces regular spiking (Fig. 9A). Our fractional model replicates the same result when we reset the memory trace after every spike (Fig. 9B). However, our model greatly differs from the analytical solution when taking into account the cumulative effect of the memory trace across multiple spiking cycles (Fig. 9C–D). Although both the analytic and full model with memory reset can capture short term memory, they do not produce spike adaptation. Hence, the full fractional model without any memory reset captures the multi-scale processes that spans the spiking activity of neurons.
Spike timing adaptation is a widespread phenomenon throughout the nervous system [37], [39], [75]. In particular, neocortical pyramidal cells produce spike adaptation with multiple timescale dynamics [5], [37], [59]. Our model is capable of reproducing multiple sub-threshold and spike timing adaptation properties reported by different groups and with different experimental conditions. The conclusion from fitting our model to experimental results is that . This indicates that the order of the fractional derivative has to be very low for the memory trace to overcome the classical contribution of the leaky integrator. Furthermore, the fractional model is capable of producing spike trains with high adaption and reliability. Our work provides a framework to study spike adaptation as part of power-law dynamics and the techniques used here can be applied experimentally to determine if a neuron is following power-law adaptation from the sub-threshold to firing rate regimes.
The fractional model can produce different degrees of adapting electrical activities by modifying the fractional exponent . A fractional order derivative captures the long-range correlations of the a system models that results in non-local dynamics. Fractional differential equations have been used in biological systems to capture the long-term memory effects of the dynamics [33], [74], [76]–[81]. For example, fractional order derivatives have been observed in the vestibular-ocular system [27], [82] and in the gating dynamics of ion channels [44], [83]. Power law statistical distributions of a neuronal response also exhibit fractional order dynamics [37]. The voltage dynamics in the fractional order model depend on both the Markov term (immediate past) and the voltage-memory trace that integrates all past voltage values. The behavior of the voltage-memory trace is similar to the behavior of the adaptive filter in the work of Pozzorini et al. [59], although in their work this filter is described as the sum of the spike-triggered current and a moving threshold. The voltage-memory trace also corresponds to the adaptation integral used in other works [18]. In this context our fractional order leaky integrate-and-fire model is a unified mathematical and computational framework that can be used to describe power-law dynamics and long-range correlations in neuronal activity. The fractional derivative can capture relationships between the distribution of conductances that can be complicated to model using explicit techniques.
In biophysical systems, the voltage-memory trace might represent spike-triggered mechanisms that cause adaptation. For example, the voltage-memory trace might represent the slowly inactivating potassium-like current that induces the upward spike adaptation shown in Layer V pyramidal neurons of primary motor cortex [38], suggesting that these neurons are fractional differentiators. Alternatively, the voltage-memory trace can correspond to other adaptation currents such as calcium-activated after-hyperpolarization currents [5], slow sodium-channel inactivation currents [84], [85], or a combination of several adaptation currents. Many studies have used models with slow adaptation currents and exponential functions to analyze spike time adaptation [5], [54], [84]. Some of these models can produce similar properties found in fractional dynamics. For example, the Hodgkin-Huxley model with slow after hyper-polarization currents can produce multiple timescale adaptation processes [37]. The Generalized leaky integrate-and-fire model with an adaptive filter (GLIF-) produces spike adaptation with power-law dynamics [59]. Both the power-law dynamics and history-dependent properties of the GLIF- model correspond to that of the fractional leaky integrate-and-fire model. However, the fractional model provides a general way by simplifying the complicated details shown in other models. The fractional model exhibits spike adaptation with power-law dynamics by integrating all the past voltage values without any additional adaptation currents. Power law functions generalize the mechanism underlying exponential processes and are better alternatives to describe scale invariant spike adaptation [17], [18], [42], [59], [86], [87]. The fractional model shows new directions for studying spike adaptation using fractional derivatives and power-law dynamics instead of classical derivatives and exponential functions.
The key parameter in our fractional model is . Experimental results have suggested that can be as small as 0.15 for neocortical pyramidal neurons [37]. Our fitting of these data also resulted in a value of . Fits to the response of Layer 5 pyramidal motor cortex neurons resulted in a value of [38]. Thus, the experimental results and our modeling analysis suggest that the biophysically important values of are when . However, it is clear that if is much closer to 0, the system takes a longer time to generate spikes or never fires spikes at all, depending on the magnitude of the stimulus, so the feasible range of the fractional exponent might be between 0.05 and 0.2. The value of might correspond to the type, function, or location of specific neurons [38]. These regional differences are not exclusive to the cortex, for example, Purkinje cells in Lobule X and Lobules III – V show different degrees of spike adaptation [75]. Thus, different values of can be used to map the general voltage and spike time adaptation properties of neurons throughout the brain.
Previous work has provided an experimental foundation to determine if a spiking neuron is a fractional differentiator [37]. Our framework provides a more general methodology to determine power-law neuronal dynamics from the sub-threshold to the spiking regime. At the sub-threshold level there are several measurements that can indicate that the membrane of the neuron follows a power-law. For example:
In the same experiment a series of protocols can also be applied to determine if the spiking activity follows power-law dynamics. Some of these measurements are straightforward, others require longer recordings. For instance:
Although power-laws are found at multiple scales of biological organization, their function and importance are still debated [88], [89]. In our work we propose that the membrane voltage of neurons can follow a power-law due to the emergent property of the combination of multiple active conductances. The value of the fractional derivative can be mapped to spike time adaptation dynamics taking place in multiple cell types across the brain. Computationally, a low value of results in spiking dynamics that are at the same time highly adaptable and reliable. Thus, neurons following power-law adaptation could have a large operational range while providing the reliability to filter out noisy signals and increase information capacity. The lack of a sub-threshold resonance frequency allows the neuron to filter signals homogeneously over a wide range of frequencies. In such a case, the fractional leaky integrate-and-fire model provides the basis to study the computational capacities and information processing properties of neurons showing high degree of spike time adaptation.
The equations were coded and implemented using our recently developed fractional integration toolbox [47] and the simulation software package MATLAB [90]. The toolbox can be downloaded at www.utsa.edu/SantamariaLab. The parameters for all simulations were fixed and are described in Table 1.
In order to compare to experiments we extracted the data from the referenced material using WebPlotDigitizer (http://arohatgi.info/WebPlotDigitizer/). Then we imported the data points into Matlab. We then ran simulations varying the value of , usually between [0.5, 1.0] at 0.1 steps. We minimized the mean squared error between the data and the simulations, , where is the number of points. We determined the 95% confidence intervals by then varying around this minimum value of and calculated when it changed for an amount larger than 5% in either direction.
In order to get only sub-threshold oscillations we used where Hz is a sigmoidal frequency function of time that varies from 0 Hz to 100 Hz in 10 seconds. The impedance as a function of frequency is defined as , where V is the membrane voltage, I is the input current, R is the resistance and X is the reactance. The absolute value of can be calculated using a fast Fourier transform and the phase as [57].
The simulations to study spike time reliability were generated by injecting a current nA, where is a Gaussian white noise with zero mean and standard deviation = 0.03 nA. The stochastic input is filtered with an alpha function with time constant ms. The spike trains were obtained from trials, and the trail-to-trail variability of those N different responses were caused by the noise while and were fixed [68], [71]. In order to avoid initial condition effects we analyzed the spike trains of the last 5 s from 10 s simulations.
The reliability measurements were computing using a correlation-based measure [68]–[71]. In brief, the spike trains obtained from N trails were smoothed with a Gaussian filter of width 3, and then pairwise correlated. The correlation-based measure reliability is defined as(10)where is the number of trials and the vectors are the filtered spike trains, filtered using ms. The values of range from 0 (lowest reliability) to 1 (highest reliability), and the reliability was computed for in the range [0.1, 1.0].
For the quantification of the reliability in the coding of an embedded signal we injected the following current nA, where nA is the mean current which generates very low firing rate, nA is the standard deviation of the intrinsic noise, and is the standard deviation of the embedded signal also generated by a Gaussian noise and varies from 0 to 6 nA. The intrinsic and embedded signals are filtered with an alpha function where ms.
We compared our fractional model to a previously developed analytical model of a fractional leaky integrate-and-fire [74]. Briefly, this analytical model is obtained with the following steps. Equation 1 can be converted to(11)where is the membrane time constant. By applying on both sides we obtained(12)Equation 12 is solved using the Fourier-Laplace transform (for details see [36], [74]) and the solution is given by(13)where is the Mittag-Leffler function [36], and for small times this function is approximated as(14)
For the simulation of this model presented in our work we used the full Mittag-Leffler function instead of this approximation. In Eq. 13 the term with the Mittag-Leffler function (right side and right term) represents the memory trace. As in the classical integrate and fire model when the voltage reaches a spike is generated and V is reset to for a refractory period . The subthreshold voltage is integrated using Eq. 13 with initial voltage = and new initial time . During each integration cycle the Mittag-Leffler function restarts from 0 since the initial time reset to a new value. We call this memory reset. Thus, this model wipes out the memory trace after every spike, in contrast to our model that integrates the entire voltage trace. As a result, the inter-spike intervals of the spike train of the analytic solution are equal. If the approximation (Eq. 14) is used to simulate the voltage, the firing rate can also be approximated analytically by combining Eq. 13 and Eq. 14 (see also [74]). Let be the time when the voltage takes to increase from to and to fire. The time is given by [74](15)Using the above the firing rate is approximated by(16)
In the Results section we compare this analytical model with our model with memory reset (re-starting the memory trace after every spike) and with the full model (integrating the memory trace from the beginning of the simulation).
|
10.1371/journal.pntd.0001217 | Contrasting Population Structures of Two Vectors of African Trypanosomoses in Burkina Faso: Consequences for Control | African animal trypanosomosis is a major obstacle to the development of more efficient and sustainable livestock production systems in West Africa. Riverine tsetse species such as Glossina palpalis gambiensis Vanderplank and Glossina tachinoides Westwood are the major vectors. A wide variety of control tactics is available to manage these vectors, but their removal will in most cases only be sustainable if the control effort is targeting an entire tsetse population within a circumscribed area.
In the present study, genetic variation at microsatellite DNA loci was used to examine the population structure of G. p. gambiensis and G. tachinoides inhabiting four adjacent river basins in Burkina Faso, i.e. the Mouhoun, the Comoé, the Niger and the Sissili River Basins. Isolation by distance was significant for both species across river basins, and dispersal of G. tachinoides was ∼3 times higher than that of G. p. gambiensis. Thus, the data presented indicate that no strong barriers to gene flow exists between riverine tsetse populations in adjacent river basins, especially so for G. tachinoides.
Therefore, potential re-invasion of flies from adjacent river basins will have to be prevented by establishing buffer zones between the Mouhoun and the other river basin(s), in the framework of the PATTEC (Pan African Tsetse and Trypanosomosis Eradication Campaign) eradication project that is presently targeting the northern part of the Mouhoun River Basin. We argue that these genetic analyses should always be part of the baseline data collection before any tsetse control project is initiated.
| Tsetse flies are insects that transmit trypanosomes to humans (sleeping sickness) and animals (nagana). Controlling these vectors is a very efficient way to control these diseases. In Burkina Faso, a tsetse eradication campaign is presently targeting the northern part of the Mouhoun River Basin. To attain this objective, the approach has to be area-wide, i.e. the control effort targets an entire pest population within a circumscribed area. To assess the level of this isolation, we studied the genetic structure of Glossina palpalis gambiensis and Glossina tachinoides populations in the target area and in the adjacent river basins of the Comoé, the Niger and the Sissili River Basins. Our results suggest an absence of strong genetic isolation of the target populations. We therefore recommend establishing permanent buffer zones between the Mouhoun and the other river basin(s) to prevent reinvasion. This kind of study may be extended to other areas on other tsetse species.
| Tsetse flies (Diptera: Glossinidae) are the sole cyclical vectors of human and animal trypanosomoses, two major plagues that are seriously impeding African development. African animal trypanosomosis (AAT) is a major obstacle to the development of more efficient and sustainable livestock production systems in West Africa. Since 2008, the Government of Burkina Faso has embarked on an ambitious tsetse eradication campaign that targets the northern Mouhoun River Basin for its first phase (http://www.pattec.bf/). The Mouhoun River Basin eradication campaign is implemented under the auspices of the Pan African Tsetse and Trypanosomosis Eradication Campaign (PATTEC), an African Union initiative that was launched in 2001 following an historic decision by the African Heads of State and Government in Lome, Togo, July 2000 (http://www.africa-union.org/Structure_of_the_Commission/depPattec.htm).
In the Mouhoun River Basin, Glossina palpalis gambiensis Vanderplank and Glossina tachinoides Westwood are the two remaining tsetse species, after the regression of Glossina morsitans submorsitans Newstead [1]–[3]. The two tsetse species remain very effective vectors of AAT [4], but local transmission of sleeping sickness (Human African Trypanosomosis (HAT)) seems to have disappeared from the Mouhoun River Basin [3]. These species inhabit the riparian forests that form habitat galleries along the rivers and the flies' relative abundance is determined by forest ecotype and its level of fragmentation and destruction [2], [5]. Their particular resilience to habitat fragmentation has been attributed to (1) their ability to easily adapt to peridomestic situations, (2) their opportunistic host feeding behaviour [6], and (3) their linear habitat that allows them to easily disperse between favourable patches, i.e. riverine forests acting as “genetic corridors” [7], [8].
Control of tsetse can be achieved through a variety of techniques [9], including traps, insecticide-impregnated targets [10], live-baits [11]–[13], sequential aerosol technique [14], and the sterile insect technique (SIT) [15]. In the past, most control efforts were not sustainable due to either flies surviving the initial interventions, or flies immigrating from untreated regions, or both [16]. The strategic choice between eradication and suppression of a tsetse population is of prime importance as it will have significant economic implications (see [17] for a review). In that respect, knowledge of the genetic structure of the target population can facilitate this critical decision making [18]–[20]. For isolated tsetse populations, eradication is undoubtedly the most cost-effective strategy, as was demonstrated with the sustainable removal of Glossina austeni Newstead from the Island of Unguja, Zanzibar in 1994–1997 [15]. On mainland Africa, the geographical distribution limits of the target tsetse populations are less clearly defined, although complete isolation was recently demonstrated for a G. p. gambiensis population in the Niayes area of Senegal that prompted the Government of Senegal to select an eradication strategy [20], [21].
In Burkina Faso, G. p. gambiensis populations inhabiting fragmented habitats are genetically structured along the rivers [22], also in the area that is the target of the national eradication campaign mentioned above. However, a certain level of gene exchange is still sustained among the various populations that inhabit the habitat fragments along the Mouhoun River. Furthermore, G. tachinoides occurs as a panmictic population along its riverine habitat in the same area, due to its more xerophylous nature allowing it to disperse more easily between suitable habitat patches [23]–[25]. As riverine tsetse populations are mainly confined to the riverbeds of the various river systems which are organised in river basins it was proposed to use the “river basin” as a unit of operation in area-wide integrated pest management (AW-IPM) programmes [28] against tsetse in West Africa. This assumed that each primary river basin (and possibly also secondary and tertiary) contained riverine tsetse populations that were geographically isolated from those belonging to adjacent river basins. If this hypothesis proves to be correct, it would be very beneficial for the present eradication campaign since it would allow limiting the control effort to the Mouhoun River Basin. However, earlier studies have indicated that riverine tsetse flies were able to disperse up to 2km into the savannah areas bordering the riparian forests [7] and a recent genetic study in Burkina Faso suggested that G. p. gambiensis was able to cross the watershed divide between the Mouhoun and the Comoe river basins that contained natural woody savannah [26]. In view of the importance of the Mouhoun eradication project, and the limited number of samples (three) used in previous study [26], it was deemed necessary to expand these studies and to obtain more data on the dispersal potential of the two tsetse species present, as evidenced through genetic structures of the various populations. A more complete picture of the exchange of genes between the various tsetse populations in the area would enable the programme managers to make informed decisions on the establishment of buffer zones between the Mouhoun River Basin and its neighbouring basins, or, alternatively, to expand the eradication campaign to these basins.
The present study includes G. tachinoides and two other river basins not considered earlier and also includes areas where the interfluve is very much fragmented, which might impact dispersal of riverine species. Genetic variation at microsatellite DNA loci was thus used to examine the structure of G. p. gambiensis and G. tachinoides populations of the Mouhoun River Basin in relation to those of all its adjacent river basins, i.e. the Niger (Bani), Comoé and Sissili River Basins (Figure 1). The objective was to assess tsetse population structuring in and between the different river basins, its relation to tsetse fly dispersal amongst adjacent river basins, and its consequences for potential AW-IPM eradication campaigns [27], [28].
The study area is located in South-Western Burkina Faso (latitude 10.2 to 12.2 N; longitude −5.5 to −2.0°W) and encompassed the Mouhoun River Basin (8 sampling sites) and three neighbouring river basins, i.e. the Comoe (3 sampling sites), the Sissili and the Niger (1 sampling site each) River Basins (fig. 1). From November 2007 to March 2008 each site was sampled using 5–10 unbaited biconical traps [29]. In each location, the maximal river length sampled was 980 m (in Darsalamy) for G. p. gambiensis and 5660 m for G. tachinoides (Fandiora), but was usually lower than 500 m (Tables 1&2).
A total of 296 G. tachinoides and 242 G. p. gambiensis flies were genotyped (see number of flies genotyped by trapping site in Tables 1&2). G. p. gambiensis was genotyped at 8 microsatellite loci: Gpg 55.3 [30], A10, B104, B110, C102 (kindly supplied by A. Robinson, Insect Pest Control Laboratory (formerly Entomology Unit), Food and Agricultural Organization of the United Nations/International Atomic Energy Agency [FAO/IAEA], Agriculture and Biotechnology Laboratories, Seibersdorf, Austria), pGp13, pGp24 [31], and GpCAG [32]. G. tachinoides was genotyped at 9 microsatellite loci: pGp13, pGp17, pGp20, pGp24, pGp28, pGp29 [31], B104, C102 and GpCAG. Of these, B104, B110, pGp13, pGp20, and 55.3 are known to be located on the X chromosome. GpCAG and C102 have trinucleotide repeats whereas the others are dinucleotides.
Three legs of each individual tsetse fly were removed, transferred to a tube to which 200 µl of 5% Chelex chelating resin was added [33], [34]. After incubation at 56°C for one hour, DNA was denatured at 95°C for 30 min. The tubes were then centrifuged at 12,000 g for two min and frozen for later analysis. The PCR reactions were carried out in a thermocycler (MJ Research, Cambridge, UK) 20 µl final volume, using 10 µl of the diluted supernatant from the extraction step as template. After PCR amplification, allele bands were routinely resolved on a 4300 DNA Analysis System from LI-COR (Lincoln, NE) after migration in 96-lane reloadable (3x) 6.5% denaturing polyacrylamide gels. This method allows multiplexing by the use of two infrared dyes (IRDye), separated by 100 nm (700 and 800 nm), and read by a two channel detection system that uses two separate lasers and detectors to eliminate errors due to fluorescence overlap. To determine the different allele sizes, a large panel of about 70 size markers was used. These size markers had been previously generated for G. p. gambiensis by cloning alleles from individual tsetse flies into pGEM-T Easy Vector (Promega Corporation, Madison, WI, USA), but were generated for G. tachinoides for this study. Three clones of each allele were sequenced using the T7 primer and the Big Dye Terminator Cycle Sequencing Ready Reaction Kit (PE Applied Biosystems, Foster City, CA, USA). Sequences were analyzed on a PE Applied Biosystems 310 automatic DNA sequencer (PE Applied Biosystems) and the exact size of each cloned allele was determined. PCR products from these cloned alleles were run in the same acrylamide gel as the samples, allowing the allele size of the samples to be determined accurately [35]. The gels were read twice by two independent readers using the LIC-OR Saga genotyping software.
All datasets were processed with Create V 1.1 [36] and converted into the appropriate format as needed.
Wright's F-statistics [37] were estimated with Weir and Cockerham's unbiased estimators [38] under Fstat V 2.9.4 (Goudet 2003, updated from [39]). FIS is a measure of local inbreeding of individuals relative to inbreeding of subsamples. It is therefore also a measure of reproductive strategy and varies from -1 (all individuals are heterozygous for the same two alleles within each subsample) to +1 (all individuals are homozygous with at least two alleles in subsamples) and equals 0 when all subsamples conform to genotypic proportions expected under panmixia. It is thus also a measure of deviation from the random mating model within populations. FST measures inbreeding of subsamples relative to the total inbreeding resulting from subdivision. It is therefore also a measure of differentiation among subsamples. It varies between 0 (no differentiation) and 1 (all subsamples fixed for one or the other allele).
The significant departure from 0 of these parameter estimates was tested by randomisation procedures under Fstat. For this, alleles are randomly exchanged between individuals in each subsample and the proportion of times when a FIS estimate was equal to or higher than the observed one provided the exact P-value of the test. For differentiation between populations, individual were randomised across subsamples and the statistic used here was the log-likelihood ratio G as recommended [40].
Linkage disequilibrium (LD) between loci was also tested through randomising association between each locus pair. For each pair of loci the tests were combined across subsamples with the G-based procedure as recommended [41]. All these randomisations (10000 in each case) were undertaken with Fstat 2.9.4.
For LD, there were as many tests as there were loci pairs (here possibly 36), we therefore tested the probability of obtaining a proportion higher than the expected one (5%) with a binomial test with k tests, mean 0.05 and ks success (the number of significant pair in linkage disequilibrium at level α = 0.05) with MultiTest V 1.2 [41].
More than three levels (i.e. individuals, sub-populations and total) exist within the samples of each tsetse species. Individuals were caught in different traps, in different sites (i.e. locations) within three different river basins (Comoé, Mouhoun and Sissili for G. tachinoides and Comoé, Mouhoun and Niger for G. p. gambiensis). Hierfstat version 0.03–2 [42] is a package for the statistical software R. This package computes hierarchical F-statistics from any number of hierarchical levels [42]. FTrap/Site represents the homozygosity due to the subdivision into different traps in each site and was tested by randomising individuals between traps within each site. FSite/Basin represents the homozygosity due to subdivision into different sites within each river basin and was tested by randomizing traps (with all individuals contained) between sites within the same river basin. FBasin/Total measures the relative homozygosity due to the geographical separation between river basins and was tested by randomizing sites (with all traps included) between the three river basins. In all cases we undertook 1000 permutations and the log likelihood ratio as for the FST analysis was the statistic used. These tests were performed with Hierfstat. A user friendly step by step tutorial of how to use HierFstat is available [43].
Some microsatellite loci, noted with an X as last letter, are X linked. These loci were coded as missing data for FIS and null allele analyses and coded as homozygous for the allele present on the X for differentiation and LD tests.
Significant FIS can be due to null alleles, stuttering or short allele dominance. We used MicroChecker V 2.2.3 [44] for stuttering and null alleles. We tested how null alleles can explain the observed FIS using estimates of null allele frequency following either Brookfield's second method [45] or to the method of van Oosterhout et al. [44] as given by MicroCheker. We used these estimates to compute expected blank (non amplified null homozygotes) frequency assuming panmixia. For each locus, the sum of all expected blanks across subsamples was compared to the sum of all observed ones with an exact unilateral binomial test with the alternative hypothesis: there were not enough observed blank genotypes as compared to what would be expected under the hypothesis of null alleles in a panmictic population. For X linked loci we also used null allele frequencies (estimated from females) directly as the expected proportion of blank (unamplified) males expected at these loci and this quantity was also compared with observed blanks with the same method as described above for females at other loci.
Confidence intervals (CI) were obtained using the standard error of estimates obtained by jackknife over subsamples or by bootstrap over loci, using Fstat, as described in [46].
Sex-biased dispersal was assessed using three tests implemented in Fstat. First, Weir and Cockerham's estimate of FST, was calculated separately in each sex. Next, tests based on the mean (mAIc) and the variance (vAIc) of Favre et al.'s corrected assignment index AIc [47] were performed (see Prugnolle and De Meeûs [48] for more details on these tests). All three tests are based on a permutation procedure; the sex of each individual is randomly re-assigned in each population (10,000 permutations). The observed difference between male and female FST, the ratio of the largest to the smallest vAIc and the AIc-based t-statistics defined by Goudet [49] were then compared to the resulting chance distributions. For the sex that has a higher dispersal rate, FST and mAIc are expected to be smaller and vAIc is expected to be higher than for the sex that has a lower dispersal rate. This choice of statistics is motivated by the work of Goudet et al. [49] where vAIc was shown to be the most powerful statistic when migration is low (less than 10%), while FST performs better in other circumstances. We also chose to keep mAIc because it may be more powerful in case of complex patterns of sex specific genetic structures [50], [51]. Tests were all bilateral.
Isolation by distance was inferred with Rousset's procedure [52] through the regression FST/(1-FST)∼a+bLn(DG). FST/(1-FST) is a modified measure of differentiation between two subpopulations, a is a constant, Ln(DG) is the natural logarithm of the geographical distance between subpopulation pairs for two dimensional data and b the slope of the regression that is related to the product Deσ2 of reproducing (effective) adults local density (De) by the dispersal surface σ2 (σ is the mean distance between reproducing adults and their parents) by the equation Deσ2 = 1/4πb because the neighbourhood size Nb = 1/b = 4πDeσ2 [52]. In that case, the effective number of immigrants per neighbourhood can be computed as Nem = 1/2πb [52]. For one dimensional data, the model becomes FST/(1-FST)∼a+bDG and Deσ2 = 1/4b [52]. The significance of the signal was tested with a Mantel test [53] and bootstrap over loci gave 95% confidence intervals for the slope. All isolation by distance procedures were implemented using Genepop 4 [54] with 1,000,000 iterations. For the sake of power, traps were used as sub-population units for isolation by distance procedures.
Effective population sizes were estimated following Waples and Do's method based on linkage disequilibrium and implemented in LDNe [55], linkage disequilibrium and heterozygosity as implemented by Estim 1.2 [56] and following Balloux's method based on heterozygote excess in dioecious populations [57] assuming even sex ratio.
For G. tachinoides, since no sub-structuring was observed at the site level, areas of sites were assimilated to the rectangle defined by the approximate gallery forest width (∼100 m) and the mean maximal distance between the two most distant traps in a site (∼1000 m), being aware that it is a conservative value. This surface S = 100,000 m2 was thus used to divide effective population sizes to compute densities. For G. p. gambiensis densities were computed by dividing the population size by the mean minimum distance between two traps (∼100 m) in one dimension along rivers, or by the surface of the rectangle defined by this distance and the approximate gallery forest width (∼100 m), hence S = 10,000 m2, for two dimensions. This distance of 100 m also corresponds to the range of attraction of a biconical trap, and thus the smallest river section that can be sampled irrespective of the sampling protocol used [58].
HierFstat analysis only found one significant hierarchical level of population structure in the G. tachinoides samples, i.e. subdivision by sites FSite/Basin = 0.026 (P-value = 0.001). Traps (P-value = 0.179) and river basin (P-value = 0.707) did not significantly contribute to the genetic structure of G. tachinoides. To check for possible disturbing effect of substructuring within sites that may not be detected by HierFstat, we also tested isolation by distance between traps in each of the four sites with the model FST/(1-FST)∼a+bDG, appropriate for one dimensional data (along the river). This analysis was feasible in view of the large amount of data available for the Mouhoun River. Absence of population sub-structuring was confirmed by the total absence of any isolation by distance between traps within the Mouhoun River (all slopes ≤0, all P-values>0.49). In further analyses we only considered sites as subpopulation units for G. tachinoides, except for isolation by distance as explained above.
For G. p. gambiensis, two hierarchical levels appeared to contribute significantly to genetic structure, the trap in each site (FTrap/Site = 0.0117, P-value = 0.033) and the site in each river basin (FSite/Basin = 0.0379, P-value = 0.001). The analysis therefore revealed that river basins were not important for the genetic structuring of the G. p. gambiensis populations (P-value>0.6). For all further analyses with G. p. gambiensis, the trap was considered as the subpopulation unit and, for population structure analyses (sex biased dispersal, isolation by distance), each site was considered separately, except when specified otherwise.
For G. tachinoides, LD tests were carried out with all the 9 loci (36 pairs tested) and with the six most polymorphic loci, i.e. loci with no allele at frequency above or equal to 0.9 (pGp28 and pGp29 excluded, hence 21 pairs remaining). In the first case three pairs appeared in significant linkage and two pairs in the second case, which is not significantly above the 5% level in each case (binomial P-values are respectively 0.27 and 0.28). For G. p. gambiensis only one test was significant at the 5% level, which is not significantly above the proportion expected under the null hypothesis (P-value = 0.7628).
There was a strong and highly significant heterozygote deficit (FIS = 0.227, 95% CI = [0.067, 0.429] in G. tachinoides due to loci pGp17, pGp20X, pGp24, pGp28 and B104X (Figure 2). The four remaining loci, pGp13X, pGp29, C102 and GPCAG, together provided a pattern conforming with genotypic proportions expected under random mating: FIS = −0.005, P-value = 0.5661. For the other loci, stuttering was observed for pGp17 in all the eight subsamples, and in one subsample for pGp20X. Moreover, null alleles can reasonably explain all FIS as can be seen from Table 3. Consequently, it was assumed with confidence that stuttering and null alleles totally explained the heterozygote deficits observed at these five loci and we can confidently conclude that the G. tachinoides subsamples conformed to the random mating hypothesis.
For G. p. gambiensis the FIS is slightly lower (FIS = 0.137, 95% CI = [0.071, 0.219]) but still highly significant (P-value = 0.0001) (Figure 3). According to MicroChecker analyses, null alleles provided a reasonable explanation (Table 4). Nevertheless, individually non significant loci alone still provided a significant positive FIS = 0.042 (P-value = 0.0356). Thus neither null alleles nor Wahlund effects alone can explain the pattern observed in this species, as it is often the case for G. p. gambiensis [18], [22], [26].
As can be seen from Table 5, there is a significant genetic signature of sex biased dispersal in G. tachinoides, with the female flies having a lower dispersal rate (male biased dispersal).
For G. p. gambiensis several sex biased dispersal tests were carried out:between sites over all river basins and between sites within the Mouhoun river basins, between traps within the Mouhoun river basin and between traps within sites. For the first and second tests, only one male and one female per trap were used, or only a single individual if only one sex was available, per trap and individuals of the same site considered as belonging to the same subpopulation. This data reduction was done to limit as much as possible the confounding effect of the significant differentiation that exists between traps in this species (see [51] for comments on that matter). A single test resulted in a significant P-value (Table 6), with the mAIc indicating a female biased dispersal. However, it can be seen from Table 6 that biased dispersal genetic signatures are inconsistent across parameters in the same analysis or across analyses for the same parameter. As previously observed [26], the most obvious conclusion, is that no genetic signature of sex biased dispersal could be detected in G. p. gambiensis at any level.
There was a highly significant isolation by distance across traps over the total G. tachinoides sampling zone (P-value = 0.0001) with a slope b = 0.015. This results in a neighbourhood size Nb≈67 individuals. Estim did not provide a usable effective population size. Effective population sizes were relatively convergent across Waples and Do's and Balloux's methods. With Waples and Do's method, three sites (two in Comoe and one in the Mouhoun Basin) provided outputs different from infinity, with mean Ne = 99.4. Balloux's method gave Ne = 100. We then assumed an effective subpopulation size of ∼100. A mean sampling surface as defined above as S∼0.1 km2, resulted in an effective population density of De = Ne/S≈1000 flies per km2. Rousset's model [52] indicated a mean dispersal per generation of around 73 m for this species, or a migration rate between neighbouring sites of m = 1/2πb = 0.11.
For G. p. gambiensis, there was no evidence for isolation by distance in any site along rivers. But this may be due to the very short length of river portions covered in each site. As some sites were however very distant, we further used isolation by distance in a two dimensional framework. Over the entire sampling zone, a significant isolation by distance was detected (P-value = 0.022) with slope b = 0.015 and a resulting neighbourhood size Deσ2≈67 individuals identical to G. tachinoides. Estim provided an estimate of Ne = 81 and m = 0.286 in one trap of the Mouhoun Basin. LDNe provided only usable values for Ne in four traps of the Mouhoun Basin, with mean Ne = 149. The surface defined above S∼0.01 km2 leads to an effective density of G. p. gambiensis De = Ne/S≈8000 (for Ne = 80) or De = 15000 (for Ne = 150) G. p. gambiensis per km2 in the study area. Mean dispersal per generation is thus σ = 26 m or σ = 19 m for Ne = 80 and Ne = 150 respectively, corresponding to migration rates of 0.13 and 0.07 respectively (with Rousset's 1997 model in two dimensions) between neighbouring subpopulations (traps).
Using the island model of migration with even sex ratio, published by Vitalis [59], and in particular using equation 10 from his paper, we checked which parameters could lead to the sex biased dispersal observed in G. tachinoides and the observed difference in FST between female and male flies. As can be seen in Table S1, the best fit of the model parameters would indicate a very low female migration rate (less than 0.01 and most probably around 0.0001), a moderate male migration rate around 0.12 (between 0.1 and 0.15) and subpopulation sizes around 100 individuals (between 80 and 120 individuals). The number of subpopulations and the mutation rate had a small influence on the results. Thus, even if some care must be taken with these values coming from an island model of migration, parameters seem quite convergent with what was inferred from G. tachinoides isolation by distance population structure.
The population genetics data presented here suggest that the savannah area of the watershed divide between two adjacent river basins does not seem to represent a significant barrier to gene flow for the two riverine tsetse species studied. The results corroborate data from an earlier preliminary study that assessed gene flow (but without clear quantification) between three populations of G. p. gambiensis inhabiting two tributaries of the Mouhoun and Comoé river basins in Burkina Faso [26]. For both species, isolation by distance between sites of different river basins (or even at a micro-scale for G. p.gambiensis) was evidenced, without a particular role of river basins. Nevertheless, for G. palpalis gambiensis, dispersal along rivers (in one dimension) is still more efficient than across them (i.e. in two dimensions). During the rainy season, riverine tsetse fly species disperse in the savannah areas neighbouring the river [7], probably in search of suitable hosts, like cattle, that during that time of the year do not have to enter the riparian forests to find drinking water. It is conceivable that after some days without rain, remaining flies in the savannah areas are quickly forced to find resting sites before facing desiccation and are therefore stimulated to disperse at a higher rate. Following environmental cues such as humidity or temperature gradients, these flies will need to venture back to the closest gallery forest, that might well belong to another river basin system. Tsetse dispersal processes are complex and simple random diffusion models have often been used to capture this complexity [60]. This approach seems to be inadequate as was recently confirmed by an analysis of dispersal data of sterile male Glossina austeni Newstead that were released homogeneously from the air. The recapture data indicated that the sterile flies congregated in the same sites that were also preferred by their wild counterparts [61]. In addition, when riverine tsetse find themselves in unsuitable sites, they are capable of dispersing up to 2km per day to reach suitable habitats (Bouyer J., unpublished data).
The analysis presented here showed that dispersal of G. tachinoides across river basins was ∼3 times higher than G. p. gambiensis, which suggests that G. tachinoides flies have the ability to disperse with ease despite the severe fragmentation of the riparian gallery forests in the study area [2]. G. p. gambiensis dispersed less along fragmented riparian forest habitat and seemed to encounter more difficulties to disperse between the remaining fragments of this suitable habitat. The fact that genetic structuring is not correlated to geographic distance at a local scale in G. tachinoides [25], and the higher level of genetic structuring observed for G. p. gambiensis populations at the micro-scale [22] corroborate these observations. G. tachinoides is more xerotolerant (i.e. tolerant for dry conditions) than G. p. gambiensis, which could lead to a different perception of habitat borders in this species [24]. Mark-release-recapture studies carried out more than 20 years ago [7] showed that, in homogeneous, unfragmented gallery forests, the two species had a similar rate of dispersal. However, capture-mark-release-recapture data do not necessarily correlate with genetic data, as was observed in morsitans group flies [62], since the former is a direct measure of all kinds of dispersal including hunting dispersal, whereas the latter is an indirect measure of only reproductive dispersal. Our data imply that habitat fragmentation seems to reduce the dispersal capacity of G. p. gambiensis much more as compared to that of G. tachinoides. Similar conclusions were drawn from recent mark-release-recapture experiments in Burkina Faso, where mean dispersal coefficients of 0.3 km2.d−1 and 0.05 km2.d−1 were observed corresponding to mean square displacements of 775 m/day and 316 m/day for male G. tachinoides (Bouyer, J., unpublished data) and G. p. gambiensis [22] respectively. The much lower effective density observed for G. tachinoides as compared to G. p. gambiensis is partially related to the location of the sampling sites, which were mostly along small tributaries of the Mouhoun. These are known to be preferred sites for G. p. gambiensis – hence the name “spring” tsetse fly [5] – but are not favoured by G. tachinoides. During the entire sampling process, the mean number of flies caught per trap per day were 1.04 (s.d. 1.06) and 0.13 (s.d. 1.31) for G. p. gambiensis and G. tachinoides, respectively.
Tsetse flies are polygynous where the reproductive investment of female flies far outreaches that of the male flies. As such and according to the three main asymmetries of dispersal/philopatry costs between genders favouring biased dispersal (i.e. the resource-competition hypothesis, the local mate competition hypothesis and the inbreeding hypothesis) a sex biased dispersal in tsetse flies (should it exist) would be biased towards greater mobility of the male sex (see [47] and references therein). Our analysis of the sex biased dispersal in G. tachinoides suggests that female flies indeed disperse very little in fragmented riparian vegetation. This seems to suggest that female G. tachinoides are very conservative in their dispersal behaviour and not only remain close to “known” suitable larviposition sites in these fragmented landscapes, but are also highly philopatric i.e. they deposit their larvae close to their own place of birth. This behaviour would reduce the risk of reinvasion, as only founding females would produce offspring for a new population. This result is at variance with classical mark-release recapture experiments where females were dispersing more than males [7]. One possibility to explain our result would be a sex specific local adaptation rendering immigrant females very unlikely to survive locally. Sex based differences in dispersal were not observed for G. p. gambiensis in the 1980's in Burkina Faso and more recently in Guinea and Burkina Faso [18], [26]. In this case, both sexes dispersed very little, which was also reflected in a high level of structuring at a more local scale [22].
In conclusion, the data presented here, combined with those from earlier studies [26], suggest that in Burkina Faso, riverine tsetse populations from adjacent river basins are exchanging genetic material, and can therefore not be considered as biologically isolated. Therefore, potential re-invasion of flies from adjacent river basins will have to be prevented by establishing buffer zones between the Mouhoun and the other river basin(s), in the framework of the PATTEC (Pan African Tsetse and Trypanosomosis Eradication Campaign) eradication project that is presently targeting the northern part of the Mouhoun River Basin. Alternatively, the campaign should be extended to adjacent infested basins to sustain the eradication.
|
10.1371/journal.pcbi.1000117 | Geometric Interpretation of Gene Coexpression Network Analysis | The merging of network theory and microarray data analysis techniques has spawned a new field: gene coexpression network analysis. While network methods are increasingly used in biology, the network vocabulary of computational biologists tends to be far more limited than that of, say, social network theorists. Here we review and propose several potentially useful network concepts. We take advantage of the relationship between network theory and the field of microarray data analysis to clarify the meaning of and the relationship among network concepts in gene coexpression networks. Network theory offers a wealth of intuitive concepts for describing the pairwise relationships among genes, which are depicted in cluster trees and heat maps. Conversely, microarray data analysis techniques (singular value decomposition, tests of differential expression) can also be used to address difficult problems in network theory. We describe conditions when a close relationship exists between network analysis and microarray data analysis techniques, and provide a rough dictionary for translating between the two fields. Using the angular interpretation of correlations, we provide a geometric interpretation of network theoretic concepts and derive unexpected relationships among them. We use the singular value decomposition of module expression data to characterize approximately factorizable gene coexpression networks, i.e., adjacency matrices that factor into node specific contributions. High and low level views of coexpression networks allow us to study the relationships among modules and among module genes, respectively. We characterize coexpression networks where hub genes are significant with respect to a microarray sample trait and show that the network concept of intramodular connectivity can be interpreted as a fuzzy measure of module membership. We illustrate our results using human, mouse, and yeast microarray gene expression data. The unification of coexpression network methods with traditional data mining methods can inform the application and development of systems biologic methods.
| Similar to natural languages, network language is ever evolving. While some network terms (concepts) are widely used in gene coexpression network analysis, others still need to be developed to meet the ever increasing demand for describing the system of gene transcripts. There is a need to provide an intuitive geometric explanation of network concepts and to study their relationships. For example, we show that certain seemingly disparate network concepts turn out to be synonyms in the context of coexpression modules. We show how coexpression network language affects our understanding of biology. For example, there are geometric reasons why highly connected hub genes in important coexpression modules tend to be important, and why hub genes in one module cannot be hubs in another distinct module. We provide a short dictionary for translating between microarray data analysis language and network theory language to facilitate communication between the two fields. We describe several examples that illustrate how the two data analysis fields can inform each other.
| Many biological networks share topological properties. Common global properties include modular organization [1],[2], the presence of highly connected hub nodes, and approximate ‘scale free topology’ [3],[4]. Common local topological properties include the presence of recurring patterns of interconnections (‘network motifs’) in regulation networks [5]–[7].
One goal of this article is to describe existing and novel network concepts (also known as network statistics or indices [8]) that can be used to describe local and global network properties. For example, the clustering coefficient [9] is a network concept, which measures the cohesiveness of the neighborhood of a node. We are particularly interested in network concepts that are defined with regard to a ‘gene significance measure’. Gene significance measures are of great practical importance since they allow one to incorporate external gene information into the network analysis. In functional enrichment analysis, a gene significance measure could indicate pathway membership. In gene knock-out experiments, gene significance could indicate knock-out essentiality. We study gene significance measures since a microarray sample trait (e.g., case control status) gives rise to a statistical measure of gene significance. For example, the Student t-test of differential expression leads to a gene significance measure. Many traditional microarray data analysis methods focus on the relationship between the microarray sample trait and the gene expression data. For example, gene filtering methods aim to find a list of (differentially expressed) genes that are significantly associated with the microarray sample trait; another example are microarray-based prediction methods that aim to accurately predict the sample trait on the basis of the gene expression data.
Gene expression profiles across microarray samples can be highly correlated and it is natural to describe their pairwise relations using network language. Genes with similar expression patterns may form complexes, pathways, or participate in regulatory and signaling circuits [10]–[12]. Gene coexpression networks have been used to describe the transcriptome in many organisms, e.g., yeast, flies, worms, plants, mice, and humans [13]–[23]. Gene coexpression network methods have also been used for typical microarray data analysis tasks such as gene filtering [19], [24]–[26] and outcome prediction [27],[28].
While the utility of network methods for analyzing microarray data has been demonstrated in numerous publications, the utility of microarray data analysis techniques for solving network theoretic problems has not yet been fully appreciated. One goal of this article is to show that simple geometric arguments can be used to derive network theoretic results if the networks are defined on the basis of a correlation matrix.
Although many of our network concepts will be useful for general networks, we are particularly interested in gene coexpression networks (also known as association-, influence-, relevance-, or correlation networks). Gene coexpression networks are built on the basis of a gene coexpression measure. The network nodes correspond to genes—or more precisely to gene expression profiles. The ith gene expression profile xi is a vector whose components report the gene expression values across m microarrays. We define the coexpression similarity sij between genes i and j as the absolute value of the correlation coefficient between their expression profiles:
Using a thresholding procedure, this coexpression similarity is transformed into a measure of connection strength (adjacency). An unweighted network adjacency aij between gene expression profiles xi and xj can be defined by hard thresholding the coexpression similarity sij as follows(1)where τ is the “hard” threshold parameter. Thus, two genes are linked (aij = 1) if the absolute correlation between their expression profiles exceeds the (hard) threshold τ. Hard thresholding of the correlation leads to simple network concepts (e.g., the gene connectivity equals the number of direct neighbors) but it may lead to a loss of information: if τ has been set to 0.8, there will be no link between two genes if their correlation equals 0.799. To preserve the continuous nature of the coexpression information, one could simply define a weighted adjacency matrix as the absolute value of the gene expression correlation matrix, i.e., [aij] = [sij]. However, since microarray data can be noisy and the number of samples is often small, we and others have found it useful to emphasize strong correlations and to punish weak correlations. It is natural to define the adjacency between two genes as a power of the absolute value of the correlation coefficient [19],[24]:(2)with β≥1. This soft thresholding approach leads to a weighted gene coexpression network. We present empirical results for weighted and unweighted networks in the main text, Text S1, Text S2, and Text S3.
Since humans are organized into social networks, social network analogies should be intuitive to many readers. Therefore, we will refer to the following ‘affection network’ throughout this article. Assume that n individuals filled out an interest questionnaire, which was used to define a pairwise similarity score sij. For convenience, we assume that the similarity measure takes on values between 0 and 1. Our definition of the affection network is based on the following assumption: the more similar the interests between two individuals, the more affection they feel for each other. More specifically, we assume that the affection (adjacency) aij between two individuals is proportional to their similarity on a logarithmic scale, i.e.,(3)This is equivalent to our soft thresholding approach aij = sijβ (Equation 2). A soft threshold β = 2 implies that the affection aij equals 0.25 if the similarity sij equals 0.5.
Many network applications use at least one gene significance measure. Abstractly speaking, we define a gene significance measure as a function GS that assigns a nonnegative number to each gene; the higher GSi the more biologically significant is gene i. We assume that the minimum gene significance is 0. For example, if a statistical significance level (p-value) is available for each gene, the gene significance of the ith gene can be defined as minus log of the p-value, i.e., GSi = −log(pi). In this article, we are particularly interested in gene significance measures that are based on a microarray sample trait, e.g., a clinical outcome. The microarray sample trait T = (T1,…,Tm) may be quantitative (e.g., body weight) or binary (e.g., case control status). Since our goal is to provide a simple geometric interpretation of coexpression network analysis, we define the trait-based gene significance measure by raising the correlation between the ith gene expression profile xi and the clinical trait T to a power β(4)Although any power β could be used in Equation 4, we use the same power as in Equation 2 to facilitate a simple geometric interpretation.
We find it convenient to express network quantities in terms of correlation coefficients since the correlation between two vectors can be interpreted as the cosine of the angle between them (measured in radians) if the vectors are scaled to have a mean of 0. Since the correlation is scale-invariant, i.e., cor(axi+b, cxj+d) = cor(xi,xj), we can assume without loss of generality that the vectors xi have a mean 0 and are of the same length. In other words, they correspond to points on a hypersphere.
The network adjacency aij is a monotonically decreasing function of the angle θij between the two scaled expression profiles if 0≤θij≤π/2. When the angle θij equals 0 or π/2, the adjacency equals 1 or 0, respectively. The network adjacency is a monotonically decreasing function of the length of the shortest path (geodesic) between the two points on the hypersphere. Soft thresholding methods (Equation 2) preserve the continuous nature of these distances. The higher the soft threshold β, the more weight is assigned to short geodesic distances compared to large distances.
Since the trait-based gene significance measure GSi = |cor(xi,T)|β, (Equation 4) is scale-invariant, the sample trait T can also be considered a point on the hypersphere. Analogous to the network adjacency, the smaller the geodesic distance between the ith gene expression profile and the trait T, the higher the gene significance of the ith gene. In other words, the smaller the angle between the sample trait and the expression profile, the more significant is the gene.
As a motivational example, we study the pairwise correlations among 498 genes that had previously been found to form a sub-network related to mouse body weight. The microarray data measure the expression levels in multiple tissue samples (liver, adipose, brain, muscle) from male and female mice of an F2 intercross. Approximately 100 tissue samples are available for each gender/tissue combination. The biological significance of this subnetwork is described in [23],[26]. Here we focus on the mathematical and topological properties of the pairwise absolute correlations aij = |cor(xi,xj)| between the genes. For each gender and tissue type Figure 1A depicts a hierarchical cluster tree of the genes. Figure 1B shows the corresponding heat maps, which color-code the absolute pairwise correlations aij. As can be seen from the color bar underneath the heat maps, red and green in the heat map indicate high and low absolute correlation, respectively. The genes in the rows and columns of each heat map are sorted by the corresponding cluster tree.
It is visually obvious that the heat maps and the cluster trees of different gender/tissue combinations can look quite different. Network theory offers a wealth of intuitive concepts for describing the pairwise relationships among genes that are depicted in cluster trees and heat maps. To illustrate this point, we describe several such concepts in the following. By visual inspection of Figure 1B, genes appear to be more highly correlated in liver than in adipose (a lot of red versus green color in the corresponding heat maps). This property can be captured by the concept of network density (defined below). The density of the female liver network is 0.39 while it is only 0.23 for the female adipose network. Another example for the use of network concepts is to quantify the extent of cluster (module) structure. In this example, branches of a cluster tree (Figure 1A) correspond to modules in the corresponding network. The cluster structure is also reflected in the corresponding heat maps: modules correspond to large red squares along the diagonal. Network theory provides a concept for quantifying the extent of module structure in a network: the mean clustering coefficient (defined below). The female liver, male liver and female brain networks have high mean clustering coefficients (mean ClusterCoef = 0.42, 0.43, 0.41, respectively). In contrast, the female adipose, male adipose, and male brain networks have lower mean clustering coefficients (mean ClusterCoef = 0.27, 0.27, 0.25, respectively). Difference in module structure may reflect true biological differences or they may reflect noise (e.g. technical artifacts or tissue contaminations).
As another example for the use of network concepts, compare the cluster tree of the female brain network with that of the male brain network. The cluster tree of the female network appears to be comprised of a single large branch, i.e., a highly connected hub gene at the tip of the branch forms the center in this network. In contrast, the cluster tree corresponding to the male brain network appears to split into multiple smaller branches, i.e., no single gene forms the center. To measure whether a highly connected hub gene forms the center in a network, one can use the concept of centralization (defined below). The female brain and male brain networks have centralization 0.34 and 0.21, respectively.
These examples illustrate that graph theory contains a wealth of network concepts that can be used to describe microarray data. But we will argue that microarray data analysis techniques can also be used to derive network theoretic results. For example, network theorists have long studied the relationship between gene significance and connectivity. Several network articles have pointed out that highly connected hub nodes are central to the network architecture [17], [29]–[32] but hub genes may not always be biologically significant [33]. To define a sample trait based gene significance measure (Equation 4), we define the gene significance of gene i as the absolute correlation between the gene expression profile xi and body weight T, i.e., GSi = |cor(xi,T)|. Figure 1C shows the relationship between this gene significance measure and connectivity in the different gender/tissue type networks. We find a strong positive relationship between gene significance and connectivity in the female and the male mouse liver networks. The positive relationship between gene significance and connectivity suggests that both variables could be used to implicate genes related to body weight. For example, we used connectivity as a variable in a systems biologic gene screening method [26]. While most network theorists would agree that connectivity is an important variable for finding important genes in a network [17],[19], the statistical advantages of combining gene significance and connectivity are not clear. Below, we use the geometric interpretation of coexpression network analysis to argue that intramodular connectivity can be interpreted as a fuzzy measure of module membership. Thus, a systems biologic gene screening method that combines a gene significance measure with intramodular connectivity amounts to a pathway based gene screening method. Empirical evidence shows that the resulting systems biologic gene screening methods can lead to important biological insights [23]–[26]. Before combining gene significance and connectivity in a systems biologic gene screening approach, it is important to study their relationship. Toward this end, we propose a measure of hub gene significance HGS as slope of a regression line (through the origin) between gene significance and scaled connectivity. As can be seen from Figure 1C, the hub gene significance is high in liver and adipose tissues but it is low in brain and muscle tissues. Below, we use the geometric interpretation of coexpression networks to characterize coexpression networks that have high hub gene significance if the gene significance measure is based on a microarray sample trait T.
One of the many biological applications of gene coexpression networks is the identification of pathways (modules) and centrally located genes (referred to as module centroids). In our applications, we define highly connected intramodular hub genes as module centroids. Weighted gene coexpression network analysis (WGCNA, [19],[24]) can be considered a step-wise microarray data reduction technique, which starts from the level of thousands of genes, identifies clinically interesting gene modules, and finally represents the modules by their centroids. The module centric analysis alleviates the multiple testing problem inherent in microarray data analysis. Instead of relating thousands of genes to a sample trait, it focuses on the relationship between a few (usually less than 10) modules and the sample trait.
An outline of WGCNA is presented in Figure 3A. The module definition does not make use of a priori defined gene sets. Instead, modules are constructed from the expression data by using a tight clustering procedure. Although it is advisable to relate the resulting modules to gene ontology information to assess their biological plausibility, it is not required. Because the modules may correspond to biological pathways, focusing the analysis on modules (and corresponding centroids) amounts to a biologically motivated data reduction method. Intramodular hub genes are centrally located in the module and thus lend themselves as candidates for biomarkers. Examples of biological studies that show the importance of intramodular hub genes can be found reported in [23]–[25],[33],[39]. Because the expression profiles of intramodular hub genes are highly correlated (in our data, r>0.90), typically dozens of candidates result. Although these candidates are statistically equivalent, they may differ in terms of biological plausibility or clinical utility.
Roughly speaking, we define network modules as groups of highly interconnected genes. As detailed in Text S1, Text S2, Text S3, and in our online R tutorials, we use a hierarchical clustering procedure to identify modules (clusters) as branches of the resulting cluster tree. A common but inflexible branch cutting method uses a constant height cutoff value. Alternatively, dynamic branch cutting adaptively chooses cutting values depending on the shape of the branch [40]. Each module is assigned a unique color label (Figure 3B). Our branch cutting algorithm only assigns module colors to branches whose size exceeds a user-specified threshold parameter. In practice, it is advisable to vary the minimum module size and other branch cutting parameters to determine how the results are affected by different parameter choices. An iterative approach for choosing the parameters could be defined by optimizing the module significance. This module detection approach has led to biologically meaningful modules in several applications [1], [8], [23]–[25], [33], [39]–[43] but our theoretical results transcend this particular module detection method. Any module detection method that results in clusters of highly correlated gene expressions could be used.
In the following, we assume that a module detection method (e.g., a clustering procedure) has found Q modules. We denote the adjacency matrix of the genes inside the qth module by A(q). Thus, A(q) represents a subnetwork comprised of the genes in the qth module. Analogously, we define GS(q) as the gene significance measure restricted to the module genes. Denote by n(q) the number of genes inside the qth module. Throughout the manuscript, we use the superscript (q) to denote quantities associated with the qth module. But for notational convenience, we sometimes omit (q) when the context is clear.
We define an intramodular network concept NCF(A(q),GS(q)) by evaluating a network concept function NCF(·,·) on the adjacency matrix A(q) and/or a corresponding gene significance measure GS(q).
For example, the intramodular connectivity is defined by(17)where the j indexes the genes in the qth module. Intramodular connectivity has been found to be an important complementary gene screening variable for finding biologically important genes [24],[25],[39].
We refer to the network significance (Equation 14) of a module network simply as the module significance measure, i.e., the module significance is the average gene significance of the module genes:(18)
The high dimensionality of gene expression data has inspired two broad categories of data reduction techniques. The first category, often used by network theorists, is to reduce the gene coexpression networks into modules. Each module can be represented by a centroid, e.g., an intramodular hub gene. The second category, often used by microarray data analysts, reduces the gene expression data to a small number of components that capture the essential behavior of the expression profiles [27], [44]–[51]. One of our goals is to understand how the two categories of data reduction methods relate to each other. Here we use the singular value decomposition [44],[45],[48] since this will allow us to define a simple measure of factorizability (Equation 24).
The module eigengenes of different modules can be highly correlated (Figure 4A). Detecting a high correlation between module eigengenes may either be of biological interest (suggesting interactions between pathways) or it may be a methodological artifact (suggesting poorly defined modules that should be merged). The correlations between two eigengenes can be used to define eigengene coexpression networks [52], e.g., a weighted eigengene coexpression network can be defined as follows(23)where E(q) and E(p) represent the eigengenes of two distinct modules. Apart from correlating the module eigengenes of different modules to each other, one can relate the module eigengenes to an external microarray sample trait T to identify trait related modules. Thus, eigengene network analysis can be viewed as a network reduction scheme that reduces a gene coexpression network involving thousands of genes to an orders of magnitude smaller metanetwork involving module representatives (one eigengene per module).
Unlike traditional microarray data reduction methods that impose orthogonality (e.g., principal component analysis) or independence (e.g., independent component analysis), gene coexpression network analysis can be considered a pathway-based data reduction method that allows dependencies between the modules. When focusing on the use of module eigengenes, network analysis can be considered a variant of oblique factor analysis.
While a high level view of modular gene coexpression networks can be viewed as a data reduction technique, many network analyses focus on the pairwise relationships of relatively few (hundreds) of correlated genes, i.e., genes that form a single module in a larger network. For example, the 498 genes of our motivational example were part of a body weight related module, which was found in a large gene coexpression network based on the female mouse liver samples [23].
The low-level analysis of a single network module may help identify key genes that may be used as therapeutic targets or candidate biomarkers. An important question of low level analysis is to efficiently describe the connection strengths between interacting module genes. We have provided empirical evidence that many module adjacency matrices, i.e., networks comprised of genes of a single module, are approximately factorizable [8]. In such networks, the adjacency between module genes i and j can approximately be factored into gene specific contributions, i.e., aij(q)≈CFi(q)CFj(q) with CFi(q) defined as the conformity of gene i. Thus, the adjacency matrix of an approximately factorizable network can be approximated using the rank 1 matrix [CFi(q)CFj(q)]. The conformity vector CF(q) can be estimated in several ways [8]; it is highly related to a single factor nonnegative matrix decomposition of A(q) [51] and it is highly related to the connectivity .
Here we define eigengene-based network concepts as a step towards a geometric interpretation of network concepts. Analogous to the case of intramodular network concepts, we define eigengene-based network concepts by evaluating the network concept function NCF(AE(q),GSE(q)) on the eigengene-based adjacency matrix AE(q) (Equation 28) and the eigengene-based gene significance measure GSE(q) (Equation 29). One can easily derive the following formulas for eigengene-based network concepts:(30)where . Under the assumptions of Observation 1, we find that A(q)≈AE(q) and GSi≈GSE,i. For a continuous network concept function NCF(·,·) this implies NCF(A(q),GS)≈NCF(AE(q),GSE). We summarize this observation as follows
Here we illustrate how the geometric interpretation of gene coexpression networks can be used to derive results, which may be interesting to microarray data analysts.
In the following, we provide several examples that illustrate potential uses of the geometric interpretation.
To facilitate the communication between microarray data analysts and network theorists, we provide a short dictionary for translating between microarray data analysis and network theory terminology. More specifically, for a subset (module) of genes that have high expression factorizability, Table 1 describes the correspondence between general network terms and their eigengene-based counterparts. While our theoretical derivations assume a weighted gene coexpression network, our robustness studies show empirically that many of the findings apply to unweighted networks as well. The summary of empirical robustness studies is described below.
In general, eigengene-based concepts are no substitute for network concepts. It is natural to use network concepts when describing the pairwise relationships between genes and to use eigengene-based network concepts when relating the gene expression profiles to a module eigengene. Since eigengene-based network concepts tend to be relatively simple, they often simplify theoretical derivations. Further, many of them allow one to calculate a statistical significance level (p-value) using a correlation or regression based test statistic.
To illustrate the theoretical results we report 4 different microarray data applications. The underlying data sets and R software code can be found on our webpage http://www.genetics.ucla.edu/labs/horvath/ModuleConformity/GeometricInterpretation/.
Network theoretic methods and concepts are increasingly used for the systems biologic analysis of microarray data. We illustrate how network concepts can be used for describing large correlation matrices and for arriving at biologically plausible data reduction techniques. Many alternative approaches for defining gene coexpression networks are possible, e.g., [13], [55]–[61]. Here we define the network adjacency and the gene significance measure in terms of correlations since this allows us to interpret pairwise relations in terms of angles between scaled versions of the variables. For example, the sample trait based gene significance measure of the ith gene is determined by the angle between the ith gene expression profile and the sample trait T (Equation 4); the scaled intramodular connectivity of the ith gene (Equation 33) is determined by the angle between the ith gene expression profile and the module eigengene; the hub gene significance (Equation 34) is determined by the angle between module eigengene and the sample trait.
The geometric interpretation of gene coexpression network analysis reveals a deep connection to other statistical methods. Since it projects the gene expressions profiles onto the hypersphere in an m-dimensional Euclidean space, network analysis can be considered a special case of directional statistics. When focusing on the use of module eigengenes, network analysis can be considered a variant of oblique factor analysis.
A high level view of modules and their centroids (eigengenes) can be used to define eigengene networks [52]. High correlations (small angles) between module eigengenes may suggest close relationships between the corresponding pathways. A low level view of a single module allows us to provide a geometric interpretation of intramodular network concepts. We use the singular value decomposition of module expression data to characterize approximately factorizable gene coexpression networks, i.e., adjacency matrices that satisfy aij(q)≈CFi(q)CFj(q). We provide an intuitive formula of the conformity CFi(q)≈|cor(xi(q),E(q))|β. Since the module eigengene E(q) summarizes the overall behavior of the module, the eigengene conformity |cor(xi(q),E(q))|β measures how well gene i conforms to the overall module. This insight led us to coin the term “conformity”. Using the singular values, we propose a measure of eigengene factorizability (Equation 24) that is analogous to the proportion of variance explained by the module eigengene (Equation 22). We provide a geometric interpretation of network factorizability in Figure 5A.
The derivation of Observation 1 in the Methods section highlights a theoretical advantage of the soft-thresholding approach (Equation 2); the resulting weighted network maintains the approximate factorizability of the underlying correlation matrix: aij(q) = |cor(xi(q),xj(q))|β≈|cor(xi(q),E(q))cor(xj(q),E(q))|β = |cor(xi(q),E(q))|β|cor(xj(q),E(q))|β.
Using multiple different gene coexpression networks from mouse tissues, brain cancer, and yeast, we provide empirical evidence that coexpression modules tend to have high eigengene factorizability and that the maximum conformity assumption (Equation 32) is satisfied for low powers of β.
We propose eigengene-based analogs of network concepts (Equation 30). While network concepts are functions of the adjacency matrix, eigengene-based network concepts are analogous functions of the eigengene conformities |cor(xi(q),E(q))|β. Algebraically, eigengene-based network concepts are closely related to “approximate conformity based” network concepts [8] but they allow for a geometric interpretation.
We use the correspondence between intramodular network concepts and their eigengene-based analogs to provide a geometric interpretation of network concepts. Observation 2 states that network concepts in weighted gene coexpression module networks are approximately equal to their eigengene-based analogs. A major theoretical advantage of eigengene-based network concepts is that they reveal simple relationships. To arrive at particularly simple relationships, we make the maximum conformity assumption (Equation 32) for the results presented in the main text. Table 1 provides a rough dictionary for translating between gene coexpression network analysis and the singular value decomposition if the underlying expression data have high eigengene factorizability (say EF(X(q))>0.95) and if the maximum conformity assumption (Equation 32) is satisfied. However, even if the maximum conformity assumption does not hold, one can still find simple relationships among the network concepts (Equation 49).
The geometric interpretation of gene coexpression networks facilitates the derivation of several results that should be interesting to network theorists. For example, we argue that highly connected intramodular hub genes cannot be intermediate between two distinct coexpression modules (Figure 5B). The geometric interpretation is particularly useful when studying gene significance and module significance measures that are based on a microarray sample trait (Equation 4). To study the relationship between connectivity and gene significance, we propose a novel measure of hub gene significance (Equation 13). We find that the hub gene significance of a module network is determined by the angle between the module eigengene and the microarray sample trait (Equation 34). Our geometric interpretation of coexpression networks allows us to describe situations when a module has low hub gene significance (Figure 5C and 5D). Our theoretical derivations for relating module significance to hub gene significance (Equation 37) assumes a gene significance measure based on a sample trait. Although this important assumption is violated for the gene significance measure (knock-out essentiality) in the yeast network, it is striking that the relationship between hub gene significance and module significance can still be observed in this application (Figure 9).
We provide a robustness analysis that shows that many of our theoretical results apply even if our underlying assumptions are not satisfied (Figures 6 and 9, Tables 3, 5, and 6, Text S1, Text S2, and Text S3). We find that the correspondence between network concepts and their eigengene-based analogs is often better in weighted networks than in unweighted networks. Further, we find that the results in weighted networks tend to be more robust than those in unweighted networks with regard to changing the network construction thresholds β and τ, respectively. Thus, weighted coexpression networks are preferable over unweighted networks when a geometric interpretation of network concepts is desirable.
The correspondence between coexpression module networks and the singular value decomposition (Table 1) can break down when a high soft threshold is used for constructing a weighted network or when dealing with an unweighted network. Thus, eigengene-based concepts do not replace network concepts when describing interaction patterns among genes.
While this article has a theoretical bent, we illustrate the results on three different microarray data sets (human, mouse, and yeast) that are described in our online R software tutorials, in Text S1, Text S2, and Text S3. Our theoretical results also apply to networks comprised of genes that are highly correlated with a sample trait. The key assumption underlying our results is high eigengene factorizability EF(X(q)). To illustrate this point, Text S4 describes a brain cancer network comprised of the 500 genes with highest absolute correlation with brain cancer survival time. Our results illustrate that the geometric interpretation of gene coexpression networks has important theoretical and practical implications that may guide the development and application of network methods.
Analogous to [8], we define a network concept function to be function of a square matrix M = [Mij] (1≤i,j≤n) and/or a corresponding vector G = (G1,…,Gn). For example, M could be the adjacency matrix (with diagonal set to 0) and G could be a corresponding gene significance measure.
We make use of the following network concept functions:(42)where the components of matrix BM in the denominator of the clustering coefficient function are given by bij = 1 if i≠j and bii = Ind(mii>0). Here the indicator function Ind(·) takes on the value 1 if the condition is satisfied and 0 otherwise.
According to our convention, the diagonal elements of the adjacency matrix are set to 1. Therefore, the diagonal elements of A–I (where I denotes the identity matrix) equal 0. Now we are ready to define the (fundamental) network concepts that are studied in this article.
Definition of Fundamental Network Concepts: The fundamental network concepts of a network A are defined by evaluating the network functions (Equation 42) on A–I and the gene significance measure GS, i.e.,
For example, the connectivity is given by(43)
We define an intramodular network concept NCF(A(q)−I,GS(q)) by evaluating the network concept function on the restricted adjacency matrix A(q) and the restricted gene significance measure GS(q).
We will now define eigengene-based network concepts. Using the eigengene-based adjacency matrix AE(q) = ae(q)(ae(q))T (Equation 28) and the eigengene-based gene significance measure GSE,i(q) = ae,i(q)ae,t(q) (Equation 29), we define an eigengene-based network concept as NCF(AE(q),GSE(q)).
As example, consider the eigengene-based connectivity given by(44)
Here we derive Observation 1, which characterizes approximately factorizable gene coexpression module networks. To simplify the presentation, we omit the superscripts (q) in the following, e.g., we will write EF(X) instead of EF(X(q)). We will argue that if the eigengene factorizability EF(X) is close to 1, the adjacencies of the weighted coexpression module network A = |cor(X)|β and the trait-based gene significance measure GSi = |cor(xi,T)|β can be factored as follows(45)where(46)(47)
Since our gene coexpression networks are defined with respect to the correlation matrix [cor(xi,xj)], which is scale-invariant, we can assume that the gene expression profiles have been scaled as follows: where m is the number of microarray samples. Then one can derive the following relationshipsNote that u1,i|d1|2u1,j/m = cor(xi,E)cor(xj,E). Using the fact that U is an orthogonal matrix, it is straightforward to show that
This equation motivates us to propose the following measure of eigengene factorizability:(48)Note that 0≤EF(E)≤1. By definition EF(E)≈1 implies thatBy raising both sides of this equation to a power β, we findThe last step highlights an important theoretical advantage of the soft thresholding method: it preserves the approximate factorizability of the underlying correlation matrix.
An alternative, possibly more direct way of motivating the observation is based on the insight that the squared singular values |dl|2 correspond to the eigenvalues of the correlation matrix COR = [cor(xi,xj)]. For high values of EF(E), the correlation matrix can be factored as followswhere u1 denotes an eigenvector of length 1.
Here we describe relationships among eigengene-based network concepts if the maximum conformity assumption does not hold (i.e., ae,max(q)<<1). It is straightforward to derive the following relationships among eigengene-based network concepts:(49)Observation 2 can be used to derive the following
In the following we provide details on our geometric interpretation of the factorizability. To simplify the notation, we sometimes drop the superscript (q) in the following expressions. We denote by θl,i the angle between the right singular vector vl (Equation 20) and the ith gene expression profile xi. The smaller the angle θl,i, the bigger the correlation cor(vl,xi) = cos(θl,i). Using , one can reexpress the eigengene factorizability (Equation 24) as follows(50)
Thus, EF(X(q))≈1 if the module gene expressions xi are approximately orthogonal (cos(θl,i)≈0) to the right singular vectors vl for l≥2, i.e., if on average the gene expression profiles point in the direction of the module eigengene v1 = E.
Under this assumption, we provide a rough geometric intuition of aij≈ae,iae,j (Equation 25) depicted in Figure 5A. We denote by θi = θ1,i the angle between the module eigengene E and the ith gene expression profile and by θij the angle between gene expression profiles i and j. Using the assumptions described in Figure 5A, θij≈|θi±θj| and sin(θi) sin(θj)≈0, we find that(51)i.e., the correlation matrix is approximately factorizable.
Here we prove that the eigengene-based heterogeneity increases with the soft threshold β (Equation 2). Recall that (Equation 30) which implies that it is a decreasing function of(52)Note that ai = |cor(xi,E)| is a nonnegative number.
To prove that the heterogeneity increases with β, it suffices to prove the following
Proposition: Let {ai, i = 1,…,n} be a group of nonnegative number and β>1 then the following inequality holds:(53)
To prove the Proposition, we will make use of the following
Lemma: Let {ui, i = 1,…,n} and {vi, i = 1,…,n} be groups of nonnegative numbers, and θ be a number 0≤θ<1. Then the following inequality holds:(54)The Lemma can be proved with Hölder's inequality, which is given by(55)We use the Lemma with θ1 = β/(2β−1), ui = ai, and vi = ai2β to deriveFurther, we use the Lemma with θ2 = (2β−2)/(2β−1), ui = ai, and vi = ai2β to deriveBy squaring the first inequality and multiplying it with the second inequality, we arrive atsince 2θ1+θ2 = 2 and 3−(2θ1+θ2) = 1. The last inequality completes the proof since it is equivalent to the inequality in Equation 53.
|
10.1371/journal.ppat.1005651 | A Miniaturized Screen of a Schistosoma mansoni Serotonergic G Protein-Coupled Receptor Identifies Novel Classes of Parasite-Selective Inhibitors | Schistosomiasis is a tropical parasitic disease afflicting ~200 million people worldwide and current therapy depends on a single drug (praziquantel) which exhibits several non-optimal features. These shortcomings underpin the need for next generation anthelmintics, but the process of validating physiologically relevant targets (‘target selection’) and pharmacologically profiling them is challenging. Remarkably, even though over a quarter of current human therapeutics target rhodopsin-like G protein coupled receptors (GPCRs), no library screen of a flatworm GPCR has yet been reported. Here, we have pharmacologically profiled a schistosome serotonergic GPCR (Sm.5HTR) implicated as a downstream modulator of PZQ efficacy, in a miniaturized screening assay compatible with high content screening. This approach employs a split luciferase based biosensor sensitive to cellular cAMP levels that resolves the proximal kinetics of GPCR modulation in intact cells. Data evidence a divergent pharmacological signature between the parasitic serotonergic receptor and the closest human GPCR homolog (Hs.5HTR7), supporting the feasibility of optimizing parasitic selective pharmacophores. New ligands, and chemical series, with potency and selectivity for Sm.5HTR over Hs.5HTR7 are identified in vitro and validated for in vivo efficacy against schistosomules and adult worms. Sm.5HTR also displayed a property resembling irreversible inactivation, a phenomenon discovered at Hs.5HTR7, which enhances the appeal of this abundantly expressed parasite GPCR as a target for anthelmintic ligand design. Overall, these data underscore the feasibility of profiling flatworm GPCRs in a high throughput screening format competent to resolve different classes of GPCR modulators. Further, these data underscore the promise of Sm.5HTR as a chemotherapeutically vulnerable node for development of next generation anthelmintics.
| Parasitic flatworms express a diverse array of G protein coupled receptors, but our knowledge of their pharmacological profile is limited. No high throughput screen of a flatworm GPCR has been reported, even though these targets have precedent for high druggability and functionality in the chemotherapeutically vulnerable excitable cell niche. The goal of this study was to establish a method for profiling flatworm G protein coupled receptors that can be scaled to high content screening. Using a cAMP biosensor, we have performed a proof of principle miniaturized screen on a schistosome serotonergic GPCR that resolves new ligands that potently and selectivity block 5-HT receptor activity in vitro, and 5-HT evoked responses in schistosomules and adult worms. This approach evidences the pharmacological divergence of a parasitic GPCR from the closest human homolog and a capacity for high content interrogation of flatworm GPCR properties and ligand specificities.
| The neglected tropical disease Schistosomiasis is the most socioeconomically devastating helminth infection, and the second most burdensome parasitic infection behind malaria, infecting over 200 million people worldwide [1]. Infected individuals are treated by the drug praziquantel (PZQ), the mainstay therapeutic for disease control.
PZQ was originally developed during the 1970s, and the continued effectiveness of this agent over four decades of usage for treating a variety of parasitic infections has proven critically impactful [1]. Indeed this clinical efficacy has ironically proven to be a factor that has restrained efforts to develop alternative therapies, and at the most basic level, define how PZQ works. However several features of PZQ remain less than ideal and require improvement. First, our lack of mechanistic understanding of how PZQ works has proved a roadblock in the rational design of new drugs. There is a need to identify new druggable targets that exploit broader vulnerabilities within PZQ-sensitive pathways [2–4]. Second, our inability to improve on PZQ by chemical derivatization of the drug: all PZQ derivatives synthesized to date are less effective than the parent compound. The need is to identify novel structural pharmacophores that impair parasite viability. Third, the inability of PZQ to kill all parasitic life cycle stages. Juvenile worms are refractory to PZQ [5,6], possibly a contributory factor driving development of drug resistance [5,7]. The need is to identify new targets expressed throughout all lifecycle stages that are ideally conserved in other PZQ-sensitive parasites. Fourth, sub-optimal cure rates in the field: PZQ requires multiple drug dosings to achieve maximal cure rates for schistosomiasis, a regimen which is not always executed in mass drug administration efforts [8,9]. Therefore, there is clear opportunity to improve on the clinical penetrance of PZQ. These issues support efforts to identify new, druggable targets for development of next generation anthelmintics.
A logical place to start is with downstream effectors within the broader PZQ interactome. Over the last few years, therefore, our laboratory has attempted to bring fresh perspective to understand how PZQ works based upon a serendipitous basic science finding. During regeneration of the planarian flatworm D. japonica–a widely used regenerative biology model [10]–PZQ miscued polarity signaling to cause regeneration of bipolar (‘two-headed’) worms with dual, integrated organ systems [11]. This visually striking phenotype, coupled with the tractability of the planarian system to in vivo RNAi, allowed us to progressively define pathways engaged by PZQ in vivo [11–14]. These studies culminated in a model where PZQ acts as an ergomimetic [13] with in vivo PZQ efficacy regulated by the opposing functionality of dopaminergic and serotonergic neurons [11–14], known regulators of muscular activity, the tissue where planarian polarity determinants reside [15]. The serotonergic and dopaminergic G protein coupled receptors (GPCRs) engaged by activity of these bioaminergic neurons therefore represent potential downstream PZQ effectors. Their engagement by ligands, as shown for bromocriptine and other ergot alkaloids, phenocopy PZQ action in vivo [13,14].
This is an important realization as flatworm G protein coupled receptors (GPCRs) are logical candidates for antischistosomal drug development efforts. Over one quarter of current therapeutics target rhodopsin-like GPCRs [16]. However, barriers have been a lack of understanding of the physiology of specific GPCRs from within the broad GPCR portfolio (~75–120 in S. mansoni [17–19]) expressed by these organisms, as well as struggles optimizing functional expression of individual flatworm GPCRs in heterologous assay systems. However several groups have now begun to define a role for specific GPCRs within the chemotherapeutically vulnerable excitable cell niche [13,20–22], highlighting the key challenge of optimizing robust platforms for pharmacologically profiling these GPCRs in a miniaturized format compatible with high throughput screening (HTS). To our knowledge, no library screen of a flatworm GPCR has yet been reported. Prior studies have simply relied on interrogation of expressed GPCRs against handfuls of ligands selected around inferred agonist specificity.
Therefore the goal of this study was to establish a method for profiling flatworm GPCRs that can be effectively scaled to HTS. Our priorities for a platform were: first, a robustness for miniaturization into a multiwall plate format to permit chemical library screening, and second, use of a proximal readout of receptor activity within intact cells to enable real time monitoring of GPCR activity that can resolve different types of modulators (full, partial and inverse agonists, allosteric modulators). One technology that fulfills these requirements employs a bioluminescent cAMP reporter to monitor the activity of Gs and Gi-coupled GPCRs, marketed as GloSensor. The assay is based upon a crucially permutated form of firefly luciferase incorporating a cAMP-binding domain from PKA, such that cAMP-binding causes a conformational change in the enzyme that enhances the luminescent signal [23]. The dynamic range and sensitivity of the biosensor has been shown to be compatible with a variety of HTS assays [23,24].
To evaluate this technology, we applied this approach to pharmacologically profile a S. mansoni serotonergic GPCR (Sm.5HTR) that has been shown in vitro to respond to 5-HT through elevation of cAMP [20]. Sm.5HTR is the parasitic homologue of the planarian serotonergic GPCR (S7.1) that we have recently shown modulates the efficacy of PZQ in vivo [13]. However, as with most flatworm GPCRs, little is known about the pharmacology of this receptor. An initial characterization revealed blockade of 5-HT evoked signals in the presence of high concentrations (100μM) of mammalian bioaminergic blockers [20]. Here, we have applied the GloSensor assay in a proof of principle pilot screen for flatworm GPCR modulation. Our data evidence the extent of pharmacological divergence between the schistosome receptor and the human 5-HT7-receptor homolog (Hs.5HT7R), and reveal new ligands and compound series selective for the parasitic GPCR. Finally, despite these differences in ligand selectivity, we demonstrate conservation of an unusual antagonist-evoked inactivation mechanism for Sm.5HTR, a pharmacological phenomenon also exhibited at Hs.5HT7R [25,26], where exposure to a subset of antagonists results in a prolonged inactivation of signaling activity from the receptor. This property enhances the attractiveness of Sm.5HTR as an anthelmintic drug target.
In schistosome parasites, 5-HT is myoexcitatory: exogenous addition of 5-HT to schistosomules causes an increase in basal contractility and 5-HT also increases mobility of adult worms [27,28]. While this action has long been known, it is only in the last several years that the relevant receptors mediating the effects of 5-HT in flatworms have been identified [13,20]. The most abundant schistosome 5-HT receptor in adult worms from transcriptomic analysis [29], is a recently characterized GPCR christened Sm.5HTR [20].
Expression of an epitope tagged Sm.5HTR construct in HEK293 cells resulted in expression of a ~56 kDa product, consistent with the predicted size (Fig 1A). To assess functionality of this receptor, we utilized the GloSensor cAMP assay as a real-time luminescent readout of cellular cAMP levels. This ‘biosensor’ monitors luminescence from a firefly luciferase that is engineered to be cAMP sensitive by incorporation of a cAMP binding domain into the recombinant luciferase. The presence of substrate and cAMP results in an enhanced luminescence from the transfected GloSensor construct (Fig 1B), allowing real time monitoring of cAMP levels within intact cells. This can be seen in HEK293 cells transfected with both Sm.5HTR and GloSensor, where application of 5-HT evoked an increase in luminescence values over time (Fig 1C). No changes in cAMP were elicited in HEK293 cells transfected with the biosensor alone (Fig 1C).
Measurements of assay sensitivity were made from 5-HT evoked luminescence signals in cells plated in 96-well plates transfected with cAMP biosensors exhibiting either high affinity (‘F20’ construct) or low affinity (‘F22’ construct) for cAMP (Fig 2A). As expected, the magnitude of the luminescence signal varied with 5-HT application in a dose-dependent manner with both biosensor constructs in Sm.5HTR transfected cells (Fig 2B). With the higher affinity 20F sensor, the EC50 for cAMP generation was 703±90nM (n = 3, Fig 2B and Table 1), with the dose response relationship shifting to higher values with the 22F sensor as previously established [23]. The magnitude of the response was greatest in media supplemented with 3-isobutyl-1-methylxanthine (IBMX, 200μM) to block cAMP degradation. In the presence of IBMX, higher overall luminescence values were recorded with peak signal to background changes of ~1.7-fold and ~15.1-fold for 20F and 22F respectively (Fig 2A and 2B), providing a good signal to background window for monitoring receptor activation.
The robustness of these cAMP assays was assessed by calculating the Z’ factor (Z’), a widely used indicator of assay quality in high throughput screening applications [30]. Z’ values over 0.5 are considered a prerequisite for executing high throughput screens. Calculations of Z’ were made at different timepoints during the agonist response, averaging 6 replicate wells within a 96 well plate. In our hands, the highest Z’ scores were obtained with the F22 sensor supplemented with IBMX (Fig 2C), and these conditions were used for all subsequent assays. Acceptable Z’ values were also obtained with cells in suspension (Table 1) and under conditions of further miniaturization to 384-well plates (Fig A in S1 Text).
Unlike assays requiring cell lysis for fixed timepoint measurement, the live cell biosensor allowed real time monitoring of cellular cAMP levels throughout ongoing experimental manipulations. Fig 2D demonstrates antagonism of 5-HT stimulated cAMP generation by the antipsychotic methiothepin [31], with cellular responsiveness demonstrable by the subsequent addition of forskolin. Dose response analyses also confirmed preferential activation of Sm.5HTR by 5-HT compared with other bioaminergic agonists (Fig B in S1 Text). Finally, we used this assay to compare responsiveness from two different isoforms of Sm.5HTR which have been isolated–the originally published sequence Sm.5HTR [20] and a longer isoform (Sm.5HTRL) containing addition sequence at the NH2-terminus and within the third intracellular loop (Fig 2E, inset). Both isoforms were activated by 5-HT, with Sm.5HTRL displaying ~10-fold greater sensitivity (EC50 0.2±0.03μM vs 2.0±0.2μM, Fig 2E) but a similar kinetic response (Fig 2F).
Sequence homology identifies Sm.5HTR as a member of the SER7 clade of serotonin receptors, clustering with planarian S7 receptors [13] and with Hs.5HT7R as the closest human homolog [20]. To characterize the extent of pharmacological conservation between the parasite and human serotonin receptor, we used the miniaturized cAMP assay to screen a commercial GPCR compound library (~250 compounds) for inhibitors of these receptors. An inhibitor screen was prioritized simply because of the improved likelihood of detecting antagonists over agonists (need to exclude false positives from stimulation of endogenous receptors), and the utility of these agents for blocking parasite motility.
The protocol for screening is shown schematically in Fig 3A. HEK293 cells transiently transfected with either the human 5HT7 receptor (Hs.5HT7R) or the schistosome receptor (Sm.5HTR) were exposed to test ligands in a 96-well plate format. After addition of test compounds, 5-HT was then added to each well at a concentration corresponding approximately to the EC80 of each receptor to assess blockade of 5-HT effects by prior compound addition. Luminescence was then read at a fixed time point (t = 60min, Fig 3A). Hits were assigned as compounds that evoked a ≥50% decrease in luminescence output at the fixed time sampling point (Fig 3B). These experiments identified 25 compounds as potential antagonists of Sm.5HTR evoked cAMP generation (Fig 3B).
Two sets of validation experiments were then performed in order to remove false ‘hits’ from the dataset. First, the same library was also screened against naive HEK293 cells as a control for responses resulting from engagement of endogenous GPCRs (Fig 3C). This analysis identified 14 compounds in the library that activated endogenous Gs-coupled GPCRs in HEK293 cells [32]. Second, to exclude ligands that inhibited either cAMP production (for example, through activation of endogenous Gi-coupled GPCRs) or directly impaired the activity of the luciferase biosensor, the library was screened against forskolin-evoked increases in cAMP (Fig 3D). This analysis identified 7 compounds that decreased luminescence values >2-fold in forskolin-treated control cells. These 21 compounds were ‘masked’ from the experimental dataset and the overall pharmacological profile of Sm.5HTR and Hs.5HTR7 were then represented as a heat map to depict ligand-evoked changes in cAMP levels (Fig 3E). This visual representation conveys in a simple manner the extent of pharmacological divergence between the human and schistosome GPCRs. Some drugs displayed a unique affinity for Sm.5HTRs, others preferentially modulated Hs.5HT7R, and some ligands blocked both receptors. Overall, 23 compounds were retained for subsequent validation as antagonists of Sm.5HTR with only a minor proportion of these compounds (7 ‘hits’) showing inhibition at both the human and parasite receptor (Fig 3E, inset). A simple overview of the pharmacological specificity of the compounds identified as antagonists using the ligand classification key associated with the library was also informative (Fig C in S1 Text). The types of ligand classes–if not compound identities–that inhibited each serotonergic GPCR was broadly similar. The only notable difference was blockade of Sm.5HTR by some cholinergic ligands, which was not apparent for Hs.5HT7R.
To confirm ‘hit’ validity, complete dose response relationships were then examined for all compounds that inhibited 5-HT evoked signals by ≥50%. Examples of these assays (Fig 4) confirm the designation of compounds showing selective inhibition of the parasite serotonin receptor (Fig 4A, top), blockade of 5-HT receptors from both species (Fig 4A, middle) and preferential antagonism of the human 5-HT receptor (Fig 4A, bottom). Calculation of a selectivity ratio (IC50 (Hs.5HT7R) / IC50 (Sm.5HTR)) for these antagonists (Fig 4B) revealed a broad continuum of GPCR selectivity among from the screened compounds. Four ligands demonstrated clear selectivity for Sm.5HTR (alfuzosin, orphenadrine, atomoxetine and rotundine, Fig 4A and 4B), of which rotundine displayed the most sensitive IC50 value (IC50 = 701±207nM). These ligands also inhibited 5-HT evoked cAMP generation through Sm.5HTRL (Fig D in S1 Text). However, none of these compounds directly affected biosensor luminescence or cell viability of untransfected cells of at screened dosages (Fig E in S1 Text).
While the above data provide proof of principle for interrogation of a flatworm GPCR against a compound library in a miniaturized format, we were also curious to use the assay to investigate the properties of specific ligands prioritized by our prior work. First, we were interested in profiling specific ergot alkaloids on the basis of observations showing these compounds act as efficacious modulators of flatworm physiology [27,33,34]. Certain ergot alkaloids inhibit schistosomule contractility, while others stimulate hyperactivity [13]. In regenerating planarians, the ergopeptide bromocriptine evoked bipolarity at concentrations 100-fold less than PZQ [14], implying a potency of this class of agents against flatworm bioaminergic receptors. However, the structure-activity relationships (SAR) of ergots at flatworm GPCRs and relative selectivity over human receptors is unknown. Second, we have previously suggested an ergomimetic quality to PZQ action, raising the possibility that PZQ itself acts as a direct ligand of flatworm bioaminergic receptors likely as a serotonergic antagonist [13]. Therefore, screening for PZQ activity against Sm.5HTR was also investigated. Third, Hs.5HTR7 displays a property of pseudo-irreversible antagonism, where a subset of ligands effect a persistent inactivation of the receptor persistent beyond the duration of drug exposure [25,26]. Is this phenomenon conserved at Sm.5HTR? Finally, guided by the chemical library data, we performed a secondary screen of compounds structurally related to ‘hits’ from the initial drug screen (‘SAR by commerce’). Each of these experiments are discussed in turn below.
First, is Sm.5HTR activity modulated by ergot alkaloids? Several ergot alkaloids were screened against Sm.5HTR and these experiments revealed agonist activity of ergotamine and dihydroergotamine, which have previously shown to stimulate the basal contractility of schistosomules [13]. Ergotamine and dihydroergotamine were more potent (EC50 of 232nM and 315nM, respectively) than 5-HT (EC50 of ~1μM, Fig 5A), but with a lower maximal response suggestive of partial agonism. By contrast, other ergoline ligands, bromocriptine, metergoline and the hallucinogen lysergic acid diethylamide (LSD), an agonist at vertebrate 5-HT2A receptors, exhibited no efficacy at Sm.5HTR (Fig 5A). To investigate further the nature of these inactive ligands, the ability of increasing doses of bromocriptine (Fig 5B), metergoline (Fig 5C) and LSD (Fig 5D) to modulate 5-HT evoked cAMP accumulation at the Sm.5HTR was assessed. Each of these ergot ligands caused a right-shift in the 5-HT dose-response relationship consistent with competitive antagonism (Fig 5B–5D). At higher concentrations (>10μM) bromocriptine and LSD showed almost complete inhibition of 5-HT evoked cAMP generation. To quantify the extent of antagonism, a Schild regression analysis [35,36] was performed which yielded affinity constants (KB) of 410 nM for bromocriptine, 629nM for LSD and 4530 nM for metergoline (Fig 5F). These data show that ergot alkaloid derivatives can act as potent modulators of schistosome 5-HTRs.
Second, screening of PZQ against Sm.5HTR in this assay did not reveal any modulation of receptor activity over doses that would convey an antiparasitic effect (Fig 5A and 5E).
Third, to investigate the properties of antagonists at Sm.5HTR, we compared the action of bromocriptine (a known ‘irreversible antagonist’ of Hs.5HTR7 [26]) with the competitive antagonist cyproheptadine. While both antagonists acutely inhibited Sm.5HTR function (Fig 6A), inhibition evoked by bromocriptine persisted after antagonist wash-out while cyproheptadine inhibition was fully reversed by 1 hour after ligand removal (Fig 6B). Expanding this assay to other ligands revealed long-lasting inhibition with several ligands previously established as pseudo-irreversible antagonists at Hs.5HT7R (methiothepin, bromocriptine, lisuride, risperidone and metergoline) but not with the competitive blockers clozapine and cyproheptadine (Fig 6C and 6D). The most potent ligands were bromocriptine, methiothepin and lisuride (Fig 6E). Therefore, although ligand specificities of these GPCRs are divergent, a unique aspect of receptor phenomenology is conserved between the human and parasite receptor.
Finally, we profiled compounds structurally related to those compounds prioritized from the library screen in terms of parasite selectivity. As two of these top hits were dimethoxyisoquinoline derivatives (rotundine, alfuzosin) we focused on agents containing this moiety. Slight modifications of rotundine structure were sufficient to alter the GPCR inhibition profile (Fig F in S1 Text), as reflected by comparison of berberine/palmatine (decreased potency and selectivity for Sm.5HTR) and tetrabenazine (selectivity for Sm.5HTR retained). Similarly, comparison of the closely related structures tetrandrine and berbamine suggested a discriminating structure-activity profile for Sm.5HTR (Fig F in S1 Text).
Evaluation of structural data from all these assays provides insight to the structural selectivity between parasite and human receptors. Fig 7 arrays worm IC50 values versus human IC50 values, such that compounds with submicromolar IC50 values and selectivity for the parasite Sm.5HTR receptor fall into the bottom right quadrant. As expected, given the historical bias in ligand design for affinity toward human receptors, most compounds favor the human receptor (falling ‘above the line’ in Fig 7). For example, most of the screened tricyclic and tetracyclic antidepressants show higher affinity for Hs.5HT7R (12/13 compounds screened). Similarly, ligands with phenyl-sulfonyl groups (SB 269970, SB742457) that are potent inhibitors of Hs.5HT7R [37] (Fig 4), completely lacked activity at Sm.5HTR. In contrast, compounds exhibiting potency and selectivity toward the parasite receptor (bottom right quadrant) were the ergot alkaloid bromocriptine and several dimethoxyisoquinoline compounds revealed by our experiments (rotundine, tetrabenazine, tetrandrine).
Do these hits from the Sm.5HTR screen in vitro translate into effectiveness against parasites? To assess this issue, selected compounds were screened for effects on schistosomule contractility. Schistosomules exhibit a basal level of spontaneous contractile activity (Fig 8A) which provides a simple phenotype for assaying neuromuscular activity. In this system, 5-HT is myoexcitatory: exogenous addition of 5-HT causes an increase in the basal contractile rate in a dose-dependent manner (Fig 8B). Subsequent addition of the four compounds validated above as potent blockers of Sm.5HTR (rotundine, tetrabenazine, tetrandrine, bromocriptine) were examined. Three of these compounds–rotundine, tetrandrine and bromocriptine–all potently inhibited 5-HT evoked contractions (IC50≤1μM). Tetrabenazine was however less efficacious in vivo, effecting only a ~50% inhibition of 5-HT evoked contractility at the highest dose (100 μM). Therefore three of the four compounds prioritized by the Sm.5HTR screening data conferred an inhibitory action against schistosomules.
The action of Sm.5HTR ligands were then examined against adult schistosome worms in vitro. Isolated worms exhibited basal mobility, and application of 5-HT significantly increased the movement of unpaired male and female worms (Fig 9A). Basal movement and the magnitude of the 5-HT evoked stimulation differed between males and females (Fig 9A). These effects was quantified by performing dose-response relationships (Fig 9B). Sex differences in the magnitude of 5-HT evoked cAMP generation [38], Sm.5HTR transcript and Sm.5HTR protein expression have previously been reported [20,39]. The action of rotundine, tetrandrine and bromocriptine were then assessed against basal (Fig 9C) and 5-HT stimulated worm movement (Fig 9D). Addition of bromocriptine and rotundine markedly inhibited worm movements at rest whereas tetrandrine enhanced movements of isolated female worms (Fig 9C). Bromocriptine and rotundine also inhibited the 5-HT evoked increases in worm movement, and again tetrandrine lacked an inhibitory effect (Fig 9D). From these experiments, we conclude bromocriptine and rotundine also act as effective paralytics of adult schistosome worms.
Finally, we were interested in examining the kinetics of inhibition caused by bromocriptine and rotundine to probe whether the protracted inhibition of Sm.5HTR observed in vitro (Fig 6) was manifest in vivo. To do these experiments, worms were exposed to bromocriptine or rotundine, and then challenged with 5-HT at various points after drug removal. As expected both drugs inhibited basal worm movement, and blunted 5-HT evoked stimulation (Fig 10). The time course of reversal of these effects was then examined. For male worms, which were effectively paralyzed by both drugs (Fig 9C), the paralytic effects of rotundine were reversed within 3hrs of drug exposure (Fig 10A). However worms exposed to bromocriptine remained impaired for considerably longer, with recovery of movement being only demonstrable 24hrs after bromocriptine removal (Fig 10A). A similar timecourse of recovery from bromocriptine exposure was also resolved for female worms (Fig 10B). These data suggest that bromocriptine exposure evokes a protracted paralysis of adult schistosomes.
In this study we executed a proof of principle pilot screen to evidence the feasibility of pharmacological profiling a flatworm GPCR in HTS format. The assay system employed relied on an genetically encoded luminescent biosensor [23]. This is an appealing approach as this strategy is non-destructive and affords the ability to continually monitor the kinetics of cAMP generation from a single sample. Further by directly reporting cAMP levels, rather than transcriptional outcomes (e.g. cAMP reporter genes), this approach also reveals proximal receptor activity in real time to help discern how specific compounds are modulating GPCR activity. Acceptable Z’ scores were reliably obtained in 384 well format (Fig A in S1 Text), and we note the sensitivity of this approach has permitted responses from endogenous GPCRs to be resolved even in ultra-high throughput screening formats (3456-well plates). Obviously, this particular biosensor is suited only for Gs and Gi-coupled GPCRs, but the optimization of other biosensors—for example, genetically-encoded Ca2+ indicators [40], or reporters that are agonist independent [41]—should aid HTS profiling of other flatworm GPCRs coupling to different second messenger cascades.
The importance of unbiased profiling of flatworm GPCR targets is underscored by visualization of the entire dataset (Fig 3E) that underscores a divergence in ligand specificities between the schistosome 5-HT receptor (Sm.5HTR) and the closest human homolog (Hs.5HT7R, ~30% amino acid identity). This divergence in ligand specificity evidences concerns over use of established mammalian ligands to infer flatworm physiological mechanisms as many chemical probes used to study human 5HT7 receptors have modest effects on Sm.5HTR at similar concentrations. Examples include the sulfonyl derivatives SB269970, SB742457 and amisulpiride (Fig 4). The fact that 5-HT receptors in different organisms have evolved divergent characteristics profiles is of course unsurprising: the adult schistosome lives within the human host circulatory system, a 5-HT rich environment, where it continuously ingests and cycles 5-HT replete cells. The characteristics of the ligand binding site of Sm.5HTR may therefore necessitate adaptations to this niche. While this pharmacological divergence may limit repurposing efforts for existing serotonergic ligands that have been optimized toward human usage, it is nevertheless encouraging for de novo ligand discovery if pharmacological differences between flatworm and human receptors can be exploited to selectivity target parasite biology.
In this regard, our data identified several ligands were identified with a preference for Sm.5HTR over Hs.5HT7R, and reciprocally several ligand classes were deprioritized owing to an observed preferential selectivity for the human receptor (‘above the line’ in Fig 7B). These latter groupings included the sulfonyl derivatives discussed above, as well as tricyclic and tetracyclic antidepressants which have previously been shown to cause schistosomule hyperactivity [42], likely through inhibition of monoamine transporters [43]. Although features of these drugs that convey potency in schistosomules have been identified [42], these features do not necessarily convey selectivity (over Hs.5HT7R, or human 5-HT transporters). A similar case could be made for many ergot alkaloids, with the noted exception of bromocriptine, which was the only ergot screened to date exhibiting higher selectivity (~10-fold) for the parasitic 5-HT receptor (Fig 7). We had previously shown bromocriptine to displace 3H-dopamine in planarian binding assays [14], but clearly bromocriptine also possess potent anti-serotonergic properties in flatworms consistent with the polypharmacology of ergot alkaloids. The antagonist effect of LSD against Sm.5HTR (Fig 5B) was also unexpected, given LSD action as a serotonergic agonist in mammals. Further attention is needed to identify features of the ergoline scaffold that convey preferential modulation of parasitic 5-HTRs, given the ~100-fold range in IC50s observed (Fig 7).
In contrast to the observed human bias of many ligands, several compounds with preferential selectivity toward Sm.5HTR were resolved (Figs 4B and 6B). First, several modulators of biogenic amine transport, represented by compounds such as atomoxetine and fluoxetine. Fluoxetine is a serotonin reuptake inhibitor while atomexetine, a non-halogenated derivative of fluoxetine, is employed as a norepinephrine transporter inhibitor. Compounds of this class are known to block 5-HT GPCRs [44]. Second, and perhaps most striking, were ligands containing dimethoxyisoquinoline moieties (blue, Fig 7), several of which exhibited clear bias toward Sm.5HTR. These included two compounds with prior precedent as bioaminergic blockers—alfuzosin, a mammalian adrenergic (μ1) blocker, and rotundine which inhibits dopamine and serotonin binding at D1, D2 and 5-HT1A GPCRs [45]. Rotundine potently inhibited Sm.5HTR (IC50 ~700nM) while lacking any effect on Hs.5HT7 at concentrations up to 50μM. Chemical exploration of this scaffold is therefore further warranted, especially in light of the small number of compounds screened in this study. We also note that the stalwart anthelmintic PZQ is also a isoquinoline derivative, although direct interrogation of PZQ against Sm.5HTR did not yield any effect (Fig 5A and 5E). This does not preclude the possibility that PZQ acts as a ligand at another flatworm bioaminergic GPCR [13], one explanation for the functional antagonism observed between PZQ and 5-HT in planarians [13], schistosomules [13] and adult schistosomes [28,46].
Subsequent screening of bromocriptine (the most parasite selective ergot alkaloid, Fig 7), and the three promising isoquinoline ‘hits’ against both schistosomules and adult worms (Figs 8 and 9) revealed a clear inhibition of 5-HT evoked hypermotility from two of the four compounds (bromocriptine, rotundine) prioritized from the screen of heterologously expressed Sm.5HTR. The two other Sm.5HTR ligands (tetrabenazine, tetrandrine) were poorly effective. Such attrition of leads is expected when advancing candidates identified in vitro. For example, the stimulatory action of tetrandrine against adult worms (Fig 9), not observed in the schistosomule dataset (Fig 8), may reflect a counteracting stimulatory action at other schistosome GPCRs upregulated at the adult stage. Differences in GPCR expression [39] may also contribute to the observed differences in drug and 5-HT action between male and female worms (Fig 9). Overall, this was an encouraging translation from in vitro data to activity against different parasite life cycle stages, supporting the rational design and development of antiparasitic drugs aimed at schistosomal GPCRs.
Despite the divergence in pharmacological selectivity between the human and schistosome 5-HT GPCRs, it is worthwhile highlighting an important similarity between these receptors which may prove a boon for anthelmintic development. The human Hs.5HT7R is induced into a prolonged inactivated state by exposure to a subset of ligands, termed ‘inactivating antagonists’ [25,26]. These inactivating antagonists are structurally diverse and include the ergot alkaloid bromocriptine, risperidone, methiothepin, lisuride and metergoline [26]. Application of these ligands caused a prolonged inactivation of Hs.5HT7R activity in heterologous expression systems or in assays on endogenous receptors [25,26,47]. Data suggests this aspect of receptor phenomenology may be conserved with Sm.5HTR, the most abundant deorphanized GPCR in adult schistosome worms, when evaluated in receptor level (Fig 6) and whole organism assays (Fig 10). The predominant expression of this specific GPCR in this organism, together with conservation of this receptor property, provides a clearly targetable weakness for anthelmintic development. If transient exposure to an inactivating antagonist inhibits parasite mobility well beyond the pharmacokinetic persistence of the drug within an infected individual, this would be clearly be effective for antiparasitic action and serve to minimize dosing requirements in challenging healthcare environments. Sm.5HTR is also expressed at multiple life cycle stages, and is conserved in other PZQ-sensitive parasites. Further exploration of this property and identification of parasitic-selective ligands that convey this effect are warranted, and such activities will be facilitated by the approaches optimized in this study.
In conclusion, these data demonstrate the optimization and application of a real-time biosensor assay for interrogating flatworm GPCRs in vitro, which is capable of scaling to HTS. Application of this approach to profile Sm.5HTR revealed parasitic-selective ligands and ligand series, as well as conservation of a ligand-evoked inactivation mechanism at the most predominantly expressed S. mansoni 5HTR.
Serotonin (5-HT), 3-Isobutyl-1-methylxanthine (IBMX) and DMSO were purchased from Sigma Aldrich. The GPCR Compound Library was purchased from Selleck Chemicals (Catalog No. L2200) pre-dissolved in DMSO (10mM). 5HT7 ligands DR4485, LY215840, metergoline and 5-Carboxamidotryptamine (5-CT) were purchased from Tocris Bioscience. Methoxy-isoquninoline alkaloids (rotundine, tetrabenazine, berbine, palmatine, tetrandrine and berbamine) were purchased from Sigma Aldrich, while fangchinoline was purchased from AK Scientific.
HEK293 cells (ATCC CRL-1573.3) were cultured in DMEM supplemented with 10% fetal bovine serum (FBS), penicillin (100 units/ml), streptomycin (100 μg/ml), and L-glutamine (290 μg/ml). Cells were transiently transfected (ViaFect, Promega) as per the manufacturer’s protocol at a density of 2x106 cells per T-25 cell-culture flask. Cells were transfected with plasmids encoding GloSensor (Promega) and either Sm.5HTR (Smp_126730, GenBank accession number KF444051.1) or Sm.5HTRL (KX150867), bothGPCRs being codon optimized for mammalian expression, or Hs.5HT7a (GenBank accession number NM_000872.4, R&D Systems) subcloned into pCS2(-). Cell culture reagents were from Invitrogen. Epitope tagged constructs were used to verify expression, and untagged constructs used for all luminescence assays.
HEK293 cells were transfected with Sm.5HTR subcloned into a pCS2(-) mammalian expression vector possessing an NH2-terminal 6xmyc tag and harvested 24 hours post-transfection. Cell pellets were solubilized in 1% NP-40, protein was quantified using Bradford reagent (Pierce). Denatured sample (10 μg) was run on a Mini-PROTEAN TGX Precast Gel (BioRad) at 150V. Semi-dry transfer to PVDF membrane was performed using a Trans-Blot Turbo Mini-PVDF Transfer pack (Bio Rad) at 25V for 30 minutes. The membrane was blocked with 5% nonfat milk in TBST (Tris-buffered saline, 0.1% Tween 20) for 1hr at room temperature, incubated with anti-myc antibody overnight at 4°C (Santa Cruz, 1:500 dilution in 5% milk -TBST) prior to washing in TBST (3x10 minutes) and incubation with secondary antibody (LiCor Goat anti-mouse IRDye 800, 1:5000 dilution in 5% milk -TBST) for 1hr at room temperature. After washing (3x10 minutes in TBST), membranes were visualized on a LiCor Odyssey imaging system.
Sequence for Sm.5HTRL was determined by cloning from an Schistosoma mansoni cDNA library (adult male and female NMRI strain, BEI cat #NR-36056) using high fidelity Advantage HD DNA polymerase (Clontech) and primers described in [20]. PCR products were ligated into the pGEM-T Easy vector system (Promega) prior to DNA sequencing. Additional sequence contained in the Sm.5HTRL isoform was verified by 5’/3’ RACE (SMARTer RACE Kit, Clontech) using total RNA extracted from S. mansoni (Trizol reagent, Ambion). Products were cloned into the pRACE vector (In-Fusion HD Cloning Kit, Clontech) prior to DNA sequencing.
For assays performed on adherent cells, HEK-293 cells were transferred one day post transfection to 96 well, solid white plates (Corning, cat # 3917) coated with 0.01% poly-L-lysine (Sigma Aldrich) at a density of 5 x 104 cells / well in DMEM supplemented with 1% dialyzed FBS (Gibco). After overnight culture (37°C/5% CO2), media was decanted and replaced with 100μL / well HBSS supplemented with 0.1% BSA, 20mM HEPES (pH 7.4) and GloSensor reagent. Plates were allowed to equilibrate at room temperature for two hours prior to performing luminescence assays (GloMax-Multi Detection System plate reader, Promega). Conditions for individual assays were described as per figure legends. The standard assay to detect changes in cAMP utilized the F22 sensor in media supplemented with IBMX (200μM). Ligands were added at a concentration of 20x per well for experiments (i.e. 5μl of drug solution added to 100μl of cells). A dose of 5-HT corresponding to the [EC80] for the relevant receptor used for all antagonist screens. The average standard deviation of the 5-HT Emax in internal, vehicle treated control wells (at least 8 per plate) was 13% for Sm.5HTR and 8% for Hs5HT7R. For the resazurin reduction assay for cellular viability, cells were incubated with resazurin (final concentration, 10μM) and tested ligands with fluorescence measurements (560 nm excitation/590 nm emission) made at 1.5hr intervals.
To test putative irreversible antagonists, assays on suspension cells were performed one day post transfection. Cells in a T-75 flask were trypsinized (0.25% w/v) and transferred to a 14mL tube, centrifuged at 300 RCF, and resuspended in HBSS supplemented with 0.1% BSA, 20mM HEPES. Compounds were added at 10 μM, and after 30 minute incubation at room temperature cells were centrifuged (300 RCF, 5 minutes) and resuspended in fresh media. This wash step was repeated, and cells were resuspended in HBSS supplemented with 0.1% BSA, 20mM HEPES and GloSensor reagent. Cells were gently rotated to prevent aggregation and settling over the course of the two hour equilibration period, after which time they were transferred to 96 well plates at a density of 8 x 104 cells / 100μL per well and assays described for adherent cells.
Biomphalaria glabrata (M-line) snails exposed to Schistosoma mansoni miracidia (Strain PR-1) were obtained from BEI Resources (Cat. number NR-21961) and cercaria were shed following exposure to light (1.5 hours). Cercaria were manually transformed into schistosomula by vortexing (3 x 45 seconds, each separated by 3 minutes on ice) and tails were removed by gradient centrifugation (24ml Percoll, 4ml 10X EMEM, 1.5ml penicillin-streptomycin, 1ml of 1M HEPES in 0.85% NaCl, 9.5ml distilled water) at 500g/15min at 4°C. Supernatant containing tails was discarded, and schistosomules were resuspended in Basch media and incubated (37°C, 5% CO2) overnight before conducting mobility assays. For contractility experiments, somules were incubated in 5-HT free Basch media to resolve a basal contractility rate. To establish a dose response curve for 5-HT, serial dilutions of 5-HT were added to Basch media and somule contractile frequency recorded. In order to assess the effects of antagonists on somule movement, recordings were made of cohorts in 5-HT free Basch media, media supplemented with 5-HT (10μM), and media supplemented with both 5-HT (10μM) and the drugs indicated in Fig 8. Schistosomules were incubated in 24 well plates (~200 schistosomules / 0.5mL media per well) for 30 minutes (37°C/5% CO2) prior to acquiring videos of schistosome movement (1 minute recordings / well) using a Nikon Coolpix 5700 camera affixed to a Nikon Eclipse TS100 microscope (10x objective). The WrmTrck plugin for ImageJ was used to quantify worm mobility [21]. Briefly, the major axis of each schistosomule body length was extracted from the raw output of WrmTrck and an average length was determined for the duration of the recording. Contractions were quantified by determining the number of episodes during which the worm body length deviated from the average by over 20%. S. mansoni protocols were approved by the Iowa State University Institutional Biosafety Committee.
Female Swiss Webster mice exposed to Schistosoma mansoni cerceria (Strain PR-1) at 5–7 weeks old were obtained from BEI Resources (Cat. number NR-34792) and sacrificed 6–8 weeks post-infection. Mature S. mansoni were harvested from the mesenteric vasculature by portal perfusion [48]. Briefly, mice were anesthetized in a CO2 chamber and sacrificed by cervical dislocation. Mice were perfused with sodium citrate (25mM) and adult schistosomes were harvested from the mesenteric veins. Schistosomes were washed in RPMI media containing penicillin (1000 units/mL), streptomycin (1000 μg/mL) and 25mM HEPES, then transferred to RPMI media supplemented with 2mM glutamine and 5% heat inactivated FBS. Worms were incubated overnight at 37°C, 5% CO2 before conducting assays. Recordings of adult schistosome movement were captured using a Zeiss Discovery v20 stereomicroscope and a QiCAM 12-bit cooled color CCD camera. Videos were acquired at a rate of four frames per second for one minute. Recordings of female worms were acquired at 7.6x magnification, 30 mm field of view and recordings of male worms were acquired at a 5.1x magnification, 45 mm field of view. Analysis was performed in ImageJ as described previously [20]. Briefly, image (.tiff) stacks of each recording were imported and processed by converting to binary format so that pixel measurements represent area of the worms’ bodies. Mobility was quantified by measuring the difference in pixels resulting from subtracting the value of one frame (n) from those of the next frame in the sequence (n+1). This difference was expressed as a percentage of the pixels in the initial frame (n), providing a measurement of the worms’ movement over a period of 0.25 seconds. This calculation was performed for each frame in the video, and the results were averaged over the length of the recording. Values reported represent the mean (±) standard error of at least three independent experiments. Animal work was carried out with the oversight and approval of the Laboratory Animal Resources facility at the Iowa State University College of Veterinary Medicine.
|
10.1371/journal.pntd.0006218 | Evaluation of oxfendazole in the treatment of zoonotic Onchocerca lupi infection in dogs | The genus Onchocerca encompasses parasitic nematodes including Onchocerca volvulus, causative agent of river blindness in humans, and the zoonotic Onchocerca lupi infecting dogs and cats. In dogs, O. lupi adult worms cause ocular lesions of various degrees while humans may bear the brunt of zoonotic onchocercosis with patients requiring neurosurgical intervention because of central nervous system localization of nematodes. Though the zoonotic potential of O. lupi has been well recognized from human cases in Europe, the United States and the Middle East, a proper therapy for curing this parasitic infection in dogs is lacking. To evaluate the efficacy of oxfendazole, 11 out of the 21 client-owned dogs (21/123; 17.1%) positive for skin-dwelling O. lupi microfilariae (mfs), were enrolled in the efficacy study and were treated with oxfendazole (50 mg/kg) per OS once a day for 5 (G2) or 10 (G3) consecutive days or were left untreated (G1). The efficacy of oxfendazole in the reduction of O. lupi mfs was evaluated by microfilarial count and by assessing the percentage of mfs reduction and mean microfilaricidal efficacy, whereas the efficacy in the reduction of ocular lesions was evaluated by ultrasound imaging. All dogs where subjected to follow-ups at 30 (D30), 90 (D90) and 180 (D180) days post-treatment. The percentage of reduction of mfs was 78% for G2 and 12.5% for G3 at D180. The mean microfilaricidal efficacy of oxfendazole in the treatment of canine onchocercosis by O. lupi at D30, D90 and D180 was 41%, 81% and 90%, in G2 and 40%, 65% and 70%, in G3, respectively. Retrobulbar lesions did not reduce from D0 to D180 in control group (dogs in G1), whereas all treated dogs (in G2 and G3) had slightly decreased ocular lesions. Percentage of reduction of ocular lesions by ultrasound examination was 50% and 47.5% in G2 and G3 at D180, respectively. Despite the decrease in ocular lesions in all treated dogs (G2 and G3), oxfendazole was ineffective in reducing ocular lesions and skin-dwelling O. lupi mfs in treated dogs (G2 and G3) in a six-month follow-up period. Here we discuss the need for more reliable diagnostic techniques and efficient treatment protocols to better plan future intervention strategies.
| The genus Onchocerca (Spirurida, Onchocercidae) includes Onchocerca volvulus, which is estimated to infect at least 37 million people globally, and zoonotic Onchocerca lupi in carnivores. Infection by O. lupi has been reported in dogs and cats from several European countries and recently also in the U.S. and Canada, causing mainly ocular lesions. In humans, O. lupi displays a marked neurotropism with nematodes embedded in nodules localized in the cervical spine of infant, children and adults. Though the reported severity of infection in humans and the high prevalence detected in dogs are now well-recognized, a proper treatment regime for curing this parasitic infection is lacking, being the surgical removal of the parasitic nodule the therapy of choice in canine patients. Hence, there is an unmet medical need for treatment of this zoonotic disease in both humans and animals. In this study we evaluated the efficacy of oxfendazole under two treatment regimes in the reduction of ocular lesions and skin-dwelling microfilariae of O. lupi in naturally infected dogs.
| The genus Onchocerca (Spirurida, Onchocercidae) encompasses parasitic nematodes mainly associated to ungulates, with the exception of Onchocerca volvulus in humans and Onchocerca lupi in carnivores [1–3]. While O. volvulus is a well-known parasite estimated to infect 37 million people globally (www.cdc.gov/globalhealth/ntd/diseases/oncho_burden.html), infection by O. lupi has been reported from dogs and cats in Hungary, Greece, Germany, Portugal, Romania and Spain [4–11], and also in the U.S. and Canada [12–18]. Adult O. lupi are found within nodules embedded in ocular tissues and annexes [5, 15, 17], and such presentation commonly leads to clinical diagnosis. However, infections may be associated with no clinical signs [19], to severe ocular disease, including blindness [20]. Nonetheless, the limited number of case reports hinders a thorough understanding of the pathogenesis of canine onchocercosis, including cases with O. lupi in the retrobulbar space of the eye, with no overt sign of infection [15]. In the latter case, imaging techniques (i.e. ultrasound scans and computed tomography [21]), or the detection of microfilariae (mfs) in skin snips [14] are the only diagnostic tools available. For instance, an overall prevalence of infection by O. lupi of 8.4% was recorded by mfs counts in skin biopsy sediments in apparently healthy dogs from Greece and Portugal [19].
Humans may bear the brunt of zoonotic onchocercosis by O. lupi with patients requiring neurosurgical intervention because of nematode localization in the cervical spine of an infant [22] and children [16, 23], thus making central the development of treatment strategies of reservoir animals. However, though the zoonotic potential of O. lupi has been well recognized from cases reported in Europe, the United States and Middle East [15, 16, 24], scientific knowledge on the biology, pathogenesis and treatment of this parasitosis is minimal. A proper treatment regimen for curing this parasitic infection is lacking and the surgical removal of the parasitic nodule has been the therapy of choice in canine patients.
Drug-based treatments included various combination and dosages of melarsomine, ivermectin, topical and systemic antibiotics and prednisone [15, 17, 25]. However, proper studies on the long-term outcomes of these therapies have not yet been performed, and there is an urgent need for studies assessing the efficacy of molecules during natural infection with O. lupi in dogs.
Benzimidazole (BDZ) drugs have a broad-spectrum activity and low toxicity, and have been approved, more than 30 years ago, in human and veterinary medicine against several helminth species, including gastrointestinal parasites and lungworm infections in animals. In this class of drugs, oxibendazole and oxfendazole (OXF) have been increasingly tested as anthelmintics used in human medicine, for their potential efficacy also against tissue-dwelling larval helminths [26, 27].
In addition, benzimidazole drugs (flubendazole, mebendazole, OXF, albendazole, fenbendazole) have shown an in vivo macrofilaricidal activity against several filarial species in animal models [27]. In particular, their efficacy was assessed against larval and adult forms of Brugia malayi, Brugia pahangi, Acanthocheilonema viteae and, Litomosoides sigmodontis, in experimentally infected rodents [28–32]. Significant effects on the microfilaremia after treatment are not always correlated with adulticidal efficacy suggesting that subdoses may alter embryogenesis. However, differences in efficacy of benzimidazoles have been related to the parasite species and the route of drug administration. By subcutaneous administration, OXF has shown either no activity [33] or full protection against adults of B. pahangi [34] whereas it exhibited a marked macrofilaricidal activity against L. sigmodontis [33].
In this study we evaluated the efficacy of OXF under two treatment regimens in the reduction of ocular lesions and skin-dwelling mfs of O. lupi in naturally infected dogs.
This study was performed as a negative controlled, blinded and randomised field study in privately owned dogs conducted according to the principles of Good Clinical Practices (VICH GL9 GCP) [35]. The protocol of this study was approved by the Ethical Committee of the Department of Veterinary Medicine of the University of Bari (Prot. Uniba 1/16). All dogs were living in an O. lupi endemic area of Algarve region (southern Portugal [19]), and the study procedures on animals were performed after receiving the owner’s informed consent.
In October 2016, privately owned dogs (n = 123) were sampled via skin snip and positive animals were subjected to ultrasound examination for diagnosing O. lupi infection. All animals lived in the municipalities of Tavira, Faro and Castro Marim.
Skin samples were collected in the afternoon-evening by using a disposable punch over an area of ≈0.4 × 0.5 cm from the interscapular regions of the dogs [19]. Skin biopsies (one per dog at each time point) were soaked in saline solution for 12 h at room temperature and sediments (20 μL) were individually observed under light microscopy. Microfilariae were identified according to morphological keys [19], and their numbers were assessed from each positive animal by a blinded double-check counting of two independent operators.
Microfilariae were isolated and genomic DNA extracted using a commercial kit (DNeasy Blood & Tissue Kit, Qiagen, Germany) in accordance with the manufacturer’s instructions. Samples were molecularly processed for specific amplification and sequencing of the partial cytochrome oxidase subunit 1 (cox1) gene (~689 bp), following procedures described elsewhere [36]. Amplicons obtained from the skin sediments were purified using Ultrafree-DA columns (Amicon, Millipore, USA) and sequenced directly with the Taq DyeDeoxyTerminator Cycle Sequencing Kit (v.2, Applied Biosystems, USA) in an automated sequencer (ABI-PRISM 377, Applied Biosystems).
Sequences were aligned using the Geneious R9 software package (http://www.geneious.com) and compared (BLASTn) with those available in the GeneBank database (http://blast.ncbi.nlm.nih.gov/Blast.cgi).
Following the assessment of mfs counts, each positive animal was checked by an ultrasound examination of both eyes and retrobulbar spaces as follow. Briefly, 1–2 minutes prior to the exam oxibuprocain chloridrate (Anestocil, Laboratório Edol, Portugal) was used as local anaesthetic, placing a few drops on each eye, with animals restrained in sternal recumbence. The eye and the retrobulbar space were scanned with a portable ultrasound Logic Book–GE, equipped with two probes, one linear and one microconvex, with frequency ranges from 6–10 MHz, through transcorneal, transscleral and transpalpebral approaches, along horizontal, vertical and oblique planes to check for lesions associated with the presence of adult parasites.
Naturally infected dogs were enrolled in the study if scored positive for O. lupi mfs at skin sediment observation and further molecular diagnostic confirmation. Nodules or hyper-echogenic lesions caused by adult nematodes were also assessed. Dogs were allocated to study groups in blocks following a random treatment allocation plan on the basis of an inclusion sequence. Each dog, per block, was randomly assigned to one untreated control group (G1) and to treated groups (G2 and G3). Animals were treated with Dolthene (Boheringer Ingelheim, Germany) a commercial oral suspension of OXF for dogs containing 22.65 mg/ml of OXF, 1.5mg/ml of sorbic acid (E200), macrogol, macrogol stearate, sodium carboxymethyl cellulose, silica colloidal anhydrous, citric acid monohydrate (E330), sodium citrate and purified water. A dose of 50 mg/kg per body weight per OS once daily for 5 (G2) or 10 (G3) consecutive days was administrated. No information is available on the in vitro activity of OXF against O. lupi microfilaria or adult parasites. Furthermore in vitro activity of anthelmintic benzimidazoles in general is difficult to be assessed [37, 38]. Therefore, a relevant drug concentration to be achieved in plasma of infected animals cannot derive from in vitro efficacy studies. To maximise the chance of observing a pharmacological effect a high dose of 50mg/kg/day for 5 (G2) and 10 (G3) consecutive days was selected to achieve a plasma disposition of OXF during the whole treatment duration above approximately 1 μg/ml [39]. Efficacy of the treatment was assessed by microfilarial count and presence and size of ocular nodules, at 30 (D30), 90 (D90) and 180 (D180) days post-treatment.
The percentage (%) of reduction [s] of ocular lesions was calculated as follow [s] = [(Cs0 –Cs) / Cs0] x 100, where Cs0 is the baseline ocular size lesion before treatment and Cs was the count at Cs was the count at any time point (s). The percentage (%) of reduction [t] of mfs was calculated as follow [t] = [(Ct0 –Ct) / Ct0] x 100, where Ct0 is the baseline count before treatment and Ct was the count at any time point (t). Moreover, mean microfilaricidal efficacy (%) = [(Ct–T) / Ct] x 100, where Ct is the mean count of mfs of the control group at X time and T is the mean count of mfs of the treated animal groups at X time. The significance of the mfs count reduction and mean microfilaricidal efficacy in treated dogs was analysed by ANOVA, with standard statistical assumption. Statistical analysis was planned and conducted in compliance with current guidelines [40]. Statistical calculations and randomization were performed with SPSS statistical package for Windows, version 13.0, and nQuery+nTerim 3.0 (StatSols), Statistical Solutions Ltd. 2014, Microsoft. A post-hoc power calculation on the mfs counts and ocular lesions at day 0 and at the several days of measurement has been calculated by the software GPower 3.1.9.2, using the module F-test, ANOVA model for repeated measures with between-within factor interaction, setting the power at 80% and significance level at 0.05 and the sample size was evaluated in function of effect size.
In order to assess the pharmacokinetic of OXF, and the metabolites fenbendazole (FBZ) and fenbendazole-sulfone (FBZSO2), heparinised blood samples were collected by cephalic vein puncture prior to the start of treatment from all dogs (G1, G2 and G3 groups) and, once a day, at different time points (i.e., +1, +5, +6, +7 day post treatment (pt) for G2, and +1, +5, +10, +11 and +12 day pt for G3). Samples were collected immediately prior to the daily drug administration.
A 20 μL samples of whole blood sample from each animal was transferred into 96-well polypropylene plate and added with 20 μL of blank dog plasma. After mixing a volume of 400 μL of acetonitrile was added to each well. The plate was than mixed and centrifuged at 2100 g for 20 min and 2 μL of supernatant was directly injected in the LC-MS/MS system. Calibration standards in the range of 1 to 5000 ng/mL and added with 20 μL of blank dog whole blood were included in duplicate at each run.
OXF, FBZ and FBZSO2 were detected and quantified using an Agilent 1100 series HPLC connected to a API4000 QTRAP Mass Spectrophotometer (SCIEX, Applied Biosystems, USA). Chromatographic separation was achieved using a Kinetex C18 analytical column (50*3.0mm, 2.6 mm. Phenomenex, USA) column maintained at 40°C and eluted with a gradient of ammonium acetate (10 mM) and acetonitrile. The run time was 0.6 min. The detection and quantification of the three compounds was performed in the tandem mass spectrometry operated in positive electro-spray ionisation and multiple reaction monitoring mode using the transition range of m/z 316–191.2 for OXF, 300.1–268.2 for FBZ and 332.1–300.3 for FBZSO2. The ion source and gas parameters were: curtain gas 20 psi, ion source gas 45 (GS1) and 40 (GS2), source temperature 450°C and collision gas set to medium. The optimized acquisition parameters for the three analytes were: declustering potential 95 V for OXF and FBZSO2 and 120 V for FBZ); entrance potential 10 V for all analytes; collision energy 30 V, 28 V and 31 V for OXF, FBZ and FBZSO2, respectively and collision cell exit potential 15 V for OXF and f FBZSO2 and 10 V for FBZ.
The performance of the LC-MS/ MS method was tested using a short validation protocol. Linearity of calibration was confirmed at concentrations ranging from 2.5 to 5000 ng/mL (r2 = 0.9975) for OXF, from 1.0 to 1000 ng/mL (r2 = 0.9964) for FBZ and from 1.0 to 2500 ng/mL (r2 = 0.995) for FBZSO2. The mean extraction recovery was not less than 80% for all analytes tested and the accuracy (expressed as %Bias) and precision (expressed as % CV) of the method ranged from -2.9 to 3.4% and 2.5 to 6.7% respectively for OXF, from -1.6 to 2.0% and 2.8 to 6.4% for FBZ and from -10.4 to 6.4%, and 0.1 to 9.0% for FBZSO2. The lower limit of quantification (LLOQ) was 2.5 ng/mL for OXF and 1.0 ng/mL for FBZ and FBZSO2. Results were expressed as mean (± s.e.m.). Differences in drug blood concentration at different sampling times were determined using one-way ANOVA and results were considered statistically significant when p<0.05.
Of the 123 animal sampled, 21 (17.1%) scored positive for O. lupi mfs, out of which 11 were enrolled in the efficacy study and assigned to groups G1 (dogs 1–4), G2 (dogs 5–8) and G3 (dogs 9–11). The mean number of O. lupi microfilariae in skin sediment was homogenous amongst the three groups (p<0.05). The morphological identification of mfs was molecularly confirmed, and nucleotide sequences obtained from mfs DNA (GenBank; accession number: MG677940) displayed 100% identity with those of O. lupi from Portugal (GenBank; accession number: EF521410).
The count number of O. lupi mfs at each study day is reported in Table 1. The percentage of reduction of mfs was 78% for G2 and 12.5% for G3 at D180. Mean microfilaricidal efficacy of OXF in the treatment of canine onchocercosis by O. lupi was 41%, 81% and 90%, respectively at D30, D90 and D180 in G2 compared to G1 and 40%, 65% and 70%, respectively at D30, D90 and D180 in G3 compared to G1. Differences in mean microfilaricidal efficacy in animals in G2 and G3 compared to the control group (G1) were not statistically significant at all time points, except at D90 between G2 and control group (p<0.05).
On D0, eight dogs (i.e. nos. 1, 4, 5, 6, 7, 9, 10, 11) had hyper-echogenic lesions (0.7–2.5 mm wide) in the retrobulbar space overlapping the localization and the dimensions of O. lupi adult nematodes. At the ultrasound examination, retrobulbar lesions did not reduce from D0 to D180 in dogs of the G1, whereas one dog of G2 (no. 7) cleared the ocular lesion and all the other dogs of treatment groups had a slightly decreased size of ocular lesions (Figs 1 and 2). Percentage of reduction of ocular lesions by ultrasound examination was not statistically significant, being of 50% and 47.5% at D180 in G2 and G3, respectively.
All samples collected from untreated dogs were below the limit of detection of the method. The plasma concentrations time curves of OXF and its metabolites following oral administration of 50 mg/kg for 5 or 10 days are shown in Figs 3 and 4. At the zero-time point, prior to the first administration, the mean of OXF plasma levels is above zero, this likely caused by a carryover effect. Standard deviations of measured plasma levels indicate high variability from day 1 to day 5 and day 10, while OXF is administrated to animals. Among others, the heterogeneity of treated animals, in regards of species, weight and age is likely to be the cause for such variability. Over the 5-days period blood mean concentration of OXF varied at trough level 0.49±0.24 μg/mL on day 1 to 0.98±0.74 μg/mL on day 5 and over the 10-days period from 0.23±0.13 μg/mL on day 1 to 0.78±0.17 μg/mL on day 5. By 1 day after the last dosing (day 6 and day 11, respectively) the drug blood concentrations were maintained at similar levels than those recorded during the treatment and they fell to 0.005±0.002 μg/mL and 0.003±0.001 μg/mL 2 days after the last drug administration (day 7 and day 12 respectively). No significant differences among the blood drug concentrations were recorded from the first day treatment to the first day post-treatment (one-way ANOVA). In conclusion, the overall exposure of OXF was sustained in all animals during the whole treatment duration. In both administration protocols, the blood concentration of FBZSO2 followed a similar but lower pattern to OXF. Over the 5-days period, the mean concentration percentage of FBZSO2 compare to OXF varied from 27% on day 1 to 34% on day 6; over the 10-days period, a similar variation is observed: 31% on day 1, to 17% on day 5 and to 31% on day 11. The metabolic biotrasformation of the parent drugs into the reduced metabolites fenbendazole was negligible.
In this study, OXF was ineffective in reducing ocular lesions and skin-dwelling mfs of O. lupi in naturally infected dogs in a six-month follow-up period. Though one treated dog cleared the ocular lesions, OXF did not show any major efficacy in reducing the size of hyper-echogenic areas in the retrobulbar space. Ultrasound examination has several limitations in the detection of the parasites, including lack of sensitivity and specificity, but represents the only option available to detect nematodes with retrobulbar localization [21].
Only at D90, but not D180, OXF showed a significant efficacy in reducing the number of mfs in animals from G2 compared to the control group. Moreover, the increased number of mfs in 6 out of 11 dogs from D30 to D180 together with the lack of significant decrease of size of ocular lesions stands for a lack of efficacy in treatment of canine onchocercosis. In addition, this variation in number of mfs from November (D30) to April (D180) may indicate the existence of a seasonal pattern, which may match with the behaviour of the as-yet-unknown vector species. Up to now, the detection of DNA in wild-caught Simulium tribulatum indicates this simulid species as putative vector in California [41], though further studies are needed to ascertain species of vectors involved in the epidemiology of O. lupi. The seasonal variation in O. lupi mfs would add further information to their circadian rhythm [42], also considering that seasonal pattern has been well studied in the life-cycles of O. volvulus and Dirofilaria immitis [43–46].
The highest percentage of reduction of mfs in dogs treated for 5 days rather than for 10 days (78% vs 12.5%) is unexpected. OXF is a macrofilaricide and the evaluation of its efficacy solely based on mfs counting may be troublesome, also considering that their lifespan in the definitive hosts is unknown. Moreover, the enrolment of naturally infected animals in this trial was challenging and resulted in a limited number of dogs included in the study. Indeed, it would have been desirable a large effect size (large difference between first observations and subsequent follow-ups) like the one observed, albeit variances of the model resulted very small to achieve 80% power, i.e. a higher level of probability that there is an effect if the differences are statistically significant, a greater sample size should be recruited, with a lower effect size that is still clinically interesting.
Nevertheless, benzimidazoles, including OXF, have been described, as tubulin inhibitors, preventing polymerisation of the tubulin subunit α and ß [47]. Though this mechanism of action is compatible with inhibition of the embryogenesis [48, 49], this effect did not seem to occur in the present study. Further explanations for the lack of efficacy of OXF may be the pharmacokinetic differences, though not statistically significant, in dogs from G2 and G3 (mean blood concentration 0.49–0.97 in G2 vs 0.23–0.78 μg/mL in G3). The concentration of OXF, FBZSO2 and FBZ, measured at the trough level is in the same range of previous assessment of the plasma disposition in dogs at the same dosage (i.e. 50mg/kg) after a single administration [39]. The metabolic biotransformation of OXF occurs via oxidation of the sulfoxide group of OXF to form FBZSO2. This is consistent with oral pharmacokinetic studies conducted in cattle, sheep, dog and pig, where such pattern was observed as well over a period of 24 hours [50]. However, this is not case in horses and rats where plasma concentration of OXF is less abundant compared to FBZ and FBZSO2 respectively, over a period of 24 hours [50]. In sheep, an equilibrium between FBZ and OXF has been described in vivo [51]. This was not observed in the current study (Figs 3 and 4) where formation of FBZ is negligible as reported by [39]. Following 5 and 10 consecutive days of drug administration, OXF blood concentration did not vary being almost undetectable after the second day from the last administration (i.e. at day 7 and day 12 in G2 and G3, respectively). This data may be beneficial for future efficacy studies and confirm the shorter blood retention time of OXF in dogs’ plasma (see also [39]).
Despite a sustained exposure of OXF was observed in all dogs in treatment group, the drug concentration in the nodules is unknown. Also, vascularization of O. lupi nodules is not described and it is possible that OXF plasma concentration is not representative of the drug concentration that will eventually reach the adult parasites. Although OXF is a broad spectrum anthelmintic effective for several filarial species [52], it may be not effective against O. lupi.
The percentage of dogs positive (17.1%) for O. lupi mfs represents the highest prevalence of canine onchocercosis detected, as in an epidemiological study performed in dogs from Algarve the 8.3% of the examined animals scored positive [19]. Remarkably, animals in the abovementioned survey and those in the present study were clinically healthy, drawing the attention on the role of dogs as suitable reservoir of O. lupi in endemic area.
The absence of palpable nodules together with the lack of ocular lesions at ultrasound examination in some of the study animals, which harboured mfs in their skins, imply the potential presence of adult nematodes in other anatomical localization. For instance, gravid females of O. lupi were found in the thyroid cartilage of a dog in Portugal, with no involvement of ocular tissues [53]. Additionally, in a retrospective study of cases from dogs in New Mexico (U.S.), the 67% of animals treated with melarsomine, ivermectin and doxycycline had recurrent ocular disease [17], which was in contrast with a similar study in Greece, where no recurrence was noticed after drug administration [5]. This apparent challenge in treating animals from the U.S. has been attributed to the presence of a single haplotype circulating in this country [17]. Again, humans infected by O. lupi from the U.S. displayed more severe diseases (e.g. spinal and orbital localization) but not subconjunctival nodules [16]. A recent phylogenetic analysis indicates that Onchocerca species form a monophyletic group encompassing three clades, one of which composed of Onchocerca gutturosa, O. linealis and Onchocerca ochengi of domestic bovids, O. volvulus of humans and O. lupi [3]. In addition O. ochengi and O. volvulus are sister species, with O. lupi being basal to this clade [3]. In that study based on a single-gene analysis O. lupi showed a large genetic intraspecific variability, suggesting the existence of two clades, one detected only from Portugal and all the others distributed in Europe and in the U.S. and this is consistent with either two- or seven-gene analysis [3, 13]. Nonetheless, epidemiological and clinical studies coupled with the genetic traits of O. lupi should be conducted to elucidate whether haplotypes occurring in different geographical areas could play a role in the disease ecology and treatment efficacy.
Non-surgical treatment strategies may include the use of microfilaricide and anti-symbiont drugs. For instance, among available pharmaceutical options ivermectin showed to be effective against O. volvulus mfs [54, 55].
With the exception of Onchocerca flexuosa, Wolbachia symbionts have been detected in all the species of the Onchocerca genus [56], including O. lupi [12, 57] making this species a potential focus for symbiont-targeted therapy. Among others, these bacteria favour the survival of the mfs and interact with the cell of the immune system modulating responses to inflammation. Hence, treatments aiming at Wolbachia results in sterilization and death of the adult worms and first-to-third larval moulting blockage [58]. Nonetheless, though the clade in which O. lupi clusters have the strongest co-evolutionary pattern with their Wolbachia symbionts [3], few studies included anti-Wolbachia treatments as potential target for therapy [17].
Undoubtedly, a defined treatment protocol for this infection is still lacking and the therapies employed up to now mostly derive from clinical experience on the treatment of heartworm disease in dogs [59, 60]. Infections with O. lupi can inflict important hardship on the health of people, and there is an unmet medical need for treatment of this zoonotic disease in both humans and animals. Future efficacy studies are, therefore, urgently needed and should take into account the difficulties in the detection of adult and larval stages of this zoonotic filarioid.
|
10.1371/journal.pntd.0001236 | C-Terminal Domain Deletion Enhances the Protective Activity of cpa/cpb Loaded Solid Lipid Nanoparticles against Leishmania major in BALB/c Mice | We have demonstrated that vaccination with pDNA encoding cysteine proteinase Type II (CPA) and Type I (CPB) with its unusual C-terminal extension (CTE) can partially protect BALB/c mice against cutaneous leishmanial infection. Unfortunately, this protection is insufficient to completely control infection without booster injection. Furthermore, in developing vaccines for leishmaniasis, it is necessary to consider a proper adjuvant and/or delivery system to promote an antigen specific immune response. Solid lipid nanoparticles have found their way in drug delivery system development against intracellular infections and cancer, but not Leishmania DNA vaccination. Therefore, undefined effect of cationic solid lipid nanoparticles (cSLN) as an adjuvant in enhancing the immune response toward leishmanial antigens led us to refocus our vaccine development projects.
Three pDNAs encoding L. major cysteine proteinase type I and II (with or without CTE) were formulated by cSLN. BALB/c mice were immunized twice by 3-week interval, with cSLN-pcDNA-cpa/b, pcDNA-cpa/b, cSLN-pcDNA-cpa/b-CTE, pcDNA-cpa/b-CTE, cSLN, cSLN-pcDNA and PBS. Mice vaccinated with cSLN-pcDNA-cpa/b-CTE showed significantly higher levels of parasite inhibition related to protection with specific Th1 immune response development, compared to other groups. Parasite inhibition was determined by different techniques currently available in exploration vacciation efficacy, i.e., flowcytometry on footpad and lymph node, footpad caliper based measurements and imaging as well as lymph node microtitration assay. Among these techniques, lymph node flowcytometry was found to be the most rapid, sensitive and easily reproducible method for discrimination between the efficacy of vaccination strategies.
This report demonstrates cSLN's ability to boost immune response magnitude of cpa/cpb-CTE cocktail vaccination against leishmaniasis so that the average parasite inhibition percent could be increased significantly. Hence, cSLNs can be considered as suitable adjuvant and/or delivery systems for designing third generation cocktail vaccines.
| Cutaneous leishmaniasis (CL) is the most common form of leishmaniasis with an annual incidence of approximately 2 million cases and is endemic in 88 countries, including Iran. CL's continued spread, along with rather ineffectual treatments and drug-resistant variants emergence has increased the need for advanced preventive strategies. We studied Type II cysteine proteinase (CPA) and Type I (CPB) with its C-terminal extension (CTE) as cocktail DNA vaccine against murine and canine leishmaniasis. However, adjuvants' success in enhancing immune responses to selected antigens led us to refocus our vaccine development programs. Herein, we discuss cationic solid lipid nanoparticles' (cSLN) ability to improve vaccine-induced protective efficacy against CL and subsequent lesion size and parasite load reduction in BALB/c mice. For this work, we evaluated five different conventional as well as novel parasite detection techniques, i.e., footpad imaging, footpad flowcytometry and lymph node flowcytometry for disease progression assessments. Vaccination with cSLN-cpa/cpb-CTE formulation showed highest parasite inhibition at 3-month post vaccination. Immunized mice showed reduced IL-5 level and significant IFN-ã increase, compared to control groups. We think our study represents a potential future and a major step forward in vaccine development against leishmaniasis.
| Leishmaniasis is one of the most important vector borne infections that can cause a spectrum of diseases, ranging from a clinically silent process to a fatal progressive disease in human. It is a major public health crisis in many countries including Iran resulting in an estimated of 12 million new cases occurrence, each year (World Health Organization website, http://www.who.int/vaccine_research/diseases/soa_parasitic/en/index3.html). This parasitic disease is diagnosed as three clinical forms named cutaneous, mucocutaneous and visceral leishmaniasis. Clinical manifestation of the disease depends on both the species involved and the host. The tissue lesion in cutaneous form, can last for months or years before healing. An interesting feature is that despite the disappearance of the lesion and resistance to reinfection, residual parasites remain in the host, probably for a very long time, if not forever [1]. Current curative therapies for cutaneous leishmaniasis are costly, often poorly tolerated and not always effective. This disease is one the few parasitic diseases likely to be controllable with vaccination [2]. Generally; vaccination is largely protein-based and requires direct administration of dead or attenuated parasite, recombinant proteins, or virus-like particles. For targets resembling intracellular pathogens like Leishmania species, such vaccines generate incomplete immune responses and fail to induce protective affects as usually generate only antibody-mediated (humoral) immune responses and often require periodic booster injections [2], [3]. However, cell-mediated immune responses are required for clearance of such parasite and generation of cytotoxic T-lymphocyte (CTL) cells that kill infected cells. Currently, only attenuated live organism vaccines generate significant cell-mediated immune responses, but these are associated with certain safety concerns and can be difficult to manufacture consistently [2], [3]. DNA vaccination offers an attractive alternative to traditional non replicating vaccine strategies. Intracellular production of antigens from delivered DNAs can result in both humoral and cellular immune responses. DNA-based vaccines also offer practical advantages as well, mostly because of the capability of developing countries to cheaply and rapidly produce pDNA from bacteria. Furthermore, it is possible to formulate several antigens from different stages of the parasite life cycle or different subspecies as one shot of vaccine [4]. However, despite decades of research, safe and efficient delivery of pDNA to initiate proper immune responses remains one of the major drawbacks in bringing DNA vaccination into clinical trials. Synthetic particle carrier systems are known as one of the important tools for improvement of current performed DNA vaccines. In such approaches, pDNA is encapsulated into, or complexed via electrostatic interaction with a synthetic carrier, resulting in fabrication of particles with a size ranging from nanometers (nm) to few micrometers. The pDNA release from these particles has proven to be a very effective delivery strategy as the passive targeting to antigen presenting cells (APCs) by size exclusion mechanism, protection from nuclease degradation, cellular uptake enhancement, antigen depot formation at the injection site and controlling the release rates of pDNA that might be important for timing immune responses, are all likely to be occurred [5].
Lipid-based delivery systems represent one of the most advanced drug delivery technologies to date. Different lipid-based adjuvants are introduced and evaluated for pDNA formulation, e.g. Solid lipid nanoparticles (SLN) [based on class I of lipids], liposomes, Transfersomes, niosomes and virosomes [based on class II of lipids] and Micelles, emulsions [based on class III of lipids] [6]. In general, the common ground of these systems for transfection is the need for cationic lipids such as 1, 2-dioleoyl-3-trimethylammonium (DOTAP) to facilitate pDNA binding. A neutral helper lipid such as L-alpha-dioleoyl phosphatidylethanolamine (DOPE) or cholesterol is also required to increase the transfection properties of the pDNA/lipid complexes. Several authors have administered liposomes, niosomes and other lipid based systems for pDNA delivery [6]. But despite offering a number of technological advantages over other existing transfection reagents, SLN utility for DNA vaccination has not been conveniently investigated in vivo. As SLN can be manufactured in large scale and under favorable technological parameters without the need for organic solvents and have an acceptable stability that facilitates their manipulation for different processes such as lyophilization and steam sterilization [7], , they may become a potentially valuable addition and promising alternative to the well-established dossier of non-viral transfection agents leading by cationic liposomes.
Leishmania express large quantities of cysteine proteinases (CP) which are members of the papain superfamily [9]. In Leishmania major (L. major), two most important CPs have been described. CPA is a type II cysteine proteinase which is expressed at higher level in amastigote stage and stationary phase promastigote. CPB is a type I cysteine proteinase which present maximally at the amastigote developmental stage and is encoded with an unusual C-terminal extension [10].The presence of this highly variable CTE differentiates Leishmanial rCPB from the other CPs in the papain superfamily [10], [11], [12]. CTE can be glycosylated and partially removed by proteolytic cleavage during processing of the enzyme to its mature form [11]. Hence, CTE fragment is not crucial for enzyme activity and intracellular trafficking, although it is highly immunogenic and responsible for immune evasion and play a role in the diversion of the host immune response [11], [12]. We have reported that antigenic rCTE of L. infantum elicitated a predominant IgG2 response in asymptomatic dogs and in vitro proliferation of PBMCs. Immunization with CTE also displayed both type 1 and 2 immune signatures in experimental murine model of L. infantum infection and therefore is not protective as a vaccine candidate [13].
Furthermore, we had demonstrated that the cpa/cpb cocktail is more protective against cutaneous leishmanial infection than the separate forms [14]. Therefore, there was still a need to study the effect of CTE deletion in this cocktail vaccine against L. major.
Despite the proven antigenicity and immugenicity of these DNA vaccine candidates, the largest drawback of this kind of vaccination is the obscurity in intracellular delivery of pDNA that causes low levels of gene expression (transfection) which in turn limits the resulting immune responses [15]. Therefore, pDNA must use with a proper and safe formulation in order to coordinate innate and adaptive immune responses and generate strong immunity.
Recently, several immunoadjuvants like BCG, G-CSF and CpG-ODN and also various delivery systems like PLGA microspheres and liposomes have been used to potentiate the immune responses against Leishmania antigens [16].
Herein, we used cationic solid-lipid nanoparticle (cSLN) as a non viral transfection agent for delivery of the cocktail DNA vaccine. In our previous studies, cSLN formulation as a delivery system have revealed comparable efficiency/cytotoxicity ratio to linear PEI-25 kDa-pDNAs polyplexes, protected cpa, cpb-CTE and cpb genes from extracellular enzymatic degradation and also exhibited considerable low cytotoxicity [17]. In this study, these characterized formulations of cocktail vaccine candidates were evaluated for their immune induction potential in BALB/c mice as sucesptible animal model.
All solutions were prepared using MilliQ™ ultrapure (Milli-Q-System, Millipore, Molsheim, France) and apyrogenic water to avoid surface-active impurities. Cetyl palmitate, tween 80 and cholesterol were purchased from Merck (Darmstadt, Germany). N-[1-(2,3-Dioleoyloxy) propyl]-N,N,N trimethylammonium chloride (DOTAP), Sodium dodecyl sulfate (SDS) were purchased from Sigma–Aldrich (Deisenhofen, Germany) The materials applied for PCR, enzymatic digestion and agarose gel electrophoresis were acquired from Roche Applied Sciences (Mannheim, Germany). Cell culture reagents including Fetal Calf Sera (FCS), M199 medium, HEPES, L-glutamine, adenosine, hemin, gentamicin, and RPMI were sourced from GIBCO (Gibco, Life Technologies GmbH, Karlsruhe, Germany) and Sigma (Germany) respectively.
The cSLN suspension was produced by a technique previously described by Doroud et al. [17]. Briefly, desired amount of DOTAP (0.4% w/v) was dissolved in hot aqueous phase which was then added to the melted cetyl palmitate and cholesterol (5.1% w/v) phase containing tween 80 as a nonionic surfactant at 3.2:1 molar ratio. Emulsification was carried out by stirring the mixture at 2000 rpm by a mechanical stirrer (IKA®, Germany) for 10 min at 90°C. Samples were then homogenized using a high shear homogenizer (IKA®, Germany) at 18,000 g for 15 min. cSLN dispersion was obtained by direct cooling of hot O/W microemulsion on an ice-bath while stirring at 1000 rpm. cSLNs were washed by centrifugation (6000 g, 10 min, three times) using 100 kDa Amicon® Ultra centrifugal filters (Millipore, Schwalbach/Ts, Germany) to purify the suspension from the excess amounts of surfactant. Endotoxin concentration in the cSLN formulation was determined by limulus amoebocyte lysate (LAL) assay (LAL Kit, Charles Riever Endosafe, T2092 CTK7, and USA).The physicochemical stability of the formulations were evaluated at 4±1°C, 25±1°C at dark for 1 month at regular time intervals via observation of any changes in suspension clarity, particle size and zeta potential assessments.
pGEM-cpa, pGEM-cpb and pGEM-cpb-CTE were available from our previous studies [17] and each of antigenic fragments were subcloned into pcDNA 3.1(−) vector (Invitrogen). Plasmid DNAs were transformed into the DH5α E. coli strain and purified by alkaline lysis method (QIAGEN Endofree Plasmid Giga Kit) and then confirmed by PCR and digestion (data not shown). In all three constructs (pcDNA -cpa, pcDNA -cpb, and pcDNA –cpb-CTE), the cpa, cpb, and cpb-CTE open reading frames were under control of the CMV promoter, inserted downstream of a Kozak consensus sequence and in frame with an initiation codon. The total concentration and purity of pDNAs were determined by NanoDrop® ND-1000 spectrophotometer (Labtech, UK).
Construct corresponding to pQE-cpa and pQE-cpb were produced in fusion form with an N-terminal histidine (6XHis-tag) for expression and purification of rCPA and rCPB, as previously described [14]. The cpb-CTE gene was subcloned into the cloning site of the bacterial expression vector pET-23a expression plasmid, downstream of the T7 promoter. The E. coli strain BL21 (DE3) was transformed with pET-cpb-CTE and grown at 37°C in 100 ml LB medium supplemented with 100 µg/ml ampicillin and 25 µg/ml chloramphenicol.
The culture was induced with 1 mM IPTG at an OD600 of 0.8 and grown for a further 4.5 h at 37°C. Cells were centrifuged at 8000 rpm for 20 min. Bacterial pellets were dissolved in lysis buffer [50 mM Tris–HCl (pH = 8), 100 mM NaCl and 1 mM EDTA] and frozen overnight at −20°C. After centrifugation (10,000 g, 15 min at 4°C) pellets were washed extensively with washing buffer [20 mM Tris–HCl (pH 8), 20 mM NaCl and 1 mM EDTA]. Inclusion Bodies (IB) were purified by imidazole-SDS-Zn reverse staining method. The purified recombinant protein was concentrated by ultrafiltration using Amicon Filter (MWCO: 10 kDa) and dialysed against PBS. Protein concentration was determined with BCA assay kit (Pierce, Rockford, USA). Purified recombinant proteins were analyzed by SDS-PAGE and Coomassie blue staining to assess the integrity and purity of proteins. These proteins were recognized by previously prepared rabbit anti-CPB antisera using western blot technique [12].
cSLN–pDNA complexes were prepared by adding volumes corresponded to 650 µg of each purified pDNA (pcDNA-cpa, pcDNA-cpb, pcDNA-cpb-CTE) to cSLN suspension at a DOTAP:pDNA ratio of 6∶1 (w/w) and 60 min incubation at room temperature separately, as described before [17]. The final formulations were named as Spa, Spb and Spb-CTE respectively. Complete condensation of pDNAs, complexation with cSLN and the ability of the formulation to protect pDNAs from DNase I digestion were analyzed as previously demonstrated by agarose gel electrophoresis [17].
Statistics were performed using Graph-Pad Prism 5.0 for Windows (Graphpad Software Inc 2007, San Diego, Calif., USA). All the data analyzed with one way ANOVA (Multiple comparisons Tukey post test) when required, with the exception of size and zeta potential measurements, which were analyzed with a Student's t-test. A p-value of ≤0.05 was considered as significant difference between the groups. “n” represents number of mice per group or samples per assay.
cSLN were produced by using the modified microemulsion and high shear homogenization method, cetyl palmitate and cholesterol as matrix lipid, DOTAP as charge carrier and Tween 80 as surfactant. Obtained nanoparticles were approximately 257±23 nm in size and positively charged with a zeta potential of +52±8 mV in milli Q water and size distribution of 0.34±0.08. This suspension was stable for 30 days (p<0.05). The SLN-pcDNA stable complexes (Spa, Spb and Spb-CTE) were prepared by pDNA adsorption on the surface of cSLNs via direct complexation with pcDNA-cpa, pcDNA-cpb and pcDNA- cpb-CTE, respectively. These formulations were also characterized according to their size, zeta potential and poly dispersity index (PDI, Table 1). The results indicated that Spa, Spb and Spb-CTE cationic formulations had an average size of 244±12, 250±15 and 237±12 nm, respectively. There was no significant difference in the size of different preparations (p>0.05). Spa, Spb and Spb-CTE had a mono disperse formulation as the PDI value was about 0.2 for all of them. The observed zeta potential revealed all the formulations were cationic (ζ potential = 22 to 27 mV) that is suitable for interaction with the negatively charged cell surface and the cell entry.
The agarose gel electrophoresis analysis was used to test Spa, Spb and Spb-CTE formulations for their ability to condense pDNA through electrostatic interactions after preparation (data not shown). Gel retardation assay for SLN-pDNAs confirmed complete complexation between pDNA and cSLN at a DOTAP:pDNA ratio of 6∶1. Cationic SLN were able to protect pDNA against DNase I digestion as previously reported [17]. Spa, Spb and Spb-CTE formulations were stable at refrigerated temperature (4±1°C) over one month storage. Endotoxin concentration in the cSLN formulations was 0.215 EU/50 ug.In our previous in vitro studies on COS-7 cells, we demonstrated a very low degree of toxicity of both cSLN and cSLN–pDNA complexes. Furthermore, flow cytometry analysis confirmed SLN-pDNA complexes were able to promote transfection of COS-7 cells at least for 72 hrs after treatment of these cells without a significant reduction in cell viability and in vitro CPs expression capacity [17].
The efficacy of cocktail pDNA vaccines containing pDNAs encoding CPA/CPB, CPA/CPB-CTE, and the same cocktail pDNA vaccines formulated with cSLN (Spa/b and Spa/b-CTE) were evaluated by their capability to induce protection against Leishmania infection in the BALB/c mice model. It is worth to mentin that, no apparent sign of local intolerance such as redness, swelling, bruising, pain observed at the site of injection 1 and 24 hr after vaccine administration. Parasite inhibition in different groups was assessed on both FP and LN of animals. For this purpose, dynamic measurement with a metric caliper, parasite load determination by flowcytometry and imaging techniques were done on the infected mice FP. In parallel, the parasite load was assessed in the LN via microtitration and flowcytometry methods.
As shown in Figure 1, FP swelling of seven different groups of mice n = 10)was measured after a challenge inoculation with EGFP-transfected stationary phase promastigotes of L. major. All the mice which have received the empty cSLNs (G5), pDNA vector without any insert formulated with cSLN (G6) or PBS (G7); showed significant lesions by 7 weeks post-challenge. There was a significant (p<0.05) difference between animals received formulation Spa/b and Spa/b-CTE (G1 and 3) and animals immunized with non-formulated cocktail vaccines (G2 and 4) at 9th week post challenge. The latter groups did not show any significant difference with the control groups. There was also a significant (p<0.05 difference between group 1 and 3, confirming that the higher level of protection observed in group 3. However, this difference between the cocktail vaccine compositions was not significant in group 2 and 4 and this difference could not be simply described due to the presence of CTE in the texture of the Spa/b cocktail vaccine. It seems that utilizing cSLNs for the formulation of pDNAs could potentiate the manifestation of this difference between the vaccine compositions. Vaccination with cSLN-pDNA encoding cpa, cpb (G 1) or cSLN-pDNA encoding cpa, cpb-CTE (G3) delayed FP swelling when compared to immunization with cSLN and PBS (G5 and 7), at this time point. For a certain time, immunization with the pcDNA-cpa/b and pcDNA-cpa/b-CTE cocktail vaccines had a significant effect in delaying FP swelling. However, this effect was not long lasting and 9 weeks after challenge, there was a significant difference (p<0.05) in protective effect between groups 3 and all of the other groups including group 1. Furthermore, the mean FP lesion size in group 1, was still significantly (p<0.05) lower than that in the control groups. Totally, vaccination with a cocktail of cpa, cpb-CTE formulated with cSLNs (G3) resulted in control of the infection progress compare to the control groups as all of the animals in this group developed significantly smaller FP lesions (2.02±0.7 mm) at 9th week after challenge.
To assess the reliability of this conventional and time consuming experiment at 9th week post challenge, we also detected EGFP-expressing L. major in the FPs of mice using fluorescence imaging system that is supposed to give a precise two-dimensional image from the extent of infection, independent of the inflammatory responses at the FP injection site. This experiment enabled us to clearly detect the parasite load in the infected FPs (Fig. 2). As shown in this Figure, the GFP fluorescence in the control group of animals was not localized to the site of the inoculation and the parasites spread to the whole FP. We also observed that the increasing thickness of the infected FPs was not correlated with the intensity enhancement of the detected GFPs in the tested animals (Fig. 2). The sum green intensity (pixel) from the imaging studies were higher in infected FP of the control groups. Furthermore, only group 3 of the tested animals revealed significantly (p<0.05, n = 4) lower GFP intensity compare to the other groups (Fig. 2).
EGFP expression in L. major amastigotes resident in the FPs was also monitored by flow cytometry technique. High expression levels of EGFP were observed in PBS treated group (G 7, Fig. 3). Fluorescence activated cell sorting (FACS) analysis indicated a clear quantitative separation between parasite transfected and normal cells. The frequency of EGFP positive cells of the FPs determined by FACS analysis using the appropriate gating are shown in Fig. 3. Group 3 of the vaccinated animals had the lowest infection rate (12.77%±0.62) compare to the other groups. The percentages of GFP-expressing parasites were significantly lower in the FP cells of the group 1 and 3 of the vaccinated animals.
The infection progression was also followed by determining the total parasite load in the LNs of challenged mice at 9 weeks post-infection by the microtitration method (Fig. 4A) and flowcytometry (Fig. 4B). According to microtitration method and compare to control groups, parasite load in the LNs of vaccinated groups with Spa/b-CTE (G3) decreased significantly (p<0.05, n = 3) (Fig. 4A). The expression of EGFP in the LNs was readily evident from the intense green fluorescence of the parasites (Fig. 4B). The parasites reside inside the LN cells were quantified by monitoring EGFP expression via flowcytometry analysis. As shown in Fig. 4B, Group 1 and 3 both expressed significantly (p<0.05, n = 3) lower percent of EGFP that is correlated with the least amastigote parasite existence in the lymphatic cells. On the other hand, parasite burden in the group 3 was significantly (p<0.05) lower than group 1.
It is noteworthy that, the rate of infection in all groups was in concordance with the delay in the appearance of lesions, the thickening peaked at 9 weeks post-infection and the parasite load in the LNs and FPs determined by microtitration, flow cytometry and imaging methods. Mice vaccinated with formulated cocktail plasmids i.e. Spa/b and Spa/b-CTE (G1 and G3) showed a significant decrease in the FP tissues parasite load as well as LN, compared to the control groups. The ability of the vaccines to inhibit the infection progress and parasite replication [parasite inhibition (PI) %] was predicted according to the assessment of the FPs and LNs parasite load via imaging and flow cytometric method in terms of the decrease in intensity of green fluorescence observed in vaccinated animals, as well as parasite burden of the FPs and LNs by flowctometry and microtitration method. As shown in Table 2, the given rates were calculated by comparing the precentage of fluorescence intensity or parasite burden in FPs and LNs of vaccinated mice to the lowest and highest intensity of fluorescence in the non vaccinated groups. Average of PI% rates were more expressive, when FP imaging and LN flow cytometric techniques were used. The average PI% in G3 was significantly higher than other vaccinated groups when determined by different methods, executed on FPs and LNs. The mean average of parasite reduction in this group was 87.11% (IC95%, 85.9–88.33) by FP imaging, 65.64% (IC95%, 61.29–69.98) by FP flow cytometry, 41.43% (IC95%, 35.05–47.81) by LN microtitration and 86.77% (IC95%, 87.78–85.75) by LN flow cytometry (Table 2). The profile of the alteration in the PI% results in all of the vaccinated groups were correlated with the results of the LN parasite burden via microtitration assay and footpad swelling. Neither imaging nor flow cytometry assays could discriminate between the parasite inhibition results of G1 and G3 of the vaccinated animals, when manipulating the animals' footpads (Table 2).
In order to compare the induced immune responses by different DNA cocktail vaccination strategies and explore the protective effects against L. major challenge in BALB/c mice, IFN- γ and IL-5 production were assessed after in vitro stimulation of LN cells with both Leishmania soluble antigen (SLA) and recombinant CPs, pre and post challenge. According to IFN-γ production, pooled cells from three mice of groups 1, 2, 3 produced significantly (p<0.05) higher levels compared to control groups before infection in response to the rCPs (Fig 5A). Although group 3 produced higher amounts of IFN-γ (522.98±11.99 pg/ml), but was not significantly higher than the other vaccinated groups. Low IFN-γ production was detected in supernatants of LN cells of all three control groups in response to rCPs, before challenge. At 9th weeks post challenge, the rprotein specific IFN-γ production level increased only in group 1 and 3 of the animal vaccinated with Spa/b and Spa/b-CTE formulations, respectively. Although, the difference in IFN-γ production level was only significant (p<0.05), in group 3 (773.29±16.78 pg/ml) of animals compared to control groups (Fig. 5B). No IL-5 was detectable in the supernatant of cells from all groups after stimulation with rCPs before challenge (data not shown). In contrast, significant levels of IL-5 were detected in the supernatant of cells from group 5, 6 and 7, in response to rCPs SLA at 9th week after infection, compared to all the vaccinated groups of animals (G1, 2, 3 and 4, Fig. 5C). There were no significatnt difference in Con A-induced cytokine production, among the tested groups.
Further analysis of the induced cytokines profile by means of IFN-γ/IL-5 ratio revealed that only formulated cocktail DNA vaccines (Spa/b and Spa/b-CTE) could clearly induced strong Th1 immune responses (Fig. 6). As shown in this Figure, this ratio was significantly (p<0.05) higher in group 3 of vaccinated mice compared to the other groups suggesting a higher level of protective immunity.
To compare IgG isotypes in protected and non protected vaccinated groups, sera were collected before and 9 weeks after challenge and assessed for IgG1 and IgG2a. To determine the antibody specificity, all sera of vaccinated and control mice were pooled and assayed by ELISA method before (Fig. 7A, B and C) and after (Fig. 7D, E and F) challenge with L. major promastigotes. As it is shown in these Figures, the group of mice that developed an effective protective response (e.g., group 3) had substantially higher levels of CPA-specific IgG2a antibody compared to unprotected mice, both before (n = 13) and after (n = 10) challenge (Fig. 7A, C). This is consistent with the results obtained above that SLN formulated cocktail DNA vaccine encoding cpa and cpb-CTE (Spa/b-CTE) preferentially induced a Th1 response. According to the Fig. 7D, higher levels of CPB-specific IgG2a was produced in G1 and 2 of the vaccinated mice. This might be described by the immunogenic nature of CTE fragment. The productions of IgG1 in the control groups were significantly higher than IgG2a after challenge, when stimulated by SLA (Fig. 7E).The ratio of IgG2a/IgG1 in sera of mice immunized with Spa/b-CTE formulation was higher than the other groups when titrated against rCPA (Fig. 8). This ratio was also higher in sera of mice immunized with the pcDNA-cpa/bCTE formulation than all the other vaccinated groups.
Challenged with GFP expressing L. major promastigotes induced high level of IgG1 antibody titers against SLA in all groups (Fig. 7E). The sera of mice immunized with Spa/b-CTE formulation showed higher levels of specific IgG2a antibody compared to the IgG1, when titrated against SLA and higher IgG2a/IgG1 ratio was only observed in this group of vaccinated animals.
Although vaccination in the endemic populations is the most cost-effective tool to diminish the burden of Leishmaniasis, an effective vaccine to control this disease is not commercially available yet [2]. It is unlikely that an effective anti-Leishmania vaccine based on the use of a single antigen will be achieved. This might be due to the complex and biphasic life cycle of Leishmania parasite. Therefore, a rational approach toward developing an effective cocktail vaccine should be the use of extracellular and intracellular parasite antigens resulting in a valuable cumulative immune response [3], [14]. In this regard, ease of combining different pDNA vaccine candidates has made genetic vaccination an attractive platform for vaccination strategies [3], .
However in most cases, even multivalent or cocktail DNA vaccines have failed to achieve the required level of protection possibly due to the lack of an appropriate delivery system and/or adjuvant [3], [16]. Therefore, there is still an urgent need for development of new, safe and improved adjuvant and/or delivery systems to enhance the immunogenicity of the available vaccine candidates. There are versatile conflicting reports about DNA vaccination effectiveness against leishmaniasis. Most of DNA vaccine candidates have been tested as single vaccine regimens, but there are also some reports about using combination of genes [1], [3]. Up to date despite of the heterogeneity of vaccination protocols, mean average of parasite load reduction was determined to be 59.24% (IC95%, 47.75–70.73) [3]. This might be due to the reason that adjuvants and delivery systems were rarely added to the formulations containing candidates of the mentioned third-generation vaccines [3]. Approved adjuvants for human vaccines are poor inducers of antigen-specific Th1 responses that are necessary for an intra cellular parasite like Leishmania [21]. Therefore, several strategies, including live vectors; saponins; Freund's and montanide ISA 720 water-in-oil emulsions; oil-in-water emulsions (MF59); dendritic cells (DC) and liposomes have been utilized in different studies and trials to deliver antigens and redirect the immune responses towards desired Th1 pathway [22]. However, most of these failed to provide both long-term immunity and safety for human vaccines.
Therefore, it is still crucial to develop a potentiated delivery system with Th1 stimulating, safe and cost-effective properties for such a promising vaccination technology. Nanoscale vehicles are able to boost the quality and magnitude of an immune response in a predictable, designable trend that can be applied for wide-spread use of genetic vaccination, for developing vaccines for diseases such as cutaneous leishmaniasis, which is currently managed only through relatively ineffectual therapeutic regimens. Nanoparticles as vaccine delivery systems, promote bioactive vaccine candidate protection against extracellular degradation and modulate cellular and humoral immune responses via targeting antigens to APCs such as DCs and therefore would be potentially useful as effective vaccine adjuvants [15]. Cationic lipid-based systems could be formulated as emulsions, liposomes or cationic solid lipid nanoparticles (cSLN). These Lipid-based delivery systems are able to protect the nucleic acid payload and significantly reduce its degradation and extend its activity, improve the pDNA pharmacokinetic characteristics and thus induce more potent immune responses due to a depot effect in which persistence of pDNA at the site of delivery allows uptake by local immune cells, enhance intracellular uptake and delivery to target APCs [15]. Amongst these, SLNs have offered a number of technicopractical advantages including proper storage stability, easy production procedure, steam sterilization and lyophilization possibility and acceptable safety profile [7], [8]. In regards to vaccination studies, significant enhancement was reported when using cSLN-pDNA formulation for cholera toxin and lipid A delivery so that well tolerated particles with sizes greater than 100 nm exhibited higher adjuvant activity enhancing both T-helper types of immune responses compared to the FIA [8]. Hence, in this study we considered SLN for formulating cpa and cpb or cpb-CTE genes. In our previous study, these genes as DNA prime boost vaccination regimen have induced partial protective responses in susceptible BALB/c mice [14]. Heterologous prim-boost immunization approaches have complex logistics and high costs associated with manufacturing a second vaccine platform. Therefore, utilizing a suitable delivery system might be an important strategy to eliminate the need for boosting with the recombinant proteins and enhance third generation vaccine's potency by preventing rapid elimination of the administered pDNA from the circulation. In this regard, cSLN formulations were prepared and characterized regarding their size, zeta potential, nuclease protection, in vitro transfection efficiency, and cell viability, in our earlier report [17].
To assess the cSLN utility prospect as a leishmanial vaccine delivery system, we exploited GFP-transfected Leishmania for generating experimental cutaneous leishmaniasis in BALB/c mice. This model was reported as a novel dynamic immunopathogenic tool allowing visualization and correlation of fluorescence intensity with parasite burden [23], [24]. Some concerns might be raised according to an anti-GFP immune response which could be induced against expressed GFP by the parasites. Several studies indicated that no immune responses have been detected in animals immunized with recombinant EGFP. In other words, recombinant EGFP is not able to stimulate APCs, nor do it induce a significant T-cell response or anti-EGFP antibody production [25]. In our investigations for precise judgment about the feasibility of this delivery system, we looked for a more rapid, sensitive and easily reproducible method to predict average parasite inhibition (PI) in the FP and LN of the vaccinated animals. Therefore, different techniques were used in this regards and the outcomes were correlated to the conventional standard caliper-based method and microtitration parasite burden as well as cytokine and antibody responses to choose the most sensitive, precise and less time-consuming technique for following Leishmania infection, in mice model. As illustrated in Table 2, to evaluate the protection rate; FP swelling (Fig. 1), imaging (Fig. 2) and flowcytometric analysis on the FP (Fig. 3) and LN (Fig. 4B) were assessed in immunized mice and the results were compared to each other and the control groups. The LN parasite burden was also determined by microtitration conventional method (Fig. 4A), at the same time point. The results demonstrated that the size of lesion in mice immunized with Spa/b and Spa/b-CTE at week 9 post challenge were significantly (p<0.05) smaller compared to control groups. Interestingly, the mice immunized with cpa/cpb and cpa/cpb-CTE had also revealed significant difference in FP swelling compared with the group of mice immunized with the formulated cocktail of pDNA-cps (p<0.05). However, there was no significant difference between the groups of mice received non-formulated cocktail vaccines. The same results were obtained through LN analysis by microtitration and flowcytometry. The mentioned methods disclosed the significant PI discrepancy between Spa/b and Spa/b-CTE formulations, while FP analysis couldn't discriminate the effective vaccination strategy between these formulations. However, although FP caliper and LN microtitration based procedures were accurate, precise and capable of differentiation between effective formulations in terms of tissue prasitism, they were time-consuming, rather difficult and not absolutely reproducible as the risk of operator errors are more probable in these experiments. Therefore according to the presented data, it seems that parasite inhibition can be directly estimated by flowcytometry analysis performed on LN cells; as a more rapid, sensitive, and easily reproducible method for screening anti Leishmania vaccine candidates and delivery systems.
The number of viable L. major was quantitated in the FPs and LNs of vaccinated groups of mice after challenge and compared to the control groups, and used as an indication of the protection rate or average parasite inhibition percent (PI%). The significantly (p<0.05) higher mean PI% was seen in group 3 (73.36±15.94%) which was immunized with Spa/b-CTE formulation. This result further revealed the adjuvant effect of the cSLN for potentiating the immunogenicity of this genetic vaccination strategy (Fig. 1, 2, 3, 4A and B) and (Table 2). It seems that a/b-CTE cocktail vaccine induced lower levels of the protection and needs a suitable delivery system to maintain and enhance its immunoprotective activity (44.38±12.59%). This obtained protection rate is in accordance with the percents of reduction of parasite load, reported to obtain with DNA vaccination, without a booster injection (59.24%) [3]. However, despite using cSLN, Spa/b formulation couldn't provoke the same PI%. as did Spa/b-CTE (58.31±17.61%). This difference could be better described by the antigenic nature of the pDNAs used in this experiment and further confirms our previous results that CTE is highly immunogenic but not protective and more favorable to direct the immune system responces towards Th2 type. Based on the presented data, cSLN formulation conferred immunoprotecting activity to a/b-CTE genes which were non-immunoprotecting in their free form, and possibly enhance the immunostimulatory activity of these genes, by effectively inducing TLR-9 mobilization in the endosomal compartment. According to the data presented in Table 2, despite utilizing highly sensitive methods to determine parasite burden, there was no significant difference between cpa/cpb and cpa/cpb-CTE cocktail DNA vaccines (G2 vs G4; Table 2). On the other hand, the discrepancy between these cocktails was significant when they were formulated with cSLN and the parasite burden was determined by LN manipulation (G1 vs G3, Table 2). This data further supports the importance of the lymph nodes as one of the most relevant tissues involved in the parasite-host interface during the stages of L. major infection as the cellular and humoral immune responses in the LN are able to better describe the major immunological changes due to parasite persistence during infection. Therefore, LN may reflect the profile of the host's immune response and the parasite burden intensity throughout L. major infection via both conventional (microtitration) and novel (flow cytometry) techniques. PI% were singnificantly different between the groups received formulated and non formulated cocktail vaccines (G1 vs G2 and G3 vs G4). These results emphasis induction of the immune responses by using delivery systems in such an extent that even the effect CTE deletion could be detectable in the disease progression.
To further evaluate the precision of the obtained data, the cytokines produced by antigen-specific T-cells, were evaluated to determine the profile of an elicited antibody response. IL-5 is associated with high levels of IgG1, whereas production of IgG2a is dependent on IFN-γ. Our results demonstrated that the immune response elicited by Spa/b and Spa/b-CTE formulations was dominantly Th1 response denoted by the higher ratio of IFN-γ/IL-5 secretion after stimulation with SLA and rCPs. However this ratio was significantly higher in Spa/b-CTE vaccinated animals. This ratio is approximately 3-fold higher when SLN has been used as a delivery system. T-cell immunogenicity of CPs had been shown in previous studies, where immunizations with L. mexicana recombinant CP resulted in the development of a potentially protective Th1 cell line, and that recombinant CPB from L. major efficiently induced CD8+ T-cells [9], [26]. Therefore, SLN proceeded as a Th1 stimulator adjuvant and increased the cocktail vaccine efficiency in elicitation of protective responses.
IgG1 and IgG2a antibody titers were also used as an indicator of Th2 and Th1 immune responses, respectively. The significantly (p<0.05) highest IgG2a was seen in the sera of group 3 of mice immunized with Spa/b-CTE before challenge against recombinant CPA compared to other groups. This might be an indication that CTE-domain deletion in Spa/b-CTE formulation redirected the immune responses toward increasing IgG2a production. However, using cSLN formulation for this cocktail vaccine consequently induced a more potent antibody response compared with free pDNAs (Fig. 7A, C and G3, G4). At week 9th after challenge, the IgG2a/IgG1 ratio in mice vaccinated with Spa/b-CTE formulation (G3) was correlated with a Th1 response and further confirmed that could induce a potent Th1 type of immune response and protection against leishmaniasis, at least in murine model. Only G3 showed significantly (p<0.05) highest ratio of IgG2a/IgG1, revealing the induced protection in this group confirmed by a significant smaller FP swelling (p<0.05), lower LN parasite load (p<0.05), highest IFN-γ/IL-5 ratio (p<0.05) and maximum average of parasitic inhibition percent by flowcytometry and imaging methods.
Despite, using cSLN as a delivery system, the immune responses to the selected antigens (cpb and cpb-CTE) of the cocktail vaccine were differed. Generally, the titres of specific antibodies raised by DNA vaccination are lower than those obtained after vaccination with a recombinant protein. As it is shown in this study antibodies are induced to a very low extent against CPB-CTE, especially in comparison to antibody levels against CPB (Fig. 7B, D). This is definitely attributed to the presence of immunogenic CTE-domain in the vaccine formulations administered to G1 and G2 that affect the character and potency of the responses against defined antigens in the mentioned cocktail vaccines. As a part of our experiments, we have performed an ELISA test in which pooled sera of G1 and G2 were tested against CPB-CTE and sera of G3, G4 tested against recombinant CPB. As a result, G1 and G2 revealed reduced antibody responses while G3, G4 did not show any difference in antibody responses (data not shown). Thus, to avoid presenting exaggerated responses in the benefit for Spa/b-CTE formulation, the presented data show the results of the experiment in which each group plated against its own set of antigens. Therefore, CTE-domain deletion is shown to be an appropriate approach to design a protective vaccine candidate against L. major as well as L. infantum infectious challenge [13].
Above mentioned protective Th1 response demonstrated in G3 that was characterized by increased titres of IgG2a in sera and elevated IFN-γ production by LN cells both before and after challenge, was further supported by the report by Brewer J M et al. indicated that lipid vesicles with a mean diameter >225 nm preferentially induces Th1 responses in BALB/c mice [27]. On the other hand, pcDNA-cps also possesses several immunostimulatory CpG motifs within pcDNA3.1 vector backbone. These CpG motifs might also facilitate priming of CTL responses by activating DCs [28]. Nevertheless, this effect is often temporary because of the rapid degradation of DNA [5], [15] and consequently repeated administrations or much higher doses are required to achieve the desired effects. Moreover, the pDNA delivery to the intracellular compartments for recognition by TLR-9 is hardened [5], [29] without a delivery system, as shown for G2 and G4. Therefore, we can conclude that cSLN as a delivery systems improved the storage stability [17], transfection efficiency [17] and immunostimulatory effects of pDNA-cps (Table 2).
This was possibly the result of pDNAs protection from nuclease activity in vivo, as we have previously reported this potential in vitro [17] and facilitation of pDNAs delivery to the cytoplasm because of cSLN positive zeta potential and the presence of cholesterol domains in cSLN formulation that enhances transfection efficiency by facilitating membrane fusion [17], [30]. In addition, phagocytosis of cSLN-pDNAs by APCs as well as localization of them in the draining LNs occurs easily following SC administration due to the composition and physicochemical characterization of these nanoparticles. The presence of Tween 80 in the formulation enhances this phenomenon as recently Seeballuck and co-workers have also demonstrated that, this surfactant would increase lymphatic uptake by promoting chylomicron formation [31]. Since the draining LNs contain a greater number of cells that express TLR9, localization of pDNAs in the draining LNs possibly will be an important mechanism by which cSLN formulation has enhanced the immunological activity of these antigens. Another important aspect in this formulation could be the presence of DOTAP in this cSLN formulation that can also activate the dendritic cells through a common binding partner with LPS [32]. Therefore, this formulation not only act as a delivery system but also as an adjuvant for Leishmania vaccine by improving the uptake of loaded antigens and also stimulating immune cells in specific way.
In conclusion, this paper clearly demonstrates that cSLN is a promising and adaptable delivery system that can be modified rationally towards specific vaccine targets by varying composition. Simplicity, reproducibility and the scale up possibility of the manufacturing process together with the appropriate immunostimulatory effects of this formulation as a delivery system might be utilized to create a stronger protective vaccine in combination with Leishmania CPs. The current data, in murine model of L. major infection, showed promising role of cSLN as an adjuvant to enhance stronger immune response against Leishmania infection. The mean average of parasite load reduction for such a cocktail pDNA vaccination was determined to be 38% [1]. Here in this study, we report that the percent of parasite inhibition by a particulate cocktail DNA vaccination technology could be increased up to 73.36±15.95%, according to the precise methods for parasite burden determination in the different organs of the challenged animals via both conventional (i.e. microtitration, Fig. 4A) and more novel (i.e. flow cytometry, Fig. 3, 4B and imaging, Fig. 2) techniques. In the entire mentioned techniques parasite burden gave a discriminative view among control groups and Spb-CTE vaccinated animals. Amongst the disscussed methods, direct LN flowcytometry was found to be the most rapid, sensitive, and easily reproducible method for screening vaccination strategies. These promising data warrant further investigations in this regard. Our future studies are being designed to expand cSLN passive targeting to an active targeting to increase the vaccination efficiency. Coating cSLN harbouring pDNA-cps with ligands (such as mannan) are our main future visions to increase the cellular immune responses.
|
10.1371/journal.pntd.0004834 | Identification of Chalcones as Fasciola hepatica Cathepsin L Inhibitors Using a Comprehensive Experimental and Computational Approach | Increased reports of human infections have led fasciolosis, a widespread disease of cattle and sheep caused by the liver flukes Fasciola hepatica and Fasciola gigantica, to be considered an emerging zoonotic disease. Chemotherapy is the main control measure available, and triclabendazole is the preferred drug since is effective against both juvenile and mature parasites. However, resistance to triclabendazole has been reported in several countries urging the search of new chemical entities and target molecules to control fluke infections.
We searched a library of forty flavonoid derivatives for inhibitors of key stage specific Fasciola hepatica cysteine proteases (FhCL3 and FhCL1). Chalcones substituted with phenyl and naphtyl groups emerged as good cathepsin L inhibitors, interacting more frequently with two putative binding sites within the active site cleft of the enzymes. One of the compounds, C34, tightly bounds to juvenile specific FhCL3 with an IC50 of 5.6 μM. We demonstrated that C34 is a slow-reversible inhibitor that interacts with the Cys-His catalytic dyad and key S2 and S3 pocket residues, determinants of the substrate specificity of this family of cysteine proteases. Interestingly, C34 induces a reduction in NEJ ability to migrate through the gut wall and a loss of motility phenotype that leads to NEJ death within a week in vitro, while it is not cytotoxic to bovine cells.
Up to date there are no reports of in vitro screening for non-peptidic inhibitors of Fasciola hepatica cathepsins, while in general these are considered as the best strategy for in vivo inhibition. We have identified chalcones as novel inhibitors of the two main Cathepsins secreted by juvenile and adult liver flukes. Interestingly, one compound (C34) is highly active towards the juvenile enzyme reducing larval ability to penetrate the gut wall and decreasing NEJ´s viability in vitro. These findings open new avenues for the development of novel agents to control fluke infection and possibly other helminthic diseases.
| Despite the widespread prevalence of liver fluke infections and many efforts aimed at developing a preventive vaccine just one drug, triclabendazole, is effective in killing all parasite stages. This is aggravated by the increasing number of countries reporting parasite isolates resistant to triclabendazole treatment, highlighting the urgent need to develop new agents to control fluke infections. Here, we report the discovery of chalcones that effectively inhibits key cysteine proteases essential for parasite development and survival within the host. We further characterized the mode of inhibition of the most active compound against cathepsins secreted by both adult and larval stages of the parasite. This class of compounds is generally considered safe for clinical use and we showed that the most effective derivative is not cytotoxic to bovine sperm cells. Importantly, in vitro approaches showed that it reduces host penetration and larvae viability, finally leading to parasite death. These findings point at this derivative as a good starting point for the development of novel agents to control fluke infection and possibly other helminthic diseases.
| Parasitic flatworms are the causative agents of serious human and livestock infections many of which have been considered neglected tropical diseases in urgent need for research efforts. Liver flukes (Fasciola spp.) cause fasciolosis, traditionally a relevant disease of cattle, sheep and goats [1], currently considered an emerging zoonotic disease by the World Health Organization due to an increased incidence of human infections [2]. Despite many efforts to develop a vaccine to prevent mammalian infection [3], chemotherapy is the only Fasciola control mechanism currently available. Triclabendazole is the first choice drug since it is effective in killing juvenile and mature parasites, but resistance is emerging in several countries [4, 5]. This highlights the urgency of finding novel strategies and target molecules for developing innovative drugs to treat fluke infections.
Many virulence factors have been identified as primary targets for parasite control, since they can be used for developing therapies based on drugs or immunogens. Cysteine proteases play essential roles in numerous protozoan (like Trypanosoma cruzi and Plasmodium falciparum) and other helminth parasites [6], were they have been explored as appropriate targets for antiparasitic chemotherapy [7, 8, 9, 10, 11]. Liver flukes secrete high amounts of cysteine proteases being key players in host-parasite interaction at all stages of their life cycle [12, 13, 14, 15]. Cathepsin L3 is expressed by the newly excysted juveniles (NEJ) and consequently take part in the first steps of mammalian host infection [12, 13, 15, 16]. A different set of cathepsins L are produced by adult flukes residing in the bile ducts [15,16] and are involved mainly in digesting nutrients and developing immune responses mechanisms [17, 18]. Therefore, F. hepatica cathepsins are interesting targets for drug development in an effort to avoid parasite infection or reduce parasite burden and the pathogenic effects of the infection.
Due to their role in human disease and tumour progression, inhibitors targeting cysteine proteases have been extensively studied. Most efforts were focused on peptidic inhibitors with different substituents such as aminoacetonitriles, heterocyclic ketones, nitriles, epoxides and vinyl sulfones [19, 20, 21]. Many of these small molecules contain electrophilic groups that bind in the active-site through covalent interactions with the catalytic cysteine either in a reversible or irreversible way. Non-peptidic compounds have also been reported as cathepsin inhibitors, which are considered a better strategy for in vivo inhibition in order to avoid degradation by proteases. Among these, chalcones and other flavonoids can be found [22, 23, 24]. Flavonoids are biologically active compounds that possess remarkable properties, being presented as antioxidant, anticancer, antidiabetes, anti-inflammatory, antiprotozoal, antiHIV, antituberculosis, among many other interesting activities [25, 26, 27]. What is more, several flavonoids, particularly chalcones, have shown good pharmacological potential and have been approved for clinical use or tested in humans [27]. There have also been described flavonoid derivatives with cathepsin L-like cysteine protease inhibitory activity [23, 28, 29, 30, 31] as well as some natural flavonoids with fasciolicide activity [32, 33]. However, up to date there are no reports of in vitro screening for non-peptidic inhibitors of Fasciola cathepsins (FhCLs) or in vitro screening of synthetic chalcones with fasciolicide activity.
Taking this into account, we performed a search for small molecular weight compounds from our own library of synthetic flavonoids that may inhibit key Fasciola cysteine proteases as FhCL3 and FhCL1. FhCL3 is the only cathepsin L found in the early NEJ excretion/secretion products [12, 13, 14, 16], and FhCL1 is the main cathepsin expressed by adult flukes [14, 15]. This kind of compounds was also previously assayed as anti-tumoral and cancer chemopreventive agents showing non-mutagenic effects and being well tolerated in vivo [34, 35, 36]. Here, we identified novel inhibitors of F. hepatica cathepsins with in vitro fasciolicide activity which shall contribute in the design of novel drugs to control fluke infection.
Since flavonoids have been reported as able to inhibit cysteine protease family enzymes, we evaluated 39 synthetic flavonoids (S1–S3 Tables) from our chemical library. In order to test a variety of chemical entities, we included chalcones without (C1-C8, C20 and C21) or with (C9-C19 and C22-C26) a 2'-substituent in the A ring, chalcones with extended aromaticity (C27-C35) and flavones (C36-C39) along with the natural flavonol quercetin (C40).
FhCL1 and FhCL3 recombinant enzymes were expressed in the yeast Hansenula polymorpha as previously described [17, 39]. Briefly, yeast transformants were cultured in 500 mL YEPD broth (1% glucose, 1% tryptone, 1% yeast extract) at 37°C to an OD600 of 2–6, harvested by centrifugation at 3000xg for 10 min and induced by resuspending in 50 mL of buffered minimal media (0.67% yeast nitrogen base; 0.1M phosphate buffer pH 6.0; 1% methanol) for 36 h at 30°C. Recombinant propeptidases were secreted to the culture media, and recovered by 20–30 fold concentration of culture supernatants by ultrafiltration with a 10 kDa cut-off membrane. The proenzymes were autocatalytically activated to the mature form by incubation for 2 h at 37°C in 0.1 M sodium citrate buffer (pH 5.0) with 2 mM DTT and 2.5 mM EDTA, dialyzed against PBS pH 7.3 and stored in aliquots at -20°C until used. Functionally active recombinant enzyme was determined by titration against the cysteine protease inhibitor E-64c.
A 10 mM stock of each evaluated compound was prepared in dimethylsulfoxide (DMSO). Enzyme activity assays were conducted to evaluate the inhibitory capacity of the different compounds. FhCL1 and FhCL3 were used at nanomolar concentrations to measure initial rates during the first 10 min of the assay and compounds were tested at a moderate fixed dose of 10 μM. Briefly, each enzyme and compound were preincubated 5 minutes in a 96-well plate in 0.1M sodium phosphate buffer pH 6, 1 mM DTT and 1 mM EDTA at room temperature. The reaction was initiated by the addition of 20 μM of substrate, a synthetic peptide conjugated to the fluorophore 7-amino-4-methylcoumarin (AMC). Enzyme activity was measured by the increase in AMC fluorescence as peptide substrates were hydrolyzed (Z-VLK-AMC for FhCL1 and Tos-GPR-AMC for FhCL3) at an excitation wavelength of 340 nm and emission wavelength of 440 nm using a spectrofluorometer (Varioskan Thermo). Enzyme activity was expressed as RFU/s (relative fluorescence units of AMC released per unit of time). Each compound was tested in duplicate. A progress curve without enzyme was performed to control for non-catalyzed reactions between substrates and inhibitors and a spectrum from 300 to 450 nm wavelength was done for each inhibitor to verify that none of the compounds has optical activity in the measurement range. The percentage of enzyme inhibition was calculated as: 100 - (vi/vo) x 100, where vi and vo correspond to the initial rate of AMC fluorescence increase (RFU/s) with and without inhibitor, respectively.
IC50 was calculated at 12 different concentrations of the inhibitor compound, 0, 0.625, 0.937, 1.25, 1.875, 2.5, 3.75, 5, 7.5, 10, 15 and 20 μM. The measurement of enzyme activity was performed in triplicate in a 96-well plate as previously described. We plotted initial rates (RFU/s) versus log10 of inhibitor concentration and obtained the IC50 value from a linear regression of the data. We also carried out slow-binding assays to evaluate if inhibition was time-dependent. C34 at 10 μM was incubated with each enzyme for increasing lengths of time from 3 to 120 min and the percentage of inhibition was determined as described for the screening assay. To test for reversibility of the enzyme-compound interaction, we performed rapid dilution assays [37]. Briefly, samples containing each enzyme at a 100-fold concentration (compared with standard assays) were preincubated with 10-fold the IC50-equivalent concentration of the inhibitor for 20 min at room temperature. Control reactions for each enzyme without inhibitor were carried out in parallel. Samples were then diluted 100-fold with assay buffer containing the appropriate substrate to initiate reactions, and the time course of AMC release was measured as previously describe.
FhCL1 and FhCL3 structures previously obtained by homology modelling were used [38, 39] (Robinson 2011, Corvo 2013). In order to improve structures accuracy for molecular docking, MD simulations were performed using the pmemd module implemented in the AMBER12 package [40], with the ff03.r1 force field [41]. Hydrogen atoms and sodium ions (to neutralize charge) were added to each protein with the leap utility. Each system was placed in a truncated octahedral box of TIP3P explicit water [42], extended 10 Å outside the protein on all sides. The structures of FhCL1 and FhCL3 were treated as follows: a) water and counterions were relaxed to minimize energy during 2,500 steps (500 steepest descent steps, SD, and 2,000 conjugate-gradient steps, CG) with the protein restrained with a force constant of 500 kcal/molÅ2; b) the system was minimized without restraints during 20,000 steps (5,000 SD and 15,000 CG). Long range electrostatic interactions were considered using the particle-mesh Ewald (PME) method [43] and a non-bonded interactions cutoff of 10 Å was used. After minimization, each system was gradually heated in a NVT ensemble from 0 to 300 K over 100 ps using the Berendsen coupling algorithm [44]. This procedure was followed by 20 ns of NPT simulations at 300 K and 1 atm pressure using the Langevin dynamics algorithm [45]. All bonds involving hydrogen atoms were constrained using the SHAKE algorithm [46]. The equations of motion were integrated with a time step of 2.0 fs and coordinates of the systems were saved every 2 ps. Representative structures of FhCL1 and FhCL3 from the last 30 ns of the trajectories were obtained through cluster analysis using the average-linkage algorithm [47] and used for subsequent docking calculations. Clustering, RMSD, RMSF and hydrogen bond analysis were performed using the cpptraj module in AmberTools14. For trajectories visualization the VMD program was used [48].
Compounds 1–39 (S1–S3 Tables) were fully optimized at the ωB97XD/6-31+G(d,p) level [49, 50] in water using the IEF-PCM continuum model [51] with Bondi atomic radii. The nature of the optimized structures as stable species was inspected checking the eigenvalues of the analytic Hessian matrix, calculated at the same level of theory, to be positive in all the cases. All these calculations were performed using the Gaussian09 software [52].
To predict the binding site of flavonoids C1-C39 into FhCL1 and FhCL3 flexible-ligand docking was performed using a grid box of 126×94×116 points with a grid spacing of 0.60 Å in order to cover the entire protein surface (blind docking). The grid box was centered on the macromolecule. Results differing by less than 2.0 Å in root-square deviation were grouped in the same cluster. The conformation with the lowest binding energy was chosen from the most populated cluster and the corresponding ligand-protein complex was used for further MD studies. All docking calculations were done with the AutoDock 4.2 [53] software package using the Lamarckian genetic algorithm. A population size of 150 individuals and 2.5×106 energy evaluations were used for 50 search runs.
MD simulations of C34 with FhCL1 and FhCL3 were performed as described above using the GAFF [54] force field for the ligand. RESP partial charges [55] for C34 were derived using Gaussian09 at the HF/6-31G* level and the antechamber module in AMBER12 was employed to obtain the force field parameters. 40 ns of productive MD were simulated and coordinates of the systems were saved every 10 ps.
F. hepatica metacercariae were acquired from Baldwin Aquatics Inc. (Monmouth, Oregon) for in vitro NEJ treatment and from Instituto Miguel C. Rubino (DILAVE, MGAP, Uruguay) for the gut invasion assay. NEJ were obtained by in vitro excystement as previously described with minor modifications [56], with no differences observed between metacercariae from different origin. Briefly, 100 metacercariae were incubated with 1% sodium hypochlorite for 5 min at room temperature to remove the outer cyst wall and then washed exhaustively with PBS. Metacercarie were activated by incubation at 39°C in a medium prepared by mixing equal volumes of solution A (0.4% sodium taurocholate, 120 mM NaHCO3, 140 mM NaCl pH 8.0) and solution B (50 mM HCl and 33 mM L-cysteine). A 100 μm filters were used to retain the cyst wall as NEJ began to emerge. The excystment process was monitored for m90-180 min under the microscope. Collected NEJ were washed three times with RPMI-1640 supplemented with 200 U/mL Penicillin G sulfate, 200 mg/mL streptomycin sulfate, 500 ng/mL amphotericin B, 10 mM HEPES, counted and divided in groups of around 20 parasites that were transferred to 12 wells tissue culture plates. Parasites were maintained at 37°C, 5% CO2 in modified Basch’s medium [57]. At day 1, C34 (50 μM) was added to treated groups and 0.5% DMSO to control groups, each condition tested by duplicate. Medium was changed twice a week and fresh compound was added, control NEJ were cultured for 20 days. NEJ behavior was monitored under a light microscope (Olympus BX41), every day each well was recorded for a minute in order to assess parasite motility and registered using the following score: 3-normally active; 2- reduced activity (sporadic movement); 1- immotile (adapted from [58]). For the gut invasion assay we performed an in vitro excystement of metacercarie as previously described and incubated the NEJ for 4 h either in the presence of 50 μM C34 or 0.5% DMSO (control group) and immediately transferred them to gut sacs. For gut preparation we used the protocol previously described by Burden et al. with some modifications [59]. Brifely, 6 weeks old male Wistar rats were euthanized by cervical dislocation and approximately 25 cm of small intestine was excised using the caecum as a reference. The intestine was washed several times with RPMI-1640 media warmed at 37°C and cut in 5 cm sacs. One end of each sac was ligated, 20–30 NEJ from treated or control groups were pipetted inside the sac and then the other end was ligated. Gut sacs were maintained in RPMI-1640 media at 37°C, 5% CO2 in 12 wells plates and parasites that migrated through the mucosal wall within 3 h were recovered at the plate bottom and counted. Each condition was assayed in triplicate.
Semen samples were obtained from a healthy fertile Hereford bull and kept frozen in 0.5 mL straws (extended in Andromed, Minitube, Germany) under liquid nitrogen until use. The semen used belonged to a single freezing batch that was obtained during a regular collection schedule with an artificial vagina. Samples from three straws were thawed and a sperm pool was prepared in PBS at a concentration of 40 million spermatozoa per mL, then 50 μL of this sperm suspension was carefully mixed with 50 μL of C34 diluted to 100, 50, 25, 12.5 and 6.25 μM or with 1% DMSO in control experiments. Each condition was assayed by duplicate in 96-well plates and controls were assayed by triplicate. Plates were incubated at 37°C for 1 h with moderate shaking. The motility analysis was carried out using a CASA (Computer Assisted Semen Analyzer) system Androvision (Minitube, Tiefenbach, Germany) with an Olympus BX 41 microscope (Olympus, Japan) equipped with a warm-stage at 37°C. Each sample (10 μL) was placed onto a Makler Counting Chamber (deph 10 μm, Sefi-Medical Instruments, Israel) and the following parameters were evaluated: percentage of total motile spermatozoa (motility >5 μm/s) and velocity curved line (VCL, >24 μm/s). At least 400 spermatozoa were analysed from each sample from at least four microscope fields.
We evaluated forty flavonoids from three structural clusters of our own chemistry library as inhibitors of FhCL1 and FhCL3. Cluster I contained twenty-six chalcones modified in ring B (C1-C26, Fig 1A), cluster II contained naphtolchalcone derivatives (C27-C35, Fig 1B) and cluster III was composed of flavones modified in ring B (C36-C40, Fig 1B). Initially, we screened the compounds at a low concentration of 10 μM measuring initial rates during the first 10 min, finding activities ranging from 0% to 75% enzyme inhibition (Fig 1). Roughly, a larger amount of compounds decreased FhCL1 activity considerably, while FhCL3 proved more difficult to inhibit. Thus, eleven compounds showed more than 50% inhibition of FhCL1 (C3, C10, C22-C24, C27, C30, C31, C33, C34 and C35) (Fig 1), while two of them, belonging to cluster II, were above this cut-off for FhCL3, C34 and C35 (Fig 1B), being C34 the best global inhibitor. None of the assayed flavones (cluster III) displayed relevant activities against both FhCL1 and FhCL3 (% inhibition at 10 μM lower than 20%, Fig 1B) for this reason the studied population of flavones was lower than the chalcones one.
In general, chalcones containing heterocycles as B-ring in cluster I (C22-C26) and those with extended aromaticity belonging to cluster II showed the highest inhibition percentages. Clearly, this kind of compounds inhibited more readily FhCL1 than FhCL3 (fourteen showing at least 50% inhibition of FhCL1 vs two with FhCL3 at 10 μM). These might be explained considering the active site pocket structure and substrate preferences of the enzymes. While FhCL1 has a wide S2 pocket which easily accommodates bulky and aromatic residues, FhCL3 has a narrow and restricted site that preferentially interacts with small moieties like proline and glycine [39]. Strikingly, C34 substituted by a hydroxyl at position 2 of the A-ring and with a naphthyl on both rings, was the best inhibitor of FhCL3, showing 75% inhibition for FhCL1 and 65% for FhCL3 in the screening. Likewise, the highest percentages of FhCL3 inhibition correspond to the same type of compounds (C28, C30, C31, C35) (Fig 1B). We hypothesized that the narrower conformation of FhCL3 active site compared to FhCL1 diminishes the number of compounds that manage to accommodate in such a position as to allow enzyme inhibition.
There are some interesting compounds showing selectivity towards FhCL1, for example derivatives C4, C13 and C22-C25 exhibiting around 50% inhibition of the former enzyme and poor or no inhibition of FhCL3. These compounds belong to the group containing heterocycles as B-ring (C22-C25) or 4-methoxy moiety on the B-phenyl ring (C4, C13), that might establish hydrogen bonds with specific residues of this enzyme (S4 Table). Particularly, C20, the non-hydroxylated analogue of C22, does not inhibit FhCL1, suggesting the 2´-OH might also be playing an important role in this inhibition. A chalcone with extended aromaticity (C27) also exhibits 65% inhibition of FhCL1 and only 13% of FhCL3, being the only one with a phenyl substituent as B-ring. Notably, among these bulky compounds, the less active towards FhCL1 was C29, which lacks Michael-acceptor motive susceptible of nucleophilic attack by enzyme residues (Fig 1B, S2 Table).
The p-Cl substituted chalcone C10 also showed good inhibition for FhCL1 (63%) (Fig 1A, S1 Table). Likewise, this compound offers some interesting structure-activity relationships. Here, a positive influence of the 2´-OH in the inhibition might be addressed again when comparing C10 with its de-2´OH analogue C2 that exhibits only 25% of inhibition. Besides, its bioisostere p-Br substituted on the B-ring (C11) depicts poor inhibition of FhCL1 (19%), suggesting that it is the combination of both 2´-OH and p-Cl substituents that exerts optimum inhibition of this enzyme (Fig 1A, S1 Table).
We selected C34 to perform additional characterization as it was the best inhibitor of both target enzymes. Consequently, a dose-response study was performed. The IC50 was similar with both enzymes, being 7.7 μM and 5.6 μM for FhCL1 and FhCL3 respectively (Table 1). Interestingly, the interaction mode with each enzyme seems to differ. Longer incubation times at 10 μM resulted in higher inhibition of FhCL3 but not FhCL1, thus C34 interacts in a slow-binding time-dependent fashion only with the former enzyme. Furthermore, when increasing the preincubation time with the enzyme, FhCL3 is almost completely inhibited (96%) (Table 1).
We then evaluated the reversibility of the inhibition by measuring the recovery of enzymatic activity after a rapid and large dilution of the enzyme–inhibitor complex. When the initial rate of the reaction is similar in the presence and absence of the inhibitor, it indicates a rapid recovery of activity after compound dilution meaning the reaction is reversible. This was the case for the interaction of C34 with FhCL1 (Fig 2, left panel), where the progress curve after inhibitor dilution has a slope similar to that of the control sample (enzyme incubated and diluted in the absence of inhibitor); however, we found a different behavior with FhCL3. Here, the progress curves have a lag phase followed by a linear phase (Fig 2, right panel), reflecting the slow recovery of activity as inhibitor dissociates from the enzyme and suggesting that the interaction of C34 with FhCL3 is slowly reversible being the compound tightly bound to the enzyme.
According to docking results the studied compounds most frequently interact with a putative binding site, represented in cyan for FhCL1 and orange for FhCL3, and found at a similar position within the active site cleft of each enzyme (Fig 3). A second putative binding site (represented in blue) was observed for FhCL3, which has a different orientation and is not as deep located inside the binding site (Fig 3).
To obtain a deeper knowledge of C34 interaction with target enzymes, 40 ns molecular dynamics simulations were performed allowing us to confirm the binding sites predicted by docking, as we obtained stable complexes in the last 35 ns of the simulation (S1 Fig). Thus, the high percentage of inhibition experimentally observed can be attributed to the compound locating next to residues directly implicated in substrate positioning and catalysis (Figs 4 and 5). From a population cluster analysis, we determined the most representative structure for the enzyme-inhibitor complex for FhCL1, with an occurrence of 51% (Fig 4 upper panel). We found two more clusters with occurrences of 24% and 22% that, while changing the relative position and interaction with certain residues, still occupy the active site (Fig 4 middle and lower panels, respectively), and two last clusters with lower occurrences of 2 and 1% (not shown). For FhCL3, C34 occupies the orange site (Fig 3) along the entire simulation time. A major cluster with an occurrence of 59% is observed (Fig 5 upper panel), followed by another two clusters with lower occurrences of 23% and 11% (Fig 5 middle and lower panels, respectively), and two with less occurrences of 6 and 1% (not shown), similar in location and relative position to the major cluster.
The experimentally observed variation in inhibition percentages might be explained by the high number of interactions established by the ligand with each target. It is worth noticing that the experimental results were in good correlation with computational observations. C34 is positioned along the active site cleft in both enzymes and interacts with Cys-His catalytic dyad and residues from the S2 and S3 pockets (Figs 4 and 5), determinants of the substrate specificity of this family of cysteine proteases [39, 61]. FhCL1 establishes several hydrophobic interactions with the inhibitor, particularly through residues Gly66, Gly67, Val160 and Asn161 from S2 and S3 subsites, and Cys25 from the catalytic dyad (Fig 4). Likewise, we observed many hydrophobic interactions of C34 with FhCL3, with a larger number of residues belonging to the enzyme sub-sites: Trp69, Met70, Ala135, Val160, Thr161, Val163 and Ala209, which form the entire S2 subsite, plus Gly68 belonging to S3 subsite, both decisive in FhCL3 substrate positioning for catalysis (Fig 5). In this sense, the high inhibition experimentally observed for C34 with both enzymes is confirmed.
Interestingly, Trp69 can contribute to inhibitor binding by establishing Pi-Pi stacking interactions. This residue is specific of FhCL3 and when mutated for Leu as in FhCL1 it renders the enzyme almost inactive highlighting its importance for substrate positioning for catalysis [39]. A recent in silico search for FhCL3 inhibitors provides additional evidence that the non-polar side chain of Trp69 establishes critical interaction with ligands and adopts variable conformations to accommodate different groups in the enzyme binding site [62]. Furthermore, it showed that aromatic moieties with high hydrophobicity can establish favorable interactions with non-polar residues from FhCL3 binding cleft [62]. The greater number of interactions between inhibitor and residues from FhCL3 sub-sites compared with FhCL1, might explain its lower IC50 (Table 1), its behavior as a slowly reversible inhibitor (Fig 2), and its highest percentage of inhibition reached after 120 min (Table 1).
No hydrogen bonds with a significant occupancy (greater than 50) along the simulation time were found in any of the enzyme-inhibitor complexes, except for the bond between carbonyl oxygen in C34 and Gly68 with an occupancy of 22 in FhCL1 (Fig 4), analogous to the interaction with Glu72 in FhCL3, with only 11 of occupancy (Fig 5). However, there are significant differences in the hydrogen bonds at the protein structural level, where some intra-molecular protein hydrogen bonds have an occupancy greater than 90 in the absence of inhibitor but disappear in its presence, mainly in the FhCL3-inhibitor complex (Fig 6).
If we analyze the hydrogen bonds involving residues from sub-sites S2 and S3 in FhCL1 (Fig 6 and S5 Table, residues in bold), we see a tendency to diminish its occupancy in the presence of C34, supporting its interaction along the active site alters local protein conformation. A similar behavior is observed in FhCL3, with a more remarkable loss in occupancy seen for hydrogen bonds involving Val136-Thr161, Asp137- Thr161 and His63-Arg58 (Fig 6 and S6 Table, residues in bold). Moreover, upon ligand binding new hydrogen bonds are established between backbone and side chains of sub-site residues (with occupancy greater than 50) in the proximity of C34 binding site (Fig 6, HB with ΔOcc.˂0), named His63-Gly66, Trp69-Asn62 and Ala135-Ala163 (S6 Table, residues in bold), again suggesting the interaction with C34 dramatically alters protein conformation and thus its ability for catalysis.
We evaluated the effect of C34 on NEJ parasites cultured in vitro. NEJ were incubated in the presence of 50 μM C34 and the assessment of parasite movement is shown in Fig 7.
The addition of C34 resulted in a decrease in parasite motility leading to parasite death. At day 4 the percentage of immobile flukes was considerably higher in treated vs control groups, 67% vs 0% respectively. The percentage of sporadic movement and immobile parasites increased in treated parasites and evident signs of internal and tegument damage appeared. Culture media was dirty as compared to control flukes due to the release of body components and some NEJ died protruding their anterior region (Fig 7, S1 and S2 Videos). After day 10 no movement was detected in treated NEJ while control parasites continue moving until the experiment was finished on day 20, thus, motility seems to adequately estimate worm viability.
There have been controversial findings regarding the effect of cathepsin L knock down by RNAi. While a recent work described no phenotypic changes in NEJ even after 21 days of exposure to long dsRNA or siRNA [63], an earlier report had documented immobile NEJ after a 4h interference with Cat-L and B siRNA [64]. Here, we observed a reduction in fluke motility consistent with this first report. Additional evidence supporting this result comes from an in vitro treatment of NEJ with the cathepsin L and B inhibitors E64-d and CA-074Me, which resulted in the loss of worm motility accompanied by structural damage of parasites [65]. In a similar fashion, they observed a progressive loss of motility starting 5 to 12 days after treatment initiation. As a consequence of decreased motility, a diminished ability of parasites to traverse the duodenum wall was reported [64]. In order to check if C34 treatment can also impair invasion, we performed a gut penetration assay. NEJ incubated 4h with C34 showed the expected phenotype of reduced migration through the gut wall, suggesting the impairment of cathepsin L function (Fig 8). While the in vitro treatment effect of movement reduction is seen after longer incubation times with C34, a reduced gut penetration is readily detected in the invasion assay. This is indicative that the main role of juvenile cathepsin L3 in parasite host invasion is promptly affected, and a secondary effect on parasite movement might emerge later. In the host environment, this might translate into a considerable reduction in the number of NEJ able to reach the bile ducts and establish a successful infection.
Citotoxicity assays using spermatozoa are considered a suitable approach for preclinical toxicology screening during drug development processes [66]. The cytotoxicity assay on bovine spermatozoa showed C34 is not toxic to reproductive cells even at the highest tested dose of 100 μM. We observed no significant differences on percentage of motile spermatozoa and velocity curved line between treated and control samples (S7 Table). Also, our previous studies showed C34 is not mutagenic in Ames test using Salmonella typhimurium TA98 strain with and without metabolic activation and no genotoxic effect was observed measuring DNA damage in the alkaline comet assay [36], supporting the fact that this class of compounds exhibits low toxicity and are promising agents for clinical usage.
In this work we have identified novel chalcones that inhibit Fasciola hepatica major cathepsins L from both adult and juvenile parasites. Chalcones with heterocycles as B-ring (C22-C26) and chalcones containing phenyl and naphtyl moieties (C27-C35) showed the highest inhibition percentages. Among these, C34 resulted a slow-reversible tight binding inhibitor of FhCL3, exerting an almost complete inactivation of the enzyme. In addition, we performed an extensively computational analysis characterizing C34 interaction with cathepsins providing evidence of the different inhibition seen with each target enzyme. We demonstrated C34 is not toxic to bovine sperm, exhibits in vitro fasciolicide activity over cultured NEJ and reduces larval penetration of the gut wall, suggesting is a suitable candidate for further drug development against worm infection.
|
10.1371/journal.pntd.0005860 | Vector competence of populations of Aedes aegypti from three distinct cities in Kenya for chikungunya virus | In April, 2004, chikungunya virus (CHIKV) re-emerged in Kenya and eventually spread to the islands in the Indian Ocean basin, South-East Asia, and the Americas. The virus, which is often associated with high levels of viremia in humans, is mostly transmitted by the urban vector, Aedes aegypti. The expansion of CHIKV presents a public health challenge both locally and internationally. In this study, we investigated the ability of Ae. aegypti mosquitoes from three distinct cities in Kenya; Mombasa (outbreak prone), Kisumu, and Nairobi (no documented outbreak) to transmit CHIKV.
Aedes aegypti mosquito populations were exposed to different doses of CHIKV (105.6–7.5 plaque-forming units[PFU]/ml) in an infectious blood meal. Transmission was ascertained by collecting and testing saliva samples from individual mosquitoes at 5, 7, 9, and 14 days post exposure. Infection and dissemination were estimated by testing body and legs, respectively, for individual mosquitoes at selected days post exposure. Tissue culture assays were used to determine the presence of infectious viral particles in the body, leg, and saliva samples. The number of days post exposure had no effect on infection, dissemination, or transmission rates, but these rates increased with an increase in exposure dose in all three populations. Although the rates were highest in Ae. aegypti from Mombasa at titers ≥106.9 PFU/ml, the differences observed were not statistically significant (χ2 ≤ 1.04, DF = 1, P ≥ 0.31). Overall, about 71% of the infected mosquitoes developed a disseminated infection, of which 21% successfully transmitted the virus into a capillary tube, giving an estimated transmission rate of about 10% for mosquitoes that ingested ≥106.9 PFU/ml of CHIKV. All three populations of Ae. aegypti were infectious as early as 5–7 days post exposure. On average, viral dissemination only occurred when body titers were ≥104 PFU/ml in all populations.
Populations of Ae. aegypti from Mombasa, Nairobi, and Kisumu were all competent laboratory vectors of CHIKV. Viremia of the infectious blood meal was an important factor in Ae. aegypti susceptibility and transmission of CHIKV. In addition to viremia levels, temperature and feeding behavior of Ae. aegypti may also contribute to the observed disease patterns.
| A chikungunya epidemic recently occurred in Mandera, Northern Kenya, with over 1,700 cases reported. The disease epidemics are linked to the urban vector, Aedes aegypti. This mosquito species is rapidly expanding its range and it is currently abundant in and around the major urban cities of Kenya. In this study, we demonstrated the ability of Ae. aegypti from three distinct cities of Kenya to transmit chikungunya virus (CHIKV) under laboratory conditions. Our findings showed that populations of Ae. aegypti from Mombasa, Kisumu, and Nairobi were competent vectors for CHIKV. Overall, about 60% of the Ae. aegypti that ingested ≥106.9 plaque-forming units of virus/ml became infected and about 10% of the virus-exposed mosquitoes transmitted virus to a capillary tube. Vector competence remains a prerequisite in disease risk assessment, while surveillance and control of Ae. aegypti should remain the main focus in many disease control programs and should be performed routinely.
| Chikungunya is a re-emerging mosquito-borne infectious disease caused by chikungunya virus (CHIKV), a member of the genus Alphavirus in the family Togaviridae. The disease, which may manifest as febrile illness, is notorious for inflicting severe morbidity in form of prolonged joint pain which may persists for weeks or months in some patients [1]. Originally isolated in Tanzania in 1953 [2], the virus has spread causing major outbreaks in tropical Africa, islands in the Indian Ocean basin, and South-East Asia [3]. In November 2013, CHIKV was transmitted locally in the Americas for the first time, and over 2 million cases have been reported since then [4–6]. Similarly, locally transmitted cases have been detected in Europe [7,8]. In Kenya, major outbreaks occurred between 2004 and 2005 in Lamu and Mombasa in the Coastal Region, with at least 13,500 human cases, and as much as 75% of the population in Lamu affected [9]. In addition, a recent outbreak occurred in Mandera with 1,792 human cases recorded [10]. Overall, the ongoing expansion of its range presents a worrying public health trend at both local and global scales.
Both viral and vector factors have been ascribed to the global expansion of CHIKV. Amongst these is the rapid colonization and expanding habitat of the key Aedes species involved [11]. Interestingly, chikungunya outbreaks in West and Central Africa have tended to occur in smaller scales and largely in a sylvatic cycle involving humans and non-human primates and forest-dwelling Aedes species notably Ae. furcifer-taylori group, Ae. africanus, Ae. luteocephalus, and Ae. neoafricanus [12,13]. In stark contrast, larger scale outbreaks, mainly in urban and periurban settings, have largely been associated with the peridomestic and highly anthropophilic Ae. aegypti, as has been the case in recent outbreaks in East Africa, Asia, and the Americas [14,15].
Chikungunya has been reported in Mombasa city but so far has not been documented in Kisumu and Nairobi, although the presence and abundance of Ae. aegypti has been associated with urban areas [16,17]. Because the relative vector competence of different populations of Ae. aegypti can differ greatly for CHIKV [18], we hypothesized that the differences in the histories of chikungunya outbreaks in various areas in Kenya might be explained by the relative vector competence of these populations. Therefore, we tested populations of Ae. aegypti collected in Kilifi on the Coastal region near Mombasa, Nairobi, and Kisumu for their relative ability to transmit CHIKV.
Aedes aegypti was collected from selected sites in the three major cities in Kenya. Mosquitoes were collected as eggs using oviposition cups (black cups lined with oviposition papers and half filled with water) and as larvae from water holding containers in and around houses, between March and April 2016 (Table 1). The eggs and larvae were transported to the BSL-2 insectary at the Duduville Campus, International Centre of Insect Physiology and Ecology (ICIPE) in Nairobi, where the eggs were hatched and the larvae reared to provide F0 adult mosquitoes for this study. Adult mosquitoes were identified to confirm that they were Ae. aegypti, and a portion of them were blood fed on laboratory mice (Kenya Medical Research Institute, Animal House) to provide eggs. These were hatched, reared at 28°C and provided fish food (Tetramin) as larval food to produce F1 mosquitoes. The same procedure was used to produce F2 mosquitoes. Adult mosquitoes were provided 8% glucose as a carbohydrate source, which was replaced with water 24 hours prior to virus exposure. We used F0-2 mosquitoes in this study.
The Lamu001 strain of an East/Central/South Africa lineage of CHIKV, isolated during the 2004–2005 outbreak on Lamu Island, was used for all the infection assays performed in this study. The virus was amplified in T-25 cell culture flasks (Corning Incorporated, USA) containing confluent monolayers of Vero cells (ATTC CCL-81), grown in cell culture media consisting of Minimum Essential Medium (MEM) (Sigma-Aldrich, St. Louis, MO) with Earle's salts and reduced NaHCO3, supplemented with 10% heat-inactivated fetal bovine serum (FBS) (Sigma-Aldrich), 2% L-glutamine (Sigma-Aldrich), and 2% antibiotic/antimycotic solution with 10,000 units penicillin, 10 mg streptomycin and 25μg amphotericin B per ml (Sigma-Aldrich,). The inoculated cells were incubated at 37°C for 1 hour, to allow for virus adsorption. Maintenance medium (MEM supplemented with 2% FBS) was then added, and the cells were incubated at 37°C in a 5% CO2 incubator and observed for cytopathic effects (CPE). After 24 hours, 80% CPE was observed and a portion of the CHIKV-media suspension cell suspension was harvested, added to defibrinated sheep blood (Central Veterinary Laboratories Kabete, Kenya) and used without freezing to produce an infectious blood meal used to expose mosquitoes to CHIKV.
When the laboratory-reared, F1 or F2 mosquitoes from Mombasa, Kisumu, or Nairobi were 4–5 days old (or the F0 mosquitoes from Mombasa and Nairobi were 3–12 days old), they were exposed to blood meals containing four different titers. A Hemotek membrane feeding system (Discovery Workshops, Accrington, the United Kingdom), maintained at 37°C, was covered with mouse skin (Kenya Medical Research Institute, Animal House) and used for artificial blood feeding. All feeding was performed in a BSL-2 insectary at ICIPE.
In the first experiment, 100 μl of freshly harvested virus was added to 9.9 ml of cell culture media to produce a 1:100 stock virus. 3 ml of this stock was added to 7 ml of defibrinated sheep blood. We added 2 ml of the CHIKV-blood suspension to each well of a Hemotek feeder and Ae. aegypti from the three locations were allowed to feed for about 1 hour. Immediately after making the blood virus suspension, 100 μl of the suspension were added to 900 μL of homogenization media (MEM, supplemented with 15% FBS) to make a 1:10 suspension of the time-0 blood. At the end of the 1-hour feeding period, 100 μl of the virus blood meal were removed from one of the Hemotek feeders and added to 900 μl of homogenization media to determine an end of feeding concentration.
In a second experiment, conducted a few hours later, 3 ml of the freshly grown, CHIKV cell culture suspension were added directly to 7 ml of sheep blood to create a virus suspension. All other procedures remained the same as in the first experiment.
With all other procedures remaining the same, the experiments were repeated using a 1:10 dilution and undiluted freshly grown virus with titers different from those in the first two experiments. Therefore, in four separate experiments, mosquitoes were exposed to infectious blood meals containing different titers.
After feeding, all unengorged mosquitoes were removed and the cages containing the engorged mosquitoes were maintained in an insectary at 28°C, 12:12 (L:D) photoperiod, and a cotton pad containing 8% glucose solution was placed on top of the cage. On days 5, 7, 9, and 14, a sample of the mosquitoes was removed, placed in small plastic cups (covered with a fine netting material and secured with rubber bands), and cold anesthetized by placing in a refrigerator at -20°C for about 40 seconds.
The legs and wings of each mosquito were removed and the body placed on a sticky tape. The mosquito’s proboscis was inserted into a capillary tube containing 15–20 μl of homogenization media, and left to salivate for 30 minutes. The saliva containing media was eluted to 200 μl of homogenization media and the samples stored at -80°C until assayed for virus by cell culture. The body and legs were placed separately in microcentrifuge tubes containing 1 ml of homogenization media and stored at -80°C until assayed for virus by plaque assay.
Quantification of CHIKV-blood meal, and mosquito body and leg samples was performed by plaque assay. Mosquito bodies were homogenized using a Minibeadbeater (BioSpec Products Inc, Bartlesville, OK 74005 USA) with the aid of a copper bead (BB-caliber airgun shot) and clarified by centrifugation at 12,000 rpm (Eppendorf centrifuge 5417R) for 10 mins at 4°C. Serial 10-fold dilutions were prepared and inoculated in 12-well plates containing confluent Vero cell monolayers. Each well was inoculated with 100 μl of virus/blood or body dilutions, incubated for 1 hour to allow for adsorption, with frequent agitation/rocking. The infected cells were then maintained using 2.5% methylcellulose mixed with 2X MEM and incubated at 37°C with 5% CO2. On the fourth day, the plates were fixed for 1 hour with 10% formalin, and then stained for 1 hour with 0.5% crystal violet solution. Plaques were counted on a light box. Only the legs of the positive mosquito bodies were homogenized and tested in the same way, to determine the dissemination rate.
To test for virus transmission, 80 μl of the saliva sample was inoculated into a well of a 24-well plate containing confluent Vero cell monolayers. Plates were incubated for 1 hour to allow for adsorption, with frequent agitation/rocking. The infected cells were maintained using maintenance media (1 ml per well) and incubated at 37°C with 5% CO2. Plates were observed for 7 days and the supernatant of wells showing CPE were harvested and virus quantified by plaque assay.
If the virus was detected in the mosquito’s body but not in the legs, the mosquito was considered to have a non-disseminated infection limited to the midgut. Detection of virus in the body and legs was considered as evidence of a disseminated infection [19]. All samples that contained CHIKV in their saliva were considered competent in transmitting the virus. The overall infection and dissemination rates for Ae. aegypti populations from Mombasa, Kisumu and Nairobi were compared using Chi-squared tests. Body titers for mosquitoes with disseminated and non-disseminated infections were compared using a t-test. All analysis was performed in R version 3.3.1 [20] at α = 0.05 level of significance. We used the exact (binomial) method of calculating 95% confidence intervals (C.I.) (https://measuringu.com/wald/).
Scientific and ethical approval was obtained from Kenya Medical Research Institute Scientific and Ethics Review Unit (KEMRI-SERU) (Project Number SERU 2787). The animal use component was reviewed and approved (approval number KEMRI/ACUC/ 03.03.14) by the KEMRI Animal Use and Care committee (KEMRI ACUC). The KEMRI ACUC adheres to national guidelines on the care and use of animals in research and education in Kenya enforced by National Commission for Science, Technology and Innovation (NACOSTI). The Institute has a foreign assurance identification number F16-00211 (A5879-01) from the Office of Laboratory Animal Welfare (OLAW) under the Public Health Service and commits to the International Guiding Principles for Biomedical Research Involving Animals.
The titers in the infectious blood meals ranged from 105.6–107.5 PFU/ml, with titers of pre- and post-feeding samples for each meal being nearly identical. Although the number of days post virus exposure at any given dose did not have any significant effect on susceptibility to the virus (Table 2), the overall infection rates for all three geographic populations increased with an increase in the exposure dose, with very low infection rates in all three populations when ≤105.9 PFU/ml were ingested. At the higher exposure doses 106.9–7.5, infection rates were highest in the mosquitoes derived from those collected in Mombasa. However, these differences were not statistically significant (χ2 ≤ 1.04, DF = 1, P ≥ 0.31).
In all three populations, viral dissemination was observed as early as 5–7 days of extrinsic incubation, when mosquitoes were exposed to titers of ≥105.9 PFU/ml. Although dissemination rates appeared to increase with increasing exposure doses, virtually all of this increase could be accounted for by an increase in infection rates. Also, the observed dissemination rates at 106.9–7.5 were highest in the population of mosquitoes from Mombasa but the difference was however not significant (χ2 = 3.66, DF = 2, P = 0.16). Regardless of mosquito origin, infectious dose, or period of extrinsic incubation, about 71% of the infected mosquitoes had developed a disseminated infection (Table 3).
Similar to viral dissemination, viral transmission was only detected in mosquitoes that ingested ≥105.9 PFU/ml (Table 4). Although transmission rates appeared to increase with increasing exposure doses, virtually all of this increase was accounted for by an increase in dissemination rates. The transmission rates ranged from 0–2% at a viremia level of 105.9, 3–15% at a viremia level of 106.9, and 13–19% at a viremia level of 107.5. Regardless of mosquito origin, infectious dose, or periods of extrinsic incubation, about 21% of the mosquitoes with a disseminated infection were able to transmit infectious virus to the capillary tube (Table 4).
Based on the titers detected for each population of Ae. aegypti, we observed that mosquitoes that were susceptible to infection and failed to disseminate the virus had titers at least a log lower than mosquitoes which were susceptible and had disseminated the virus (Table 5). This difference was significant (t = 8.10, DF = 4, P = 0.0012). In general, viral dissemination only occurred when body titers were ≥104 in all populations (Table 5). However, for mosquitoes with a disseminated infection, no significant difference in leg titers was observed for those that did, or did not, transmit virus by bite (t = 0, DF = 4, P = 1.0).
Populations of Ae. aegypti from Mombasa, Nairobi, and Kisumu were all competent vectors of CHIKV. The recent outbreaks of chikungunya in Africa, the Americas, Asia, and Europe [7,10,21,22], clearly demonstrate the potential for CHIKV to spread to new geographical areas and cause massive epidemics. The most likely way that CHIKV may be introduced into new areas is by a person infected in an area where CHIKV is currently being transmitted traveling to an area where susceptible people and competent vectors exist. In addition, it is possible for an infected mosquito to be transported from an area of active transmission to an area with susceptible people and vectors [21,23]. The risk of importation of CHIKV into new areas is high because chikungunya epidemics often result in high attack rates, high viremia levels among infected individuals, as well as the widely distributed Aedes vectors [14]. Due to the movement of humans to cities, these areas may be associated with higher risk of vector-borne pathogens, such as CHIKV. Aedes aegypti remains the only known urban vector for CHIKV transmission in Kenya. The CHIKV titers used in this study to expose mosquitoes are similar to published viremia levels associated with human infections (often >105 PFU/ml of blood) in nature [24].
All populations of Ae. aegypti tested were able to transmit CHIKV under laboratory conditions, indicating that mosquitoes in each of these areas were competent vectors of CHIKV. Although infection rates were higher among mosquitoes from Mombasa as compared to Kisumu and Nairobi, these differences were not statistically significant, suggesting a higher CHIKV susceptibility for the Ae. aegypti from Mombasa was not the primary reason for the increased risk of CHIKV transmission in this area. For all three Ae. aegypti populations, about 70% of mosquitoes that became infected with CHIKV developed a disseminated infection. Thus, a midgut infection barrier may be an important factor affecting vector competence, particularly at lower viremias, as was suggested in another study on the dengue virus [25]. In addition, only 18 (21%) of 84 mosquitoes with a disseminated infection transmitted CHIKV to a capillary tube indicating a significant salivary gland barrier, although a mosquito may secrete less virus into a capillary tube than it would when feeding on an animal [26]. Transmission rates are therefore often lower when they are determined by collection of saliva as compared to allowing the mosquito to feed on a susceptible animal [27]. Therefore, failure to detect CHIKV in the saliva collected in a capillary tube does not mean that the mosquito would not have transmitted the virus by bite if it fed on a susceptible human, and our transmission rates should be considered as minimum transmission rates.
Interestingly, for mosquitoes from each of the three sites, dissemination and transmission rates reached high levels by 5–7 days after virus exposure (Tables 3 and 4). Therefore, at the viremia doses and temperature (28°C) used in this study, Ae. aegypti would be able to attain peak transmission rates in less than 1 week after feeding on a viremic person. This extrinsic incubation period is shorter than those described for other viruses transmitted by Ae. aegypti, including 7 to 12 days at temperatures ≥30°C for DEN [28] and a median of 10 days at 25°C for yellow fever [29].
Although we observed a significant difference (P = 0.0012) in the body titers of mosquitoes that did, or did not, disseminate CHIKV, we did not observe any difference in leg titers for mosquitoes with a disseminated infection that did, or did not, transmit virus (P = 1.0). This suggests that the salivary gland barrier which determines the ability of the virus to penetrate into the salivary glands and be secreted into the saliva is independent of the body titer in a mosquito with a disseminated infection [30].
Although, transmission rates trended higher among the Mombasa populations, the differences were not statistically significant compared to the populations in Kisumu and Nairobi, or more importantly, biologically meaningful. The higher transmission rates observed in the mosquitoes from Mombasa are, however, consistent with the higher chikungunya epidemics in this part of the country. The lower temperatures in Nairobi (average monthly temperatures 22–28°C) as compared to those in Mombasa (average monthly temperatures 27°C—31°C) may be a major contributing factor to the absence of chikungunya in the Nairobi area, as earlier studies demonstrated that temperature plays a significant role in the susceptibility of Ae. aegypti to CHIKV [31,32]. However, temperature cannot explain the low infection rates with CHIKV in Kisumu as monthly temperatures there range from 28–30°C. Also, the Ae. aegypti populations in these urban areas of Kenya may differ in their blood feeding behavior. This may be because the subspecies present in Mombasa may be predominantly Ae. aegypti aegypti, which has been described as more anthropophagic than the more sylvatic Ae. aegypti formosus strain mostly found inland and in forests [33,34]. Increased feeding on humans would have a much larger effect on CHIKV transmission than a moderate difference in vector competence, and may partly explain why CHIKV remains essentially absent in Kisumu, despite its relatively high temperatures.
In conclusion, although all three populations of Ae. aegypti were competent laboratory vectors for CHIKV, the Mombasa population appeared to be slightly more competent than the population from Kisumu and Nairobi. Findings from this study clearly demonstrated the importance of viremia levels in Ae. aegypti susceptibility to CHIKV. Vector competence is an important prerequisite in evaluating risk of emergence of CHIKV in addition to vector densities and host preference evaluation. Surveillance and control of the domestic vector, Ae. aegypti, should remain the main focus in many disease control programs and should be performed routinely where the risk is found to be high.
|
10.1371/journal.pntd.0001046 | Use of Saliva for Early Dengue Diagnosis | The necessity of a venous blood collection in all dengue diagnostic assays and the high cost of tests that are available for testing during the viraemic period hinder early detection of dengue cases and thus could delay cluster management. This study reports the utility of saliva in an assay that detects dengue virus (DENV)–specific immunoglobulin A (Ig A) early in the phase of a dengue infection.
Using an antigen capture anti-DENV IgA (ACA) ELISA technique, we tested saliva samples collected from dengue-confirmed patients. The sensitivity within 3 days from fever onset was over 36% in primary dengue infections. The performance is markedly better in secondary infections, with 100% sensitivity reported in saliva samples from day 1 after fever onset. Serum and salivary IgA levels showed good correlation (Pearson's r = 0.69, p<0.001). Specificity was found to be 97%.
Our findings suggest that this technique would be very useful in dengue endemic regions, where the majority of dengue cases are secondary. The ACA-ELISA is easy to perform, cost effective, and especially useful in laboratories without sophisticated equipment. Our findings established the usefulness and reliability of saliva for early dengue diagnosis.
| The importance of laboratory diagnosis of dengue cannot be undermined. In recent years, many dengue diagnostic tools have become available for various stages of the disease, but the one limitation is that they require blood as a specimen for testing. In many incidences, phlebotomy in needle-phobic febrile individuals, especially children, can be challenging, and the tendency to forgo a dengue blood test is high. To circumvent this, we decided to work toward a saliva-based assay (antigen-capture anti-DENV IgA ELISA, ACA-ELISA) that has the necessary sensitivity and specificity to detect dengue early. Overall sensitivity of the ACA-ELISA, when tested on saliva collected from dengue-confirmed patients (EDEN study) at three time points, was 70% in the first 3 days after fever onset and 93% between 4 to 8 days after fever onset. In patients with secondary dengue infections, salivary IgA was detected on the first day of fever onset in all the dengue confirmed patients. This demonstrates the utility of saliva in the ACA-ELISA for early dengue diagnostics. This technique is easy to perform, cost effective, and is especially useful in dengue endemic countries.
| Dengue is one of the most prevalent mosquito-borne diseases in humans. This disease is best controlled by regular Aedes source reduction activities. However, total eradication of Aedes in a densely populated urban area where the vector has established itself is a daunting task. Dengue control must include prompt control response to dengue clusters, and early and reliable diagnosis of cases is critical to this effort, which aims to halt the DENV transmission. There has been progress in recent years in the development of dengue diagnostic tools, resulting in the availability of suitable tests for each stage of the disease.
Specific detection of dengue viral ribonucleic acid (RNA) using real-time reverse transcription (RT) polymerase chain reaction (PCR) is widely utilized to diagnose and serotype dengue infections in the early phase of the disease [1]–[7]. These techniques, while rapid and effective in providing early dengue diagnosis, are costly and require trained personnel to perform. It is thus only currently available in a limited number of clinical laboratories. The more recent development of DENV non-structural protein 1 (NS1) antigen detection in the Enzyme-linked immunosorbant assay (ELISA) and rapid lateral flow platform has offered clinical laboratories an effective tool for early diagnosis during the febrile phase of the disease [8]–[13].
The detection of anti-DENV immunoglobulin M (IgM) is the most widely used serological assay in dengue diagnosis [14]–[21]. However, anti-DENV IgM is usually detected 5 to 6 d after the onset of fever and thus could result in a delay in diagnosis. Moreover, it can persist for more than 8 mo [20], [22], [23], and in dengue-endemic countries such as Singapore, the detection of IgM in a febrile patient does not necessarily indicate an acute dengue infection. The requirement for analysis of paired samples collected at least 7 d apart, for definitive diagnosis, could delay intervention efforts.
Unfortunately, the necessity of a venous blood collection in all available dengue diagnostic assays and the high cost of the tests that are available for the viraemic period hinder the early detection of cases and clusters. Phlebotomy in needlephobic febrile individuals, especially children, can be challenging, and the tendency to forgo a dengue blood test is high.
We have therefore worked toward saliva-based techniques that could address the early phase of the disease. Saliva is known to be rich in IgA, the concentration of which is 100 times greater than that of IgM and 14 times greater than IgG, and should thus serve as a good target for early diagnosis [24]. Usage of salivary IgG for diagnosis and epidemiological studies has been described before [24]–[26]. The use of serum anti-DENV IgA as a diagnostic marker has previously been explored. Groen et al. [27] described the simultaneous increase of DENV-specific IgA and IgM in dengue patients and reported that IgA was short-lived compared to IgM [27]. An antibody-capture IgA (AAC) ELISA was used. Using the same technique, subsequent studies showed that anti-DENV IgA typically appeared after IgM did and was thus not suitable for dengue diagnostics [15], [23], [28]. The use of salivary IgA for disease detection has also been reported for Human Immunodeficiency Virus, Hepatitis A and B, Measles, Mumps, and Rubella [29]–[33].
In this prospective study, we developed a protocol that allows saliva to be used for anti-DENV IgA detection. The technique, antigen-capture anti-DENV IgA (ACA)-ELISA, not only increased the sensitivity of DENV-specific IgA detection, it also reduced the total test time to 90 min, when compared with a previously published IgA assay.
The Environmental Health Institute (EHI) is a national public health laboratory that functions as a licensed diagnostic laboratory, with an ISO9001 accreditation, as well as a research laboratory.
Three suites of characterized samples, collected in Singapore, were used in this study. WHO criteria for dengue confirmation was adhered to for the determination of dengue status in the following samples: The first (A) comprised saliva and sera collected from 10 healthy volunteers as well as dengue-confirmed patients for optimization of the protocol. The sera from healthy volunteers were previously confirmed to be dengue negative via DENV RT-PCR and PanBio IgM Capture ELISA, and their negative anti-DENV IgA status was ascertained in both saliva and sera using a previously reported DENV Antibody Capture IgA ELISA (AAC-ELISA) [23]. The samples from dengue-confirmed patients consisted of saliva and sera sequentially collected from five patients during the acute phase of their disease. The dengue status of these patients was confirmed by RT-PCR on sera collected in the first 72 h and subsequent sero-conversion as demonstrated by IgM assays. These five sets of saliva and sera samples were also previously confirmed to be DENV IgA positive in AAC-ELISA. Samples in this suite were used as reference samples to establish the ACA-ELISA.
The second suite of samples (B), for evaluation of the newly developed ACA-ELISA protocol, consisted of saliva and sera obtained from 69 DENV-PCR-confirmed patients through three consecutive collections. The first collection was within 72 h after fever onset, second collection around 3 d after first collection, and third collection within 21 d after fever onset. IgM tests, performed on all three collections of each patient, demonstrated sero-conversion of each of the patients, thus confirming their dengue status. Of the 69 patients, 37 (53.3%) had DENV1 infections, 4 (5.8%) had DENV2, and 28 (40.6%) had DENV3, as determined by RT-PCR on the first collection samples [34]. Of the 69 PCR-positive patients from suite B samples, there were 33 primary infections and 36 secondary infections. A primary dengue infection was characterized by first collection serum (first 72 h) being positive for DENV PCR but PanBio Indirect IgG ELISA negative. The absence of DENV-specific IgG in the acute phase of a dengue infection is indicative of primary dengue infection [35]–[37]. A secondary dengue infection was characterized by the first collection sample being concurrently positive for DENV PCR and IgG. These samples were collected through a research project (EDEN) from April 2005 to December 2006 [34] and were used for evaluation of the established ACA-ELISA protocol.
The third suite (C), serving as a specificity test of the ACA-ELISA, comprised three consecutive collections of saliva and sera from 75 DENV PCR-negative febrile patients (EDEN). Collected in the same manner as suite B, they were DENV RT-PCR negative patients, and IgM tests revealed no sero-conversion among all 75 cases.
Oracol saliva collection swab (Malvern Medical Development Lid, UK) was used for saliva collection. A standardized saliva collection protocol was used such that active saliva secretion was obtained. Patients were instructed to swab in a scrubbing manner their inner upper and lower cheeks 10 times each on both sides and place the swab under their tongues for 1 min. Saliva samples, together with blood, were transported on ice, processed within the same day of collection, and stored at −80°C until testing.
Use of samples in suite A was approved by NEA's Bioethics Committee (IRB004.1). Use of samples in suites B and C was approved by the National Healthcare Group Internal Review Board (DSRB B/05/013). Written informed consent was obtained from all participants.
DENV cell culture lysate antigens used in ACA-ELISA were prepared using a DENV 2 strain (SS194Y02) of the Cosmopolitan genotype, isolated from a dengue patient locally, according to the method previously described [38], and viral titre was determined via plague assay [39]. A single batch of cell lysate was prepared and utilized for this entire study.
Checkerboard dilution was performed using anti-DENV IgA positive and negative samples from suite A as reference to optimize the ACA-ELISA for serum and saliva usage. In brief, the 96-well plate maxisorp plates (Nunc, Denmark) were coated with 100 µl/well of pan-DENV monoclonal antibodies (Mab; 1.15 mg/ml; Immunology Consultants Laboratory, USA) diluted at 1∶500 in sodium bicarbonate (pH 9.5) and incubated either overnight at 4°C or 1 h at 37°C. After blocking the plate with dilutent buffer (5% skim milk containing 0.05% Tween-20), virus lysate (2.14×106 pfu/ml) in diluent buffer was added to each well and incubated at 37°C for 1 h. The plate was then washed six times using washing buffer (1X PBS-0.05% Tween-20). Either 100 µl of test serum at 1∶100 in diluent buffer or 100 µl of saliva at 1∶5 dilution was added to each well. In each plate, two positive controls, two negative controls (DENV negative human sera or saliva), and one plate control (no sera or saliva added) were included. The two positive controls were either two anti-DENV IgA positive sera (titre of 1∶256) or two anti-DENV IgA positive saliva samples (titre 1∶10). After 1 h of incubation at 37°C, the plate was washed again six times, and 100 µl of 1∶4000 rabbit anti-human IgA conjugated with horse-radish peroxidase (HRP; Dako, Denmark) was added to each well and then incubated for 30 min at 37°C. Following incubation, the plate was then washed again six times, and 100 µl of tetra-methyl-bencidine (TMB; Sigma, USA) was added to each well and incubated for 5 min at room temperature. Further color development was stopped using 100 µl 0.5 M sulphuric acid, and absorbance was measured at 450 nm against a reference filter at 620 nm. The same batch of controls, reagents, and dengue lysate was used for this entire study. All IgA assays for evaluation of the protocol were performed by a single analyst within a day, without masking the results of the reference tests. Interval between sample collection and testing ranged from 1 mo to 1 y, during which samples were kept frozen at −80°C.
All IgM and IgG assays were performed using PanBio IgM Capture ELISA and PanBio Indirect IgG ELISA. IgM, IgG assays, and RT-PCR (7) were conducted by five trained personnel of the EHI Diagnostics Unit, licensed by the Ministry of Health. While PCR tests were performed within a day of collection of samples, antibody tests on sera and saliva were performed either on the same day or within 1 mo, during which samples were frozen at −80°C.
Data analysis, including calculations of correlation coefficients and standard error of proportion, was carried out using Microsoft Excel 2002 and Statistical Package for Social Sciences (SPSS) 17.0. Sensitivity and its 95% confidence intervals were provided as estimates of the effectiveness of the in-house ACA-ELISA. Standard error of proportion was calculated using the formula √ [p(1−p)/n].
The cutoff point of saliva IgA assays was determined with suite A of saliva samples from healthy volunteers. Mean ± standard deviation optical density (OD) values of the negative controls were determined. A sample was considered negative when the OD value was less than the mean value for the negative control plus two standard deviations, equivocal when the OD value exceeded the mean value for the negative control plus 2 standard deviations but less than 3 standard deviations and considered positive when the OD value is above 3 standard deviations.
Suite B saliva and serum samples, consisting of three consecutive collections from each dengue-confirmed patient, were assayed for anti-DENV IgA using ACA-ELISA. All the serum samples were also tested with PanBio Capture IgM ELISA. Serum and salivary anti-DENV IgA levels showed good correlation (Pearson's r = 0.69; p<0.001). Figure 1 shows the sensitivities of ACA-ELISA on saliva and sera, compared to IgM Capture ELISA on sera. ACA-ELISA on saliva had an overall sensitivity of 70% in the first 3 d after fever onset and subsequently rose to over 90% between 3 and 8 d after fever onset. The same technique on sera gave similar results. More interestingly, the sensitivity of ACA-ELISA in saliva was higher than that of IgM Capture ELISA on sera, which detected only 10% of the dengue-confirmed patients after 1 to 3 d of fever, and only rose to around 90% after day 6 of fever. The specificity of the ACA-ELISA test was also found to be high at 97%. Among the 75 DENV-negative patients in suite C, only one patient tested positive, at day 7 and 27, respectively.
The data of the ACA-ELISA were further analyzed with respect to primary (n = 33) and secondary (n = 36) infections. The sensitivity of the technique is detailed in Figure 2A (primary infection) and 2B (secondary infection). Among the 33 primary cases, the sensitivity of ACA-ELISA on saliva was 36% in the first 3 d and rose to 86% in the second collection (3 to 5 d). The number of samples collected during 6 to 8 d was small. Nevertheless, the results showed that at the early phase of the disease, the sensitivity of ACA-ELISA on saliva was clearly higher than those of ACA-ELISA and IgM Capture on sera (15% and 6% in the first 3 d after fever onset, respectively). Interestingly, ACA-ELISA tested on the saliva and sera of 36 secondary cases yielded sensitivities of 100% and 94%, respectively, in the first 3 d of fever and continued to allow detection at this high rate in second and third collection (3 to 34 d). In contrast, IgM Capture ELISA on sera of secondary cases gave a detection sensitivity of 14% in the first 3 d and rose to 88% only by day 6.
This study has demonstrated the potential use of saliva for early dengue diagnostics. The sensitivity of ACA-ELISA was 70% to 92% within the first 8 d from onset of fever. Data from our diagnostic unit have revealed that dengue patients in local settings visit the primary health care physicians at an average of 3.5 d from onset of fever [13]. Ninety-five percent of the patients would visit their doctors within 0.95 to 6.03 d after fever onset. Under this setting, we have found that our in-house RT-PCR offers 72.5% specificity and the most sensitive commercial NS1 assay offers 81.7% [40]. Therefore, ACA-ELISA on saliva is comparable to these early diagnosis tests and offers the additional advantages of non-intrusive sampling and easy cost-effective laboratory procedures.
It is not surprising that among the secondary dengue infections, IgA was detected in both the saliva and sera of all individuals as early as day 1 after fever onset, while detection in primary cases was delayed. This is due to the presence of IgA memory cells from the previous dengue infection that was triggered by the secondary DENV exposure—not unlike the early IgG responses in secondary dengue cases. This has also been observed in previous studies [15], [25], [41], [42]. Due to the difference in sensitivity between primary and secondary cases, the sensitivity of the ACA-ELISA in a population will be highly dependent on the proportion of secondary cases in the cohort. About 50% of all dengue cases in Singapore are secondary [34]. Under this circumstance, we demonstrated about 70% sensitivity in the first 3 d of fever. In a population with a higher rate of secondary cases, the sensitivity could potentially be higher.
There are two possible explanations to the early detection of IgA even in primary cases. Firstly, inaccuracy in reporting of the onset dates, due to insensitivity to mild fever or bias in recalls, may contribute to the situation. Secondly, intrinsic incubation periods and response time vary among individuals. Dengue patients are infected 2 to 14 d before fever onset, a period which may have allowed IgA (and IgM) production in some individuals. Early detection of IgM for some individuals is also evident for dengue and chikungunya [43], [44]. It is highly likely that in the very early phases of dengue, IgM and IgA are present in very low levels, and time of detection lies in the sensitivity of the technique used.
Previous studies, using AAC-ELISA, reported low sensitivities when using salivary IgA, in contrast to our findings [15], [24], [25], [45]. Two strategies were designed in this study to overcome the limitation seen in previous studies. Firstly, saliva collection protocol in this study was designed to allow for the collection of actively secreted saliva. Previous studies on DENV-specific IgA in saliva used passively secreted saliva in their evaluation. Prior to this study, a comparison of the two saliva collection protocols had revealed that, in the same individuals, actively collected saliva yielded higher levels of DENV-specific IgA than passively collected ones (unpublished data, Yap G). Secondly, ACA-ELISA was designed to eliminate binding competition from high levels of non-specific IgA that are normally present in mucosal secretions to protect one from infectious diseases. A previously published technique, AAC-ELISA, captured all IgA in the first step, followed by subsequent differentiation of DENV-specific IgA from the pool of IgAs. In the event of a dengue infection, anti-DENV IgA may represent only a small proportion of IgA in saliva. The low detection rate of anti-DENV IgA could be due to preexisting non-DENV-specific IgA out-competing anti-DENV IgA. To circumvent the limitation, ACA-ELISA was designed to capture all anti-DENV antibodies in its first step, followed by detection of anti-DENV IgA. In the early phase of a primary dengue infection, DENV-specific IgM and IgG are present in low levels, and coupled with the enrichment of IgA in saliva, the approach is expected to increase the detection sensitivity. The high sensitivity of the technique could also explain the observations that in the early phases of the disease in primary cases, DENV IgA could be detected earlier than DENV IgM, appearing to go against classical immunology. The capture IgM ELISA used in this study, like the AAC-ELISA, was expected to pick up only elevated levels of DENV IgM. The difference thus lies on the sensitivity of the techniques used. The slightly higher sensitivity in saliva compared to that in serum is likely due to the dimeric structure of the secretory IgA, which could increase the amplification of signal output from the ELISA.
Even though the antigen used in ACA-ELISA of this study was DENV2 and not a tetravalent antigen, the ACA-ELISA is effective in detecting IgA illicited by the three serotypes circulating in Singapore during the study period, as DENV IgA, like IgM, is cross-reactive with all four serotypes [46]. This is supported by a study that demonstrated no significant differences in the sensitivity of ACA-ELISA when DENV2 is replaced by a tetravlent antigen (unpublished data, Yap G). A multi-country study is ongoing to evaluate the test in various epidemiology settings, particularly to establish its performance in settings with other circulating DENV genotypes and other diseases that illicit cross-reacting antibodies, which may impact its performance. Potential limitation in specificity can be circumvented using recombinant antigen specific to DENV.
This study suggests the potential of the saliva ACA-ELISA for dengue diagnosis. It eliminates the need to collect blood from dengue-suspected patients, is painless, is non-intrusive, and reduces the risk of needle stick injury. Moreover, the ELISA-based technique is simple and cost effective. Patients, especially the very young and the old, will be more willing to undergo a dengue test. Together, these benefits can potentially improve surveillance and early detection of cases, which in turn can translate to prompt dengue control effort. Due to its high sensitivity among secondary dengue infections, this technique could be very useful in highly endemic areas where the majority of the dengue cases are secondary.
|
10.1371/journal.pgen.1005790 | The LSH/DDM1 Homolog MUS-30 Is Required for Genome Stability, but Not for DNA Methylation in Neurospora crassa | LSH/DDM1 enzymes are required for DNA methylation in higher eukaryotes and have poorly defined roles in genome maintenance in yeast, plants, and animals. The filamentous fungus Neurospora crassa is a tractable system that encodes a single LSH/DDM1 homolog (NCU06306). We report that the Neurospora LSH/DDM1 enzyme is encoded by mutagen sensitive-30 (mus-30), a locus identified in a genetic screen over 25 years ago. We show that MUS-30-deficient cells have normal DNA methylation, but are hypersensitive to DNA damaging agents. MUS-30 is a nuclear protein, consistent with its predicted role as a chromatin remodeling enzyme, and levels of MUS-30 are increased following DNA damage. MUS-30 co-purifies with Neurospora WDR76, a homolog of yeast Changed Mutation Rate-1 and mammalian WD40 repeat domain 76. Deletion of wdr76 rescued DNA damage-hypersensitivity of Δmus-30 strains, demonstrating that the MUS-30-WDR76 interaction is functionally important. DNA damage-sensitivity of Δmus-30 is partially suppressed by deletion of methyl adenine glycosylase-1, a component of the base excision repair machinery (BER); however, the rate of BER is not affected in Δmus-30 strains. We found that MUS-30-deficient cells are not defective for DSB repair, and we observed a negative genetic interaction between Δmus-30 and Δmei-3, the Neurospora RAD51 homolog required for homologous recombination. Together, our findings suggest that MUS-30, an LSH/DDM1 homolog, is required to prevent DNA damage arising from toxic base excision repair intermediates. Overall, our study provides important new information about the functions of the LSH/DDM1 family of enzymes.
| Inside cells, eukaryotic DNA exists in a highly packaged structure called chromatin. Chromatin packaging often inhibits enzymes that need to access the genetic information. It is therefore important for cells to regulate chromatin structure so that the genome can function properly. Mammalian LSH (Lymphoid-specific helicase; also known as HELLS, PASG, and SMARCA6) and Arabidopsis DDM1 (Decreased DNA methylation 1) are the founding members of the LSH/DDM1 subfamily of ATP-dependent chromatin remodelers. In mammals, the LSH enzyme is required for normal development, as well as oogenesis, spermatogenesis and T-lymphocyte proliferation. Similarly, the plant protein is required for development, and both proteins are important for regulating levels of DNA methylation, an important epigenetic mark. Recent studies suggest that LSH and DDM1 are also critical for genome integrity, but their precise functions are not understood. We have carried out genetic, genomic, and proteomic analyses to investigate an LSH/DDM1 homolog in a tractable model eukaryote, Neurospora crassa. We report that the Neurospora protein works in concert with the homologous recombination machinery to maintain genome stability. Our data provide important new information about the LSH/DDM1 family of enzymes.
| Many chromatin-based processes require the activity of ATP-dependent chromatin remodeling enzymes, which can alter local chromatin structure by repositioning, removing, or restructuring nucleosomes [1–3]. Mammalian LSH (Lymphoid-specific helicase; also known as HELLS, PASG, and SMARCA6) and Arabidopsis DDM1 (Decreased DNA methylation 1) are the founding members of the LSH/DDM1 subfamily of ATP-dependent chromatin remodelers–one of 24 subfamilies that comprise the larger SNF2 enzyme family [4, 5]. In vitro, DDM1 is able to hydrolyze ATP and reposition nucleosomes on a short DNA template, demonstrating that the LSH/DDM1 subfamily includes bona fide chromatin remodeling enzymes [6]. Moreover, molecular and genetic studies have implicated LSH and DDM1 in a number of important cellular processes.
Lsh was originally identified as lymphocyte-specific; however, the gene is ubiquitously expressed in mammals [7–9]. In particular, high levels of Lsh are found in proliferating cells, suggesting that the protein might function during DNA synthesis or cell division. Subsequent studies revealed that Lsh is essential for development. Mice bearing homozygous deletions of Lsh die within 24 hours of birth, reportedly due to a host of developmental defects [8, 10]. Additional studies in chimeric mice or with tissue explants revealed that LSH is essential for both male and female meiosis [11, 12], as well as for proliferation of T-lymphocytes [13]. Thus, LSH is essential for gametogenesis and for proper development of the immune system. Notably, LSH has also been implicated in cancer [7, 14–18]. An in-frame Lsh deletion in the putative catalytic domain is frequently identified in human leukemias [7], and transplantation of hematopoietic precursors from Lsh -/- mice produced abnormal hematopoiesis and elevated rates of erythroleukemia in recipients [14]. Despite its role in these important processes, the molecular functions of LSH are not well understood.
Lsh mutant mice exhibit significantly reduced DNA methylation (5mC) at many sites in the genome [19–26]. Similarly, Arabidopsis thaliana ddm1 mutants display reduced DNA methylation and developmental defects, suggesting that at least some LSH/DDM1 functions are conserved across eukaryotic kingdoms [27–33]. Recently, studies in both plants and animals have uncovered a role for LSH/DDM1 in maintenance of genome stability. Arabidopsis DDM1-deficient mutants are hypersensitive to a variety of DNA damaging agents, including MMS (methyl methanesulfonate) [34, 35]. Similarly, mammalian Lsh-/- cells are hypersensitive to DNA damage and are unable to mount a robust DNA damage response [36]. There is some controversy regarding the relationship between the DNA methylation and DNA damage phenotypes of LSH/DDM1-deficient cells. The DNA damage-sensitivity phenotype of ddm1 plants was proposed to be an indirect effect of DNA hypomethylation [35], whereas in animals, stable knockdown of Lsh in immortalized lung fibroblasts led to hypersensitivity to DNA damage before a reduction in DNA methylation levels was observed [36]. Notably, an LSH homolog was also implicated in genome maintenance in Saccharomyces cerevisiae, an organism that lacks DNA methylation. The yeast gene, named IRC5 (Increased repair centers-5), was uncovered in a high throughput screen for deletion strains that accumulate spontaneous DNA repair foci [37]. Thus, LSH-family enzymes are important for genome stability in fungi, plants, and animals, but precisely how LSH/DDM1 homologs control DNA methylation or genome stability is not clear.
LSH-family members are absent from several model systems including Drosophila melanogaster, Caenorhabditis elegans, and Schizosaccharomyces pombe [5], but the model fungus Neurospora crassa encodes a single LSH/DDM1 homolog (NCU06306; also called Chromatin Remodeling Factor 5) [5, 38]. N. crassa is a particularly attractive model for studies of chromatin structure and function because its complement of chromatin modifications and chromatin-associated proteins is similar to higher eukaryotes. For example, hallmarks of heterochromatin such as histone H3 lysine-9 methylation (H3K9me3), Heterochromatin protein-1 and DNA methylation are shared between Neurospora and higher eukaryotes, but are all absent from S. cerevisiae [39–42]. To gain insights into the functions of the LSH/DDM1 subfamily, we performed molecular, genetic and genomic analyses to investigate N. crassa NCU06306/CRF5. We found that this LSH/DDM1 homolog is not required for DNA methylation, but is essential for survival from DNA damage. ncu06306/crf5-1 is allelic to the previously described mutagen sensitive-30 (mus-30). The encoded protein is localized to the nucleus and interacts with WDR76, a conserved WD40 domain-containing protein. Based on genetic interactions with known DNA repair components, we propose that the Neurospora LSH/DDM1 homolog functions to limit genome instability resulting from toxic base excision repair intermediates.
Neurospora encodes a single LSH/DDM1 homolog encoded by NCU06306 and given the name Chromatin Remodeling Factor 5 (CRF5) based on its predicted coding sequence [38]. Like LSH and DDM1, NCU06306/CRF5 contains a characteristic SNF2 motor domain made up of an N-terminal SNF2_N DEAD box helicase domain and a C-terminal HelicC domain, but lacks other conserved domains. To determine if NCU06306/CRF5 is important for DNA methylation in N. crassa, we first performed Southern blot analysis to examine DNA methylation levels at two well-studied methylated regions (8:A6 and 8:G3) [43]. DNA methylation levels were similar to wildtype at both regions (Fig 1A). In plant ddm1 mutants, loss of DNA methylation is gradual; 5mC levels progressively decline when ddm1 homozygous mutants are inbred for multiple generations [27]. We therefore performed MethylC-seq to examine genome-wide DNA methylation levels in f1 and f2 progeny derived from homozygous crosses of Δncu06306/crf5-1 parents. We note that NCU06306/CRF5 was not required for meiosis, in contrast to mammalian LSH. As controls, we performed methylC-seq for wildtype and dim-2 (defective in methylation-2), which lacks DNA methylation [44]. We identified methylated regions in Neurospora by calling differentially methylated regions (DMR) between wildtype and a fully unmethylated genome (generated in silico; see Methods), and we calculated the average weighted methylation level for all 5mC regions in wildtype and Δncu06306/crf5-1 isolates from f1 and f2 generations (Fig 1B and S1 Table). Average methylation levels in the Δncu06306/crf5-1 isolates were not statistically different from wildtype. We next constructed metaplots to examine the distribution of 5mC across all methylated regions for wildtype and Δncu06306/crf5-1 isolates from f1 and f2 generations (Fig 1C). The methylation profile of the Δncu06306/crf5-1 strain was similar to wildtype for both strains. These data suggest that NCU06306/CRF5 does not control the levels or the distribution of 5mC within normally methylated regions, in contrast to LSH/DDM1 enzymes in higher eukaryotes (S1 Table). Finally, we used DMR analysis to compare 5mC regions in wildtype and Δncu06306/crf5-1 strains. We identified twenty-two regions with subtle changes in the level of methylation in one or more Δncu06306/crf5-1 isolates. However, these subtle differences likely represent sequence polymorphisms between the mat A reference strain and the mat a strain (S2 Table). Together, these data demonstrate that NCU06306/CRF5 is not required for normal DNA methylation in N. crassa.
We next asked if Δncu06306/crf5-1 is required for survival from DNA damage, as proposed for other LSH/DDM1 homologs. We examined growth of Δncu06306/crf5-1 in the presence of several DNA replication and DNA repair inhibitors (Fig 1D). Δncu06306/crf5-1 cells were not hypersensitive to UV light or Hydroxyurea, which inhibits ribonucleotide reductase [45]. Similarly, Δncu06306/crf5-1 cells displayed wildtype resistance to Bleomycin, which is thought trigger double strand breaks [46], and only displayed limited sensitivity to the topoisomerase I inhibitor camptothecin (CPT) [47]. In contrast, Δncu06306/crf5-1 cells were unable to grow on medium containing methyl methanesulfonate (MMS; 0.025%), which can collapse replication forks, leading to double strand breaks [48, 49]. Δncu06306/crf5-1 were also hypersensitive to oxidative damage by tert-Butyl hydroperoxide [50].
Knockout strains have been shown to accumulate second-site mutations [51]. To confirm that the DNA damage-hypersensitive phenotype of Δncu06306/crf5-1 is caused by deletion of the ncu06306/crf5-1 gene (NCU06306), we introduced a wildtype copy of ncu06306/crf5-1+ into the deletion strain and tested for growth on MMS. Wildtype ncu06306/crf5-1+ restored growth, confirming that ncu06306/crf5-1 is required for survival from MMS-induced DNA damage (Fig 1E).
A previous screen for mutagen sensitive strains led to the identification of mutagen-senstive-30 (mus-30), which had been mapped to a region on LGIV that includes the ncu06306/crf5-1 gene [52]. Like Δncu06306/crf5-1, mus-30 FK115 is sensitive to MMS, but not to HU or UV light [52]. To test the possibility that ncu06306/crf5-1 and mus-30 are allelic, we sequenced the ncu06306/crf5-1 gene from the original mus-30FK115 isolate. The mus-30 FK115 strain contains a single base change in the ncu06306/crf5-1 locus, which is predicted to produce an Arginine to Proline substitution at position 809. This mutation is within the predicted HelicC domain [53]. We next tested for complementation in heterokaryons of Δncu06306/crf5-1 and mus-30FK115. The mus-30 FK115 strain was transformed with a basta-resistance cassette to allow construction of forced heterokaryons. Six heterokaryons of mus-30FK115;:: bar+ and Δncu06306/crf5-1::hph+ were generated from individual basta-resistant transformants and maintained on medium containing both hygromycin and basta. To test for complementation, condia were spotted on medium containing hygromycin, basta, or MMS (0.015%, 0.020%, and 0.025%). Control heterokaryons were constructed by mixing conidia of the basta-resistant, MMS-sensitive Δdim-5 strain with conidia from the hygromycin-resistant, MMS-sensitive Δncu06306/crf5-1 strain. Representative heterokaryons are shown in Fig 2A. All control heterokaryons [Δncu06306/crf5-1::hph+ + Δdim-5::bar+] were able to grow on medium containing MMS, demonstrating that the MMS-sensitivity phenotypes of Δncu06306/crf5-1 and Δdim-5 strains are recessive. In contrast, none of the heterokaryons of mus-30 FK115 and Δncu06306/crf5-1 were able to grow on medium containing MMS (six independent heterokaryons were tested), suggesting that ncu06306/crf5-1 and mus-30 are allelic.
To confirm this, we introduced a wildtype copy of the ncu06306/crf5-1 gene into the mus-30 FK115 strain by co-transformation with a basta-resistance cassette. A fraction of basta-resistant transformants are expected to integrate the ncu06306/crf5-1 sequence along with the basta-resistance cassette. Of 40 basta-resistant transformants tested, 15 displayed robust growth in the presence of MMS (Fig 2B and S1 Fig). No MMS-resistant transformants were obtained when mus-30 FK115 was transformed with the basta-resistance cassette alone. Together, these data demonstrate that ncu06306/crf5-1 is allelic to mus-30. We hereafter refer to NCU06306/CRF5 as MUS-30.
MUS-30 is predicted to function as a chromatin remodeler and is therefore expected to localize to the nucleus. To test this, we constructed a GFP-tagged version of MUS-30 using a standard “knock-in” approach [54]. GFP coding sequence was integrated by homologous recombination into the 3’ end of the mus-30 gene. Primary transformants were backcrossed to isolate homokaryons and individual mus-30-gfp strains were tested for growth on MMS to confirm that the GFP fusion construct was functional (S2 Fig). In live cells, MUS-30-GFP was localized to the nucleus, consistent with its predicted role as a chromatin remodeling enzyme (Fig 3A). Some DNA repair proteins alter their localization in response to DNA damage. We treated cells with MMS for three hours and then examined the localization patterns of MUS-30-GFP before, during, and after MMS treatment. A diffuse nuclear localization pattern was observed in the presence and absence of MMS. However, we detected an increase in overall fluorescence in some experiments, suggesting that MUS-30 protein levels may be increased in response to DNA damage (Fig 3A). To determine if MUS-30 protein levels are increased in MMS treated cells, we constructed a FLAG-tagged version of MUS-30 and performed Western blot analysis. The mus-30-3xflag strain was able to grow on MMS, indicating that the tagged version of the protein was functional (S2 Fig). Total protein isolated from wildtype and the mus-30-3xflag strain grown in minimal medium and subjected to Western blotting with anti-flag antibodies. We detected an ~106kD protein in extracts from the mus-30-3xflag strain, consistent with the predicted size of the MUS-30-3XFLAG fusion protein (Fig 3B). We next compared the level of MUS-30-3XFLAG expression in minimal medium and in the presence of MMS. MUS-30-3XFLAG levels were higher in MMS-containing medium. Under certain gel conditions, the FLAG antibody detected two bands, raising the possibility that MUS-30 is post-translationally modified. Phosphorylation of proteins is often associated with signaling in response to DNA damage [55]. To determine if MUS-30 is phosphorylated, we resolved protein extracts from the mus-30-3xflag strain on a Phos-Tag gel, which reduces the mobility of phosphorylated proteins (Fig 3C) [56]. We observed a shift in mobility of MUS-30-3XFLAG under all conditions examined. Treating extracts with lambda phosphatase eliminated the slower migrating form of the protein, suggesting that MUS-30-3XFLAG is indeed phosphorylated but that phosphorylation does not occur specifically in response to DNA damage.
Genetic interactions can provide insights into gene function. Positive genetic interactions often indicate that the products of the interacting genes function in the same pathway, whereas negative interactions suggest that two gene products perform compensatory functions in separate pathways [57]. Positive genetic interactions occur when the fitness of a double mutant is better than the expected phenotype. For example, combining two mutations that cause MMS-sensitivity is expected to produce a double mutant that has a higher level of MMS sensitivity than either single mutant. In contrast, a negative genetic interaction occurs when the fitness of the double mutant is worse than the phenotype expected from combining the two single mutant phenotypes. Genetic interaction analysis has been used extensively to place Neurospora DNA repair mutants into epistasis groups [58]. For example, members of the uvs-6 epistasis group exhibit positive genetic interactions with one another and encode components of the homologous recombination repair pathway [59]. Δmus-30 strains are highly sensitive to MMS, which generates methylated bases that can stall replication forks and indirectly lead to DSBs [48, 49, 60–63]. The primary mechanism for repair of MMS-induced damage is base excision repair (BER). We therefore crossed Δmus-30 to Δmag-1, a putative BER glycosylase that removes methyl-adenine bases generated by MMS [38], and we determined the level of MMS sensitivity in wildtype, single mutant, and double mutant progeny. We observed a positive genetic interaction between Δmus-30 and Δmag-1 (Fig 4A and 4B). The Δmus-30; Δmag-1 double mutant was more tolerant to MMS than Δmus-30 single mutants, exhibiting a level of sensitivity that was similar to the Δmag-1 strain. Consistent with its predicted role as a methyl-adenine glycosylase, Δmag-1 did not suppress the TBH-hypersensitive phenotype of Δmus-30 strains (S3 Fig). Taken together, these data could indicate that MUS-30 functions in the BER pathway downstream of base removal. However, it has been shown that BER intermediates are themselves mutagenic [49, 62–65]. Therefore, these data could also indicate that MUS-30 is required to prevent or repair DNA damage that arises when replication forks encounter toxic BER intermediates.
To distinguish between these possibilities, we asked if Δmus-30 was able to repair alkylated bases at a rate that was similar to wildtype. Cells were exposed to MMS and harvested during and after MMS exposure. To monitor repair of MMS-induced damage, genomic DNA was treated with recombinant BER enzymes to generate single-strand breaks at MMS-damaged bases and abasic sites [66]. The DNA was then resolved by alkaline gel electrophoresis to examine kinetics of repair; unrepaired DNA that contained alkylated bases or abasic sites runs as a low molecular weight smear. As expected, Δmag-1 failed to repair MMS-damaged bases, consistent with its predicted role as a BER glycosylase. Both wildtype and Δmus-30 were able to remove and repair alkylated bases with similar kinetics (Fig 4C). These data demonstrate that MUS-30 is not required for BER, but is likely important for preventing or repairing DNA damage that results from toxic BER intermediates.
It was previously reported that mammalian LSH was required for efficient double strand break repair (DSB) [36]. We asked if MUS-30 influenced how double-strand breaks are repaired in N. crassa using an established transformation assay. Ectopic DNA sequences are inserted into the N. crassa genome via the homologous recombination (HR) or non-homologous end joining (NHEJ) DSB repair pathways [67, 68]. We transformed wildtype or Δmus-30 strains with a basta-resistance cassette flanked by 5’ and 3’ sequences corresponding to the methyltryptophan resistance (mtr) locus. Cells that undergo cassette integration by HR are resistant to basta and to Fluorophenylalanine (FPA), whereas cells that undergo non-homologous integration are resistant to basta, but not FPA [67]. As expected, cassette integration occurred exclusively by HR in Δmus-52 or Δmus-53, which lack required NHEJ components, while no HR events occurred in a Δmei-3 control strain. mei-3 encodes the N. crassa homolog of yeast RAD51 and is required for DSB repair via homologous recombination (HR) [67, 69]. The frequency of homologous integration in the Δmus-30 strain was similar to wildtype, suggesting that MUS-30 is not required for HR or for NHEJ in Neurospora (Table 1). Furthermore, the transformation efficiency was similar in all strains tested, with the exception of Δmus-53, which showed a reduction in transformation efficiency as reported previously [67]. These data demonstrate that MUS-30 is not required for general DSB repair in N. crassa.
MMS-induced damage and BER intermediates can lead to collapsed replication forks that can be restarted in a RAD51-depdendent manner [49, 60–65, 70–73]. To test if MUS-30 prevents replication fork collapse or facilitates MEI-3-dependent replication fork restart, we tested for genetic interactions between Δmus-30 and Δmei-3. We reasoned that a positive genetic interaction would suggest that MUS-30 facilitates MEI-3 dependent replication fork restart, whereas a negative genetic interaction could suggest MUS-30 is required to prevent collapsed forks at MMS-damaged bases and BER intermediates. We observed a striking synthetic growth defect for Δmus-30; Δmei-3 double mutants, which was evident even in the absence of exogenous DNA damaging agents. We performed race tube analysis with multiple isolates of each genotype to quantify the linear growth rate. The growth rates of wildtype, Δmus-30, and Δmei-3 were similar, whereas all isolates of Δmus-30; Δmei-3 double mutants displayed markedly slower growth (Fig 4D and 4E). We next tested the level of MMS-sensitivity for each genotype (Fig 4F). Δmus-30 and Δmei-3 single mutants were unable to grow in the presence of 0.015% and 0.010% MMS, respectively, while Δmus-30; Δmei-3 conidia failed to grow on the lowest MMS concentration tested (0.0001%). Thus, MEI-3 is critical for repairing DNA damage that accumulates in the Δmus-30 mutant strain. Together, these data demonstrate that MUS-30 is not generally required for DSB repair and suggest that MUS-30 is important for preventing DNA damage that arises from toxic base excision repair intermediates.
Many chromatin remodeling proteins exist in multi-subunit complexes. To gain insights into the biochemical function of MUS-30, we sought to identify MUS-30-interacting proteins using a proteomics approach. We used antibodies that recognize the FLAG epitope to purify MUS-30-3XFLAG and identified co-purified proteins by mass spectrometry. As negative controls, we performed a mock purification from the wildtype strain, which does not express a FLAG-tagged protein, and we performed purifications of two components of the previously described DCDC complex, DIM-5-3XFLAG and DIM-9-3XFLAG [74]. To eliminate background hits from our list of putative MUS-30-interacting proteins, we removed proteins that were identified in the “mock” sample (no FLAG-tagged protein), the DIM-5-3XFLAG sample, or DIM-9-3XFLAG sample, and we removed proteins that were identified by a single unique peptide hit (i.e. only proteins identified by two or more unique peptides passed the filter). Purification of MUS-30-3XFLAG in buffer containing 250 mM KCl failed to identify any specific interacting proteins, suggesting that MUS-30 does not exist in a stable multi-subunit complex. However, purification of MUS-30-3XFLAG from protein extracts made with buffer containing 150 mM or 200 mM NaCl led to the identification of a protein containing a WD40 domain (NCU09302) (Fig 5A). BLAST searches using the NCU09302 protein sequence identified yeast Cmr1 (Changed Mutation Rate 1; NP_010125.1) and human WDR76 (NP_079184.2) as putative homologs. Yeast Cmr1 and mammalian WDR76 localize to a sub-nuclear compartment in response to DNA damage, and these structures were shown to be distinct from DSB repair foci [75, 76]. To confirm the interaction between MUS-30 and NCU09302, we constructed a FLAG-tagged version of NCU09302 and performed FLAG affinity purifications in buffer containing 200 mM NaCl and in buffer containing 250 mM KCl. Analysis of both purified fractions by mass spectrometry identified MUS-30, confirming that NCU09302 and MUS-30 interact in vivo. We refer to NCU09302 as WDR76 based on similarity to the mammalian protein. We note that core histones were identified following purification of both MUS-30-3XFLAG and WDR76-3XFLAG, consistent with the predicted role of MUS-30 as a chromatin remodeling enzyme; however, these proteins were removed by our filter because they were also identified by purification of DCDC.
We performed several experiments to examine the functional role of the WDR76-MUS-30 interaction. We first asked if WDR76-3XFLAG is activated by phosphorylation in response to DNA damage. WDR76-3XFLAG was resolved on a Phos-tag gel and Western blots were performed using anti-FLAG antibodies. No change in mobility was observed, indicating that WDR76 is not phosphorylated in response to MMS-induced DNA damage (S4A Fig). We also examined phosphorylation of MUS-30-3XFLAG in the Δwdr76 background and observed no change in MUS-30-3XFLAG phosphorylation (S4B Fig). We next asked if WDR76 associated with chromatin and if chromatin association was altered when MUS-30 was absent. We isolated the soluble and chromatin fraction from the wdr76-3xflag and wdr76-3xflag; Δmus-30 strains grown in the absence or presence of MMS. WDR76-3XFLAG was detected in both the soluble and chromatin fractions in the wildtype and the Δmus-30 background (S4C Fig). Chromatin association was not affected by DNA damage. We performed a similar experiment to examine chromatin association of MUS-30-3XFLAG in the wildtype and Δwdr76 strain background (S4D Fig). In both the presence and absence of DNA damage, MUS-30-3XFLAG was detected in the soluble and chromatin fractions in both strains. These data suggest that WDR76 and MUS-30 do not depend on one another in order to associate with chromatin. Finally, we asked if WDR76 impacted DSB repair by transforming a mtr::basta cassette in the Δwdr76 and Δmus-30; Δwdr76 double mutants (Table 1). In both strains the relative frequency of integration by HR or NHEJ and the transformation efficiency was similar to wildtype.
To confirm that interaction between MUS-30 and WDR76 is functionally important in vivo, we crossed Δmus-30 to Δwdr76 and tested for genetic interactions. The growth phenotype and the level of MMS-sensitivity were examined for wildtype, single mutant, and double mutant progeny. Both Δwdr76 and Δwdr76; Δmus-30 strains displayed wildtype growth on minimal medium, similar to Δmus-30. Notably, a positive genetic interaction was observed for Δmus-30 and Δwdr76 in the presence of MMS (Fig 5B and 5C), confirming that the two gene products interact functionally.
Chromatin remodelers can impact genome maintenance by regulating specific types of DNA repair, facilitating DNA replication, and enhancing propagation of DNA damage signals [3]. LSH/DDM1 homologs have been implicated in genome maintenance from yeast to humans, but how these proteins contribute to genome maintenance is not understood. Plant ddm1 mutants are hypersensitive γ-radiation, UV-light, and MMS [34, 35]. Similarly, Lsh-/- cells are hypersensitive to a number of DNA damaging agents and display muted induction of γH2A.X as well as diminished recruitment of γH2A.X-binding proteins following DNA damage. Based on these observations, it was concluded that LSH promotes efficient DSB repair. [36]. Our study provides additional evidence that LSH/DDM1 proteins are key regulators of genome stability and provides new insights into the role of an LSH/DDM1 family member in genome maintenance.
We propose that N. crassa MUS-30 plays an important role in preventing genome instability when replication forks encounter toxic base excision repair intermediates. This idea is supported by our findings that: 1) deletion of mag-1 can partially rescue the MMS-sensitivity of Δmus-30 strains, 2) Δmus-30 and Δmei-3 interact genetically, 3) MUS-30 is not required for normal BER or DSB repair, and 4) MUS-30 interacts with WDR76. It is not known if other LSH/DDM1 enzymes in other systems act to maintain genome stability independently of DSB repair, but data from yeast and animals are compatible with the idea. A high throughput study in yeast found that irc5Δ strains accumulate spontaneous Rad52-GFP foci in the absence of exogenous DNA damage and exhibit elevated recombination rates with non-sister chromatids [37]. In animals, Lsh expression is highest in proliferating tissues and was correlated with the onset of S-phase [7–9, 13]. Moreover, Burrage and colleagues showed that DNA damage in LSH-deficient cells triggers normal cell cycle arrest, followed by rapid cell death once S-phase resumes [36]. Thus, it is possible that mammalian LSH and yeast Irc5 function during S-phase to prevent collapsed replication forks at specific types of DNA lesions.
Our protein interaction studies provide additional evidence supporting a conserved role for LSH/DDM1 in different systems. We found that N. crassa MUS-30 interacts with a well conserved protein, WDR76. Not only is the WDR76 protein conserved in fungi and animals, its interaction with LSH/DDM1 family members appears to be conserved across species. Proteomic analysis of Cmr1, the yeast WDR76 homolog, identified Irc5p as a putative Cmr1-interacting protein [77]. In addition, while this manuscript was in preparation, it was reported that mammalian LSH co-purifies with WDR76 [76]. Our observation that mus-30 and wdr76 interact genetically provides compelling evidence that physical interaction of MUS-30 and WDR76 is functionally important. Although the specific functions of WDR76 and its homologs are unknown, it was recently reported that both Cmr1 and mammalian WDR76 form DNA damage-dependent foci that are distinct from DSB repair centers [75, 76]. Thus, the interaction between WDR76 and MUS-30 provides additional evidence that MUS-30 is not directly involved in DSB repair. Interestingly, in the presence of the replication inhibitor HU, cmr1Δ exhibits positive genetic interactions with gene deletions of replication fork protection components [76]. We found a similar positive genetic interaction between mus-30 and wdr76 in the presence of MMS. These data could indicate that WDR76 somehow acts to destabilize stalled replication forks. Yeast Cmr1 localizes to a unique sub-nuclear compartment that was hypothesized to promote protein degradation, consistent with this possibility [76]. Alternatively, it was proposed that yeast Cmr1 negatively regulates the DNA damage response [76]. WDR76 may target MUS-30 and other components of the DNA damage response for degradation. Increased activity of other DNA repair components in the Δwdr76 strain could explain why the Δmus-30 phenotype is rescued by the wdr76 deletion.
In plants and animals, LSH and DDM1 proteins have been extensively investigated for their role in regulating DNA methylation. It was suggested that changes in 5mC may lead to differential expression of DNA repair genes in A. thaliana ddm1 mutants [35]. In contrast, it was proposed that mammalian LSH controls DNA repair and DNA methylation through distinct mechanisms [36]. Indeed, knock down of Lsh-knockdown caused hypersensitivity to DNA damage prior to methylation loss, demonstrating that loss of DNA methylation is not indirectly responsible for the hypersensitivity to DNA damage. Here, we found normal DNA methylation levels in mus-30 strains by comprehensive MethylC-seq, clearly demonstrating that loss of 5mC does not drive DNA damage-sensitivity in Δmus-30 strains. It remains possible, however, that loss of DNA methylation in LSH/DDM1-deficient cells results in part from defective DNA repair functions. Future work is needed to fully understand how LSH/DDM1 family members function to regulate DNA methylation and contribute to genome stability.
All Neurospora strains used in this study are listed in S3 Table. Knockout strains were generated by the Neurospora gene knockout consortium [78] and obtained from the Fungal Genetics Stock Center [79]. Strains were grown at 32°C in Vogel's minimal medium (VMM) + 1.5% sucrose. Crosses were performed on modified synthetic cross medium [80]. For plating assays, Neurospora conidia were plated on VMM with 2.0% sorbose, 0.5% fructose, and 0.5% glucose. When relevant, plates included 200 μg/mL hygromycin or 400 μg/mL basta [81] or DNA damaging agents at the indicated concentration. For MMS survival curves, 200 cells were plated on minimal medium and on plates with increasing concentrations of MMS. The number of colonies was counted for each plate and plotted as a percentage of the no MMS control. At least two independent strain isolates were used for each concentration of MMS and at least three independent plating assays were performed to determine the average percent viability. Error bars depict standard deviation from the mean. Neurospora transformation [82], DNA isolation [83], protein isolation, and Western blotting [84] were performed as previously described. We performed affinity purification using M2 FLAG affinity gel (cat # A2220; Sigma-Aldrich). To separate soluble nuclear proteins from the chromatin fraction, cells were grown overnight and either left untreated or exposed to 0.015% MMS for three hours. Cells were then collected, ground in liquid Nitrogen, and resuspended in 1 mL of low salt extraction buffer (50mM HEPES-KOH pH 7.5, 150mM NaCl, 2mM EDTA, plus protease inhibitor tablets (Roche, Indianapolis, IN)). Extracts were centrifuged at 14,000 rpm and the supernatant containing soluble proteins was saved. The pellet was resuspended in 1mL of high salt extraction buffer (50mM HEPES-KOH pH 7.5, 600mM NaCl, 2mM EDTA, plus protease inhibitor tablets (Roche, Indianapolis, IN)) and subjected to sonication. Extracts were centrifuged at 14,000 rpm in a microfuge and the supernatant containing was saved as the chromatin fraction.
Protein identification by mass spectrometry was performed at the Oregon Health Sciences University proteomics core facility (Dr. Larry David) as described previously [85] except for the following modifications. Extraction Buffer (50 mM HEPES-KOH pH 7.6, 2 mM EDTA, 10% glycerol, 2 mM DTT, protease inhibitor cocktail (Roche, Indianapolis, IN)) contained either 150 mM NaCl, 200 mM NaCl, or 250 mM KCl as indicated. Total immunoprecipitated protein was run into an SDS-PAGE gel and a single band containing all immunoprecipitated proteins was excised, subjected to in-gel trypsin digestion, and analyzed on a Thermo LTQ Velos Pro linear ion trap instrument. 3X-FLAG and–GFP knock-in constructs were made by introduction of linear DNA fragments constructed by overlapping PCR using described plasmid vectors [54]. All primers, including primers for generating knock-in constructs and for amplifying and sequencing the NCU06306 gene from the mus-30 strain, are listed in S4 Table. To analyze protein phosphorylation, FLAG-immunoprecipitated proteins were resolved on a modified 5% acrylamide gel containing 25 μM Phos-Tag (cat # 304–93526, Waka Pure Chemical Industries) before Western blotting [56].
Cells were grown for 11 hours in liquid VMM prior to addition of MMS to a final concentration of 0.035% MMS. After three hours, cells were collected using Buchner funnel and washed with 500mL of liquid VMM to remove MMS. Washed cells transferred to pre-warmed VMM and allowed to recover for 4 hr. Aliquots of cells were harvested and immediately frozen in liquid nitrogen prior to MMS treatment, and hourly during and after the 3 hours MMS treatment. Genomic DNA was isolated and 300 ng was digested by AP endonuclease (cat # M0282S, New England Biolabs), human alkyladenine DNA Glycosylase (cat # M0313S, New England Biolabs), or both enzymes for 1hr and 15min at 37C. Reactions were stopped by adding alkaline DNA loading buffer (50 mM NaOH, 1 mM EDTA, 3% Ficoll). Samples were resolved on a 1.2% alkaline agarose gel (1.5 M NaOH, 50 mM EDTA). Agarose gels were run in the cold room at 25V for 17hrs and then incubated in neutralization buffer (1.5 M NaCl, 1 M Tris-Cl pH 7.6) for 45 minutes before being stained with SYBR Gold (cat # S-11494, Life Technologies) for 40min and de-stained for 30 min before imaging.
MethylC-seq libraries were prepared according to the following protocol [86]. Illumina sequencing was performed using an Illumina NextSeq500 Instrument at the University of Georgia Genomics Facility. Raw data were trimmed for adapters, preprocessed to remove low quality reads and aligned to the N. crassa (version 12) reference genome as previously described in [87]. Mitochondria sequence (which is fully unmethylated) was used as a control to calculate the sodium bisulfite reaction non-conversion rate of unmodified cytosines. Binomial test coupled with Benjamini-Hochberg correction was adopted to determine the methylation status of each cytosine. Identification of DMRs (Differentially Methylated Regions) was performed as described in [88]. Methylated regions in wild type was generated by running DMR finding between two wild type samples and an artificially created sample, which has 60X genome coverage but without any methylated cytosines. The maximum physical distance to combine two DMSs (Differential Methylated Sites) was set to 1kb. DMRs with at least 10 DMSs were reported and used for subsequent analyses. For metaplots, both upstream and downstream regions were divided into 20 bins each of 50bp in length for a total 1kb in each direction. Methylated regions were separated every 5%, for a total of 20 bins. Weighted methylation levels were computed for each bin as described previously[89].
Illumina sequence reads have been deposited into the NCBI GEO database (Accession #GSE70518).
|
10.1371/journal.pntd.0004078 | Cryptosporidiosis: A Disease of Tropical and Remote Areas in Australia | Cryptosporidiosis causes gastroenteritis and is transmitted to humans via contaminated water and food, and contact with infected animals and people. We analyse long-term cryptosporidiosis patterns across Australia (2001–2012) and review published Australian studies and jurisdictional health bulletins to identify high risk populations and potential risk factors for disease. Using national data on reported cryptosporidiosis, the average annual rate of reported illness was 12.8 cases per 100 000 population, with cycles of high and low reporting years. Reports of illness peak in summer, similar to other infectious gastrointestinal diseases. States with high livestock densities like New South Wales and Queensland also record a spring peak in illnesses. Children aged less than four years have the highest rates of disease, along with adult females. Rates of reported cryptosporidiosis are highest in the warmer, remote regions and in Aboriginal and Torres Strait Islander populations. Our review of 34 published studies and seven health department reports on cryptosporidiosis in Australia highlights a lack of long term, non-outbreak studies in these regions and populations, with an emphasis on outbreaks and risk factors in urban areas. The high disease rates in remote, tropical and subtropical areas and in Aboriginal and Torres Strait Islander populations underscore the need to develop interventions that target the sources of infection, seasonal exposures and risk factors for cryptosporidiosis in these settings. Spatial epidemiology can provide an evidence base to identify priorities for intervention to prevent and control cryptosporidiosis in high risk populations.
| The parasite Cryptosporidium is a common cause of gastroenteritis worldwide. Ineffectively focused interventions are partly why the disease remains a challenge to control. In this study, we describe the geographical, seasonal and demographic characteristics of reported cryptosporidiosis in Australia from 2001–2012. We combine this analysis of illnesses with evidence published in peer review articles and state health bulletins to identify high disease risk areas and populations. We find that rates of reported cryptosporidiosis are highest in warm, remote areas and in Aboriginal and Torres Strait Islander populations’ dominated regions. Our review of the published literature and health reports highlights a focus on short term outbreaks in metropolitan areas. This negligible overlap between areas with high disease rates and areas of public health focus is of concern. Public health interventions and promotion programs to prevent and control diarrhoea need to focus on remote and Indigenous dominated Australia to reduce the currently high rates in these regions and populations.
| Cryptosporidiosis, caused by the intestinal parasite Cryptosporidium, is an important cause of gastroenteritis worldwide, particularly in resource limited settings [1]. Infection commonly presents as self-limiting gastroenteritis, but in the immune-compromised, children and the elderly can result in persistent infection, malnutrition and, more rarely, death [2–4]. Cryptosporidium is transmitted via the faecal-oral route and is easily spread through water [5], food [6], contact with infected animals [7,8], contaminated environments [9] and through contact with infected persons [10]. The pathogen requires a low infectious dose, oocysts are immediately infectious once excreted and are resistant to traditional water treatment methods for drinking water supplies and swimming pools, such as chlorination [11–13]. These characteristics make Cryptosporidium ubiquitous in the environment and the disease a challenge to control. Despite its importance worldwide, our limited understanding of the sources of infection and pathways for spread has resulted in ineffective public health strategies to prevent human infection [14,15].
Cryptosporidiosis is recognized as a parasitic disease with suboptimal disease prevention measures resulting in high disease rates in some populations in Oceania, similar to other places [16]. Identification of areas and time periods with consistently high rates of infection can help determine the environmental, socio-economic and demographic risk factors for disease. This knowledge can inform the development of targeted public health interventions to reduce disease. To date, there are no national descriptions of the seasonal, spatial and population-specific patterns of reported cryptosporidiosis in Australia.
Cryptosporidiosis became a nationally notifiable disease in Australia in 2001 [17]. We use Geographic Information Systems (GIS) approaches to identify and visualize high risk populations across Australia from 2001 to 2012. Our analysis provides an insight into environmental, social and demographic factors at the population level that may result in localized areas and time periods of high cryptosporidiosis risk. Combined with a review of published literature on cryptosporidiosis in Australia, we identify high risk groups that may benefit from targeted interventions for disease control.
Using three electronic databases, PubMed, Web of Science and Embase, publications that focused on cryptosporidiosis in Australia were identified. The keywords used were: (“Australia”), AND (“cryptosporidiosis”, “cryptosporidium”). No language or database or time restrictions were imposed on the searches. Full-text versions of the articles that focused on cryptosporidiosis in humans were obtained and their reference lists were manually searched to identify any further relevant manuscripts. Bibliographies of reviews published on Cryptosporidium epidemiology were also examined to identify additional sources for inclusion in the review. We also searched jurisdictional health departments’ public health bulletins in order to capture Cryptosporidium spp. outbreak reports and surveillance summaries. For the relevant studies, the following information was recorded: authors and year of publication, location, outbreak or sporadic, study design, the main risk factors or potential sources of infection identified.
All cases of cryptosporidiosis reported during 2001–2012 in Australia were obtained from the National Notifiable Diseases Surveillance System, which is overseen by the Communicable Disease Network Australia and managed by the Australian Government Department of Health. Case data obtained included notification date, sex, age (in five year age groups), Aboriginal and Torres Strait Islander (ATSI) status, state and postcode of residence. Ethical approval for the study was obtained from the Australian National University prior to data release. For notifications, cases were defined according to the national case definition for confirmed cases, requiring laboratory definitive evidence only with no clinical criteria [17,18].
Annual population denominator data by age, sex, ATSI status and state were based on Estimated Resident Population (ERP) estimates produced annually by the Australian Bureau of Statistics (ABS). Geographical State and Territory and postal area boundaries were obtained from the ABS. The Postal area boundaries produced by the Australian Bureau of Statistics differ between each census, due to population changes, changes to postal distribution areas and changes to the ABS methodology in defining postal area boundaries. For the 2001 and 2006 census, postal area boundaries were based on census districts and for 2011, the methodology was changed to ensure postal areas matched Statistical Area 1 boundaries (a non-administrative, geographical unit defined by the Australian Bureau of Statistics). This resulted in postal areas being removed, added and boundary changes. The ABS has not produced correspondence files allowing postal areas from 2001 and 2006 to be matched to 2011 boundaries. Therefore in this paper, data from 2001–2005 were mapped using the 2001 postal area boundaries; data from 2006–2010 were mapped using the 2006 postal area boundaries and data from 2011–2012 were mapped using the 2011 postal area boundaries. More information on postal area boundaries can be found at www.abs.gov.au.
National, State/Territory, age and sex specific cryptosporidiosis rates were calculated using the annual ERPs for each year from the ABS and converted to rates per 100 000 population. Seasonal patterns were visualized using the total number of notifications by week over the study period for each State and Territory. As data were available at the postal area level, using the ABS index of remoteness for 2011[19], each postcode was assigned a value of “remoteness” to describe cases by location. The ABS classification for remoteness designates each area as either ‘major cities’ ‘inner regional’, ‘outer regional’, ‘remote’ or ‘very remote’. To generate maps of reported illness by ATSI status, the total number of notified ATSI cases in each region was divided by the total ATSI population from the relevant census and converted to rates per 100,000 population. It is important to note that there is variation across the states and territories in the completeness of reporting of ATSI status for cryptosporidiosis. Following the methods of the Australian Institute of Health and Welfare, to assess completeness of reporting by State and Territory, we used a cut-off of 50% completeness of ATSI status for the period 2009–2012 [20]. All States and Territories met this criterion, apart from Victoria (48% complete with 52% of records with a blank field for ATSI status) and Queensland (44% complete with 56% of records with a blank field for ATSI status). Victoria and Queensland were excluded from the calculation and visualization of rates of reported cryptosporidiosis for ATSI populations. Stata v 13.0 and Microsoft Excel 2010 were used for data management and analysis including rate calculations and confidence intervals (using the command “ci” to generate the mean, standard error and 95% confidence intervals for each category). ArcGISv10.0 was used for mapping.
We found 38 relevant published studies on cryptosporidiosis in Australia, of which 34 studies are summarized in S1 Table. The full texts of four studies published prior to 1993 were not available. We found 9 additional reports from searching the jurisdictional health bulletins. Of these, one was published as a peer reviewed research article while two descriptions reported on the same outbreak. Together, there were 41 individual reports on cryptosporidiosis in Australia. Overall, 56% (23/41) of published studies were conducted in urban settings, 14% (6/41) focused on remote regions, 27% (11/41) were state-wide analyses and the locations in which samples were obtained were unknown for three studies. Of the eight studies that used a case-control study design, six identified swimming pools as the main risk factor for outbreaks and sporadic cases in urban areas. Consumption of unpasteurized milk and calf contact at an agricultural show and animal petting farms were identified as risk factors for three other outbreaks. Of the molecular focused studies, 83% (10/12) identified cattle subtypes in human infections, two identified a kangaroo subtype in a human infection and one case report described a wildlife associated genotype. Four studies used data for over five years with data limited to state specific reported incidence.
The average annual rate of reported cryptosporidiosis was 12.8 cases per 100 000 population. Cryptosporidiosis notifications showed distinct cycles over the 12 year period from 2001 to 2012. The rate of reported illnesses peaked in 2002, 2005 and 2009, and was lower in the years 2003, 2008 and 2010 (Fig 1). The number of illnesses was highest in 2009 with 4625 notified cases nationally.
A seasonal pattern of disease notification remained consistent throughout the reporting period and was evident in most States and Territories (Fig 2). Notifications peaked towards the end of summer starting in week six (February) with the Australian Capital Territory, and ending with the Northern Territory in week 9 (early March) (Fig 2). Tasmania, Queensland and New South Wales also saw rising numbers of notifications in spring, from week 39 (September) onwards. Victoria also has a smaller, later spring peak from week 45 (late October) onwards.
Average annual cryptosporidiosis rates of reported illness by remoteness category are shown in Fig 3. The lowest rate of notifications was in the major cities (7.7 per 100 000 population), followed by inner regional areas (9.0 per 100 000 population), outer regional areas (12.7 per 100 000 population), and remote areas (17.1 per 100 000 population). The areas categorized as “very remote” had the highest average rates of notification (44.9 per 100 000 population).
Overall, the highest number of reported illnesses was seen in children aged 0–4 years, comprising 45% (15852/35455) of all reported illnesses (Table 1). Overall, males and females showed a similar distribution of illnesses with 17551 (49.5%) and 17904 (50.5%) illnesses respectively. Queensland had the highest number of reported illnesses (n = 12271), while Tasmania had the lowest (n = 706) (Table 1). Average annual rates of reported cryptosporidiosis vary by age, gender and geographic region (S1 Fig). Using annual population estimates for each age and gender group, all States and Territories had the highest average notification rates in boys aged 0–4 years and higher rates for adult females across the age groups 20 to 39 years compared to males (S1 Fig).
Over the 12 year period, rates of reported cryptosporidiosis in ATSI populations were highest in the northern, tropical region of Australia with high rates in parts of Western Australia (Fig 4). When the rates of reported cryptosporidiosis for the general population were viewed in relation to the distribution of climatic regions and remoteness categories identified by the ABS (S2 Fig), we observed consistently high rates of reported cryptosporidiosis in the warmer tropical and subtropical, remote and very remote areas (S3 Fig).
Rates of reported cryptosporidiosis in Australia are 12.8 illnesses per 100 000 population. Rates in other developed countries with similar surveillance systems are generally lower at 2.9 per 100 000 (United States) [21] and 2.7 per 100 000 (Canada) and 8 per 100 000 (England and Wales), but higher in New Zealand at 22 per 100 000 [22]. These high rates support Cryptosporidium’s status as an important parasitic infection of the Oceanic region [16]. Cryptosporidiosis patterns in Australia show marked geographic and demographic variations in disease reporting rates. The high rates of reported illness in the warmer, remote and Indigenous dominated areas in Australia combined with the lack of literature on non-outbreak disease patterns in these settings is of concern.
Environmental transmission cycles between humans and animals or contaminated environments are likely to be an important mode of disease spread in some regions. Livestock are an important reservoir of Cryptosporidium. Of interest is the increase in the number of reported illnesses in spring in Queensland and New South Wales. According to the 2011 Agricultural Census [23], these States were the leading cattle (dairy and meat) rearing regions in Australia. This seasonal pattern may be due to transmission from young livestock born at this time, as young calves harbor high loads of the Cryptosporidium strain known to cause disease in humans [24–26]. Spring peaks in reported cryptosporidiosis in humans have been documented to coincide with the presence of young livestock, notably cattle, known to carry a considerable load of Cryptosporidium spp. capable of causing human infection in New Zealand and the UK [8,24,27]. In Australia, associations between cryptosporidiosis and cattle are evident from our review of published studies which include molecular studies indicating the existence of cattle adapted strains in human cases [28–31], outbreaks related to the consumption of unpasteurized milk [32] and attendance at an animal fair [33] and an animal petting farm [34]. Of interest is the increasing molecular evidence of many different strains of Cryptosporidium in the environment and animals, whose potential for infection in humans is largely unknown [35].
Strong seasonal patterns in the number of reported illnesses are also evident in other states and territories, with increasing numbers of reported cases in December through to March; a pattern maintained throughout the reporting period. A study included in our review shows that in Brisbane, a subtropical city of Queensland, Australia, hot temperatures and dry conditions were associated with an increased risk of cryptosporidiosis [36]. Globally, cryptosporidiosis shows a strong association with rainfall and temperature variability [37,38] across a wide range of climatic regions and countries [39]. Quantifying the association between disease risk and environmental factors such as weather variability and proximity to livestock and wildlife can help identify populations at imminent risk of infection.
Residing in remote or rural regions has been identified as a risk factor for cryptosporidiosis in studies published elsewhere [40,41]. We show that in Australia, rates of reported cryptosporidiosis increase with increasing remoteness. Rural living as a risk factor for cryptosporidiosis is particularly linked to cattle density [42]. Our review of the published literature suggests that cattle are likely to be an important reservoir of human infection in some rural areas. However, remote living may also be associated with less than ideal hygienic conditions for food preparation and storage [43], poor housing and household overcrowding [44]. Indeed, in remote Indigenous communities, hygiene and public health interventions, which include handwashing with soap and water, sanitation and hygiene promotion are most likely to reduce diarrhoeal illness, especially in children [45]. In the only outbreak reported in remote Northern Territory, human to human transmission was implicated as the cause of cryptosporidiosis spread in younger age groups. Inadequate provision of drinking water and protection of drinking water sources from faecal contamination has been implicated as the source of several outbreaks of cryptosporidiosis elsewhere [46]. In remote Australia, the lack of adequate quantities of potable drinking water has been identified as a major health issue [47] and may partly be responsible for the high cryptosporidiosis rates in these areas. To develop effective interventions tailored to these settings, we need to target the environmental and social factors and individual behaviors that drive the high rates of disease in remote areas.
Consistent with published studies, strong age related patterns in reported illnesses were identified. The 0–4 year age group had the highest rates of reported cryptosporidiosis consistently over the 12 year period. Such a pattern could be a result of increased contact with sources of infection [48], susceptibility to infection due to immunological naiveté or a greater probability of seeking treatment [49]. As Cryptosporidium is transmitted through the faecal-oral route, higher disease rates in adult females between 20 and 39 years may be partly explained by nappy changing and child minding activities (25, 26) which have been identified as risk factors in previous cryptosporidiosis studies [50,51]. In low income settings, some studies have found that cryptosporidiosis in children is associated with malnutrition and was linked to delayed growth and development [3,52,53] and death [4]. While the high rates of cryptosporidiosis in preschool children is widely established [1,22], gaps in our understanding of the factors that drive this high burden of disease remain. A follow up of children with cryptosporidiosis could provide valuable information on immunity to recurrent infections and growth outcomes in remote Australian communities with high cryptosporidiosis rates.
Some studies have found that Indigenous populations and minority ethnic groups have lower rates of reported illnesses and hypothesized that such patterns could be due to poorer access to health resources resulting in lower testing and reporting rates of illnesses [54,55]. However, despite the under reporting associated with surveillance systems [56] and the lack of health resources in these areas [57], we found that that high rates of reported cryptosporidiosis in Australia cluster in areas with a high proportion of ATSI populations (parts of the Northern Territory, which has the highest proportion of ATSI people in Australia, followed by Western Australia). Thus, the real burden of cryptosporidiosis in these communities is likely to be much higher. Cryptosporidiosis is already recognized as an important parasitic disease in the Oceanic region [16] with much higher rates of reported infection in ATSI peoples [57]. Our findings support these studies and add further evidence to show that this is an important issue for public health deserving of ongoing attention with respect to interventions for disease control.
The main limitation of this study is that it was based on passive surveillance data, which is known to suffer from significant under reporting [56]. However, the seasonal patterns and marked geographic variation in the rates of reported cryptosporidiosis identified here are indicative of locality specific exposures and risk factors. Such an analysis is only possible with data that are collected across space and over time. There has been no change of reporting practices for Cryptosporidium that has occurred over the time period of analysis, either nationally or locally. The incompleteness of ATSI status reporting, particularly in Queensland where only 48% of the data were complete, indicates that detailed statistical analysis using the current ATSI status data would be unwise at the national scale. Finally, different species can be spatially and temporally distinct in their prevalence and transmission pathways, potentially resulting in different patterns of disease incidence [58]. Although we were unable to address species differences, this also represents strength of our study. Molecular methods to identify species and serotypes may differ across States and Territories accounting for inter-state differences in disease patterns, therefore analysis at genus level only is most reliable across Australia and internationally.
Rates of reported cryptosporidiosis in Australia are higher than many comparable developed countries. Through our analysis of reported illness data and summary of published and health department literature we show that the majority of these studies are short term outbreak reports and focused molecular studies, providing limited understanding of disease patterns Australia-wide.
We identified strong seasonal patterns in notification data that suggest environmental factors are important predictors of cryptosporidiosis risk. Investigating the association between cryptosporidiosis and regional weather variability and proximity to livestock will provide important information on environmental transmission of Cryptosporidium.
The finding that children under four years old, particularly male children, are at increased risk, and a trend towards higher rates for females in the 20–39 years age group, is a common pattern among enteric infections. However, cryptosporidiosis infection may be a marker for impaired development and cognition in later life. Longitudinal studies that examine how environmental and social factors affect growth and development of children are needed.
Rates of reported cryptosporidiosis are highest in the warm and ATSI population dominated regions of Australia. Moreover, an increasing risk of reported cryptosporidiosis with remote living is evident. The high rates of reported cryptosporidiosis in remote and ATSI dominated regions suggest that the real incidence of cryptosporidiosis in these communities is potentially much greater. Effective prevention and control of cryptosporidiosis requires interventions and promotions aimed at the environmental, societal and individual exposures that underlie this localization of high cryptosporidiosis risk. Isolated epidemiological investigations conducted in urban areas are unlikely to provide an insight into disease transmission in remote areas. To control cryptosporidiosis in Australia, tailoring effective health interventions/promotion for remote communities need to be a priority for public health research and action.
|
10.1371/journal.ppat.1006071 | Structural and Functional Characterization of the Bacterial Type III Secretion Export Apparatus | Bacterial type III protein secretion systems inject effector proteins into eukaryotic host cells in order to promote survival and colonization of Gram-negative pathogens and symbionts. Secretion across the bacterial cell envelope and injection into host cells is facilitated by a so-called injectisome. Its small hydrophobic export apparatus components SpaP and SpaR were shown to nucleate assembly of the needle complex and to form the central “cup” substructure of a Salmonella Typhimurium secretion system. However, the in vivo placement of these components in the needle complex and their function during the secretion process remained poorly defined. Here we present evidence that a SpaP pentamer forms a 15 Å wide pore and provide a detailed map of SpaP interactions with the export apparatus components SpaQ, SpaR, and SpaS. We further refine the current view of export apparatus assembly, consolidate transmembrane topology models for SpaP and SpaR, and present intimate interactions of the periplasmic domains of SpaP and SpaR with the inner rod protein PrgJ, indicating how export apparatus and needle filament are connected to create a continuous conduit for substrate translocation.
| Many Gram-negative bacteria use type III secretion systems to inject bacterial proteins into eukaryotic host cells in order to promote their own survival and colonization. These systems are large molecular machines with the ability to transport proteins across three cell membranes in one step. It is believed that the only gated barrier of these systems lies in the bacterial cytoplasmic membrane but it was unclear so far how this gate looks like and of which components it is composed. Here we present evidence based on in depth biochemical and genetic characterization that an assembly of five SpaP proteins forms this gate in the cytoplasmic membrane of the type III secretion system of Salmonella pathogenicity island 1. We further show that one subunit each of the proteins SpaQ, SpaR, and SpaS are closely associated to the SpaP gate and may function in the gating mechanism, and that the protein PrgJ is attached to this gate on the outside to connect it to the hollow needle filament projecting towards the host cell. Our findings elucidate a hitherto ill-defined aspect of type III secretion systems and may help to develop novel antiinfective therapies targeting these virulence-associated molecular devices.
| Type III secretion systems (T3SSs) are used by many Gram-negative bacterial pathogens and symbionts to translocate effector proteins in one step across the bacterial envelope and into eukaryotic host cells [1] where they modulate host cell physiology to promote bacterial survival and colonization [2]. The core of T3SSs is formed by the so-called injectisome, a macromolecular machine composed of up to 20 different proteins [1]. The base of the injectisome, consisting of an outer membrane secretin ring and two inner membrane ring components, anchors the system to the bacterial cell envelope [3]. A filamentous needle projects away from the base towards the host cell and serves as conduit for translocated effectors [4,5]. Five cytoplasmic proteins select and unfold the substrates, which are then handed over to the actual export apparatus [6,7] housed in a membrane patch at the center of the inner ring [8,9]. The five export apparatus components are thought to facilitate the actual secretion function of T3SSs, including energy coupling, membrane translocation, and substrate specificity switching [1]. Base, needle filament, and export apparatus are together also termed needle complex.
While analyses by X-ray crystallography and cryo electron microscopy have revealed the structure of most soluble components of injectisomes or of the related flagellar system [10,11], the structure and in particular the function of the hydrophobic transmembrane (TM) domains of the export apparatus components remain poorly defined. In the T3SS encoded within the pathogenicity island 1 (SPI-1) of Salmonella enterica serovar Typhimurium (S. Typhimurium), the export apparatus is composed of the proteins SpaP, SpaQ, SpaR, SpaS, and InvA in a 5:1:1:1:9 stoichiometry [12]. Of these components, InvA and SpaS are structurally and functionally best characterized: the atomic structures of their soluble cytoplasmic domains have been solved [13,14]. The large cytoplasmic domain of InvA (or its homologs) forms a nonameric ring with a central pore of about 50 Å in diameter [15] and has been proposed to play a role in substrate switching and translocation [16,17] while its 8 predicted TM helices have been proposed to serve in utilization of the proton motive force for secretion [18]. SpaS and its homologs play a role in switching of specificity from secretion of early to intermediate and late substrates [19]. Autocleavage of a highly conserved NPTH motif in the cytoplasmic domain of SpaS is required for this function, possibly to facilitate a high conformational flexibility of this domain for secretion of later substrates [20].
The substantially hydrophobic export apparatus components SpaP, SpaQ, and SpaR and their homologs were shown to be critical for assembly of the needle complex [9,21–23] and essential for secretion function [9,24] but their precise role in secretion is still unknown. It was suggested that SpaP and SpaR form the cup substructure of the needle complex [9]. Given the presumed central location of SpaP and SpaR at the center of the membrane patch of the needle complex and their substantial hydrophobicity, we hypothesized that these two proteins may constitute the actual substrate translocation pore of T3SSs in the bacterial inner membrane, a function that as yet has not been assigned to any T3SS component.
In this study, we have biochemically characterized a stable subcomplex formed by SpaP and SpaR, and mapped its place within the needle complex using in vivo photocrosslinking and complementary techniques. We show that an isolated complex of five SpaP and one SpaR forms a donut-shaped structure with an approximately 15Å wide recession at its center. Sole expression of the SpaP pentamer in the bacterial membrane allowed the permeation of compounds of 500 Da into the cytoplasm, suggesting that these proteins form a channel large enough for translocation of secondary structures. We further show that a complex of SpaP, SpaQ, SpaR, and SpaS assembles in vivo before incorporation into the needle complex base, and that these four export apparatus components form a compact assembly with multiple reciprocal interactions at TM helices three and four of the SpaP pentamer. We also present evidence that SpaP and SpaR interact on their periplasmic side with the inner rod protein PrgJ, which provides a basis to explain how the substrate translocation conduit is continuous from the export apparatus through the inner rod into the needle filament and suggests that the hitherto unaccounted electron density of the socket substructure is made of the periplasmic domains of SpaP and SpaR, together with PrgJ. In summary, we describe physical interactions among export apparatus components of bacterial T3SSs and identify the components that form its substrate translocation pore. This work will facilitate further structural and functional work on these machines and may help to develop novel antiinfective therapies targeting these virulence-associated molecular devices.
We previously showed that a stable complex of SpaP and SpaR can be isolated from S. Typhimurium lacking the inner ring components PrgH and PrgK [9]. For further characterization, we expressed the spaPQRS operon in Escherichia coli and purified the SpaPR complex by immunoprecipitation of epitope-tagged SpaR. The isolated complex eluted as a sharp peak from a size exclusion chromatography column at an apparent size of 400 kDa (Fig 1A). Separation of the protein complex by SDS PAGE followed by Coomassie staining or Western blotting and immunodetection of SpaP and SpaRFLAG, respectively, showed that the complex contained more SpaP than SpaR (1B). Since the masses of membrane protein complexes deduced from analysis by size exclusion chromatography are skewed by the presence of bound detergent, we analyzed the fraction of protein and detergent contained in the isolated SpaPR complexes by size exclusion chromatography-multi angle laser light scattering. This analysis determined that the SpaPR peak was monodisperse, corresponding to a size of 311 kDa with a calculated protein content of 160 kDa (Fig 1C, S1 Table, S1 File), suggesting a total of 6 molecules of SpaP (25.2 kDa) and SpaR (31.7 kDa including C-terminal 3xFLAG tag). Given a mean error of 7% (S1 Fig), these data did not allow to distinguish whether the complex composition was 4 SpaP + 2 SpaRFLAG (calc. 164 kDa) or 5 SpaP + 1 SpaRFLAG (calc. 158 kDa). Native mass spectrometry was then performed to assess the exact stoichiometry of a purified SpaPRSTREP complex. A major species of complex produced peaks of 157.882 kDa and a minor species of 158.595 kDa. These masses are consistent with a stoichiometry of 5 SpaP and 1 SpaRSTREP (calculated molecular mass of 157.280 kDa) with bound phospholipids. In summary, these results show that the isolated SpaPR complex obtained from overexpression in the absence of other needle complex components has the same stoichiometry as SpaP and SpaR assembled into complete needle complexes [12] and indicates that the isolated SpaPR complex is a relevant functional module of the needle complex.
To further validate the stoichiometry of SpaP and SpaR and to characterize the placing of this module within the assembled needle complex, we employed an in vivo photocrosslinking approach based on the genetically encoded UV-reactive amino acid para-benzophenylalanine (pBpa) [25]. pBpa was built into the predicted TM helices of SpaP and SpaR, respectively, so that possible interactions at every face of the predicted TM helices were sampled (Fig 2A and 2B). spaP or spaPQRS deletion mutants of S. Typhimurium were complemented with SpaPFLAG or SpaPQRFLAGS containing pBpa at selected positions and expressed from a low copy number plasmid. Complementation of T3SS function of these mutants was assessed by analyzing type III-dependent secretion of substrate proteins into the culture supernatant (S2 Fig). Crosslinking of pBpa to nearby interactors was induced by UV irradiation of intact bacterial cells immediately after harvesting. Subsequently, crude membranes were isolated and crosslinking patterns were analyzed by SDS PAGE and immunodetection of the FLAG-tagged bait protein. Crosslinked adducts of different sizes were identified at various positions of SpaP and SpaR (Fig 2C and 2D). To exclude crosslinking artifacts resulting from plasmid-based complementation, pBpa positions that produced representative crosslinking patterns were also introduced into the chromosome-encoded genes, and crosslinking was performed accordingly. Notably, for all tested chromosomal positions the quality of previously identified crosslinks could be confirmed while the efficiency of crosslinking improved in some cases, possibly due to a more efficient complex assembly achieved by expression of pBpa-containing proteins from its native context (Fig 2E and 2F). To identify the nature of crosslinked adducts, needle complexes with pBpa-containing SpaPFLAG or SpaRFLAG were purified, UV-irradiated, resolved by SDS PAGE, and gel slices of the positions of the crosslinks were analyzed by mass spectrometry (S3 Fig). This analysis identified crosslinks between SpaP and the export apparatus components SpaS and SpaQ, and between SpaP and the inner rod protein PrgJ. Furthermore, crosslinks between SpaR and SpaP, SpaQ, and PrgJ were also identified (S2 Table, Fig 2C and 2D). The detailed validation and interpretation of the crosslinking analysis is presented in the following three sections.
UV-irradiation of SpaPFLAG-containing pBpa at positions L7, L10, A12, F13, S14, T15, M187, S189, I193, and T195 showed a ladder of crosslinks at 40 kDa, 70 kDa, 120 kDa, and 200 kDa (Fig 2C and 2E). We reasoned that this crosslink ladder might correspond to a homo-oligomeric crosslinking of the SpaP pentamer. Two further experimental results supported this hypothesis: First, crosslinking of SpaPT15XFLAG expressed in E. coli in the absence of other T3SS components showed the same crosslink ladder (Fig 3A); and second, crosslinking plasmid-complemented SpaPT15X in an S. Typhimurium strain with chromosome-encoded SpaPFLAG also produced the 40 kDa FLAG-containing crosslink, which proved at least a bipartite SpaPT15X-SpaPFLAG interaction (Fig 3B). Several of the SpaP pBpa mutants that produced a ladder upon crosslinking (A12X, T15X, M187X, S189X, I193X) were non-functional (S2 Fig). Analysis of two of these pBpa mutants (T15X and M187X) by 2-dimensional blue native/SDS PAGE indicated that the observed SpaP-SpaP interaction occurred between SpaP assembled into the complete needle complex as well as between SpaP molecules that had not yet been yet incorporated into this structure (Fig 3C). These results suggest that the loss of function of these mutants is unlikely due to improper folding or assembly but rather due to subtle conformational changes that alter their function.
Overall, these results indicate that TM helix one and to a smaller extent the cytoplasmic face of TM helix three and four are involved in protomer contacts in the SpaP homopentamer while only few homotypic interactions were observed at positions of TM helices two and three.
To cross-validate the experimental findings, we performed an independent prediction of SpaP-SpaP interactions based on analysis of sequence co-variation using the software EV couplings [26–28]. 27 of the experimentally tested SpaP positions were predicted to be involved in SpaP-SpaP interactions with a normalized coupling score >0.80 (S3 Table). 18 of the 27 experimentally tested positions yielded indications of SpaP-SpaP interactions, 2 positions were experimentally ambiguous because of very low expression levels of the mutated proteins, and 7 positions showed no signs of SpaP-SpaP interactions. As used, EV couplings does not distinguish between intra and intermolecular interactions. 6 of the predicted but experimentally negative positions are likely to be involved in intramolecular interactions, which are not detectable by the in vivo photocrosslinking approach used (Fig 3D). Many intermolecular interactions at experimentally tested SpaP positions were predicted to connect two TM helices 1 or TM helix 1 and 3 in a parallel fashion, and TM helices 1 and 2 or TM helices 3 and 4 in an antiparallel fashion (Fig 3D), supporting a SpaP topology as depicted in Fig 2A, while only the coupling prediction of SpaPS189 (to L11) opposed this model. Overall, the bioinformatic analysis supports our experimental results, strengthens the topology model of SpaP, and provides a first picture of the buildup of the SpaP pentamer.
Mass spectrometry analysis of crosslinked SpaP and SpaR adducts produced evidence for multiple interactions among the export apparatus components SpaP, SpaQ, SpaR, and SpaS (Fig 2, S3 Fig, S2 Table). To validate these results by immunoblotting, we assayed the SpaP-SpaR as well as the SpaP-SpaS interactions by FLAG-tagging the target instead of the pBpa-containing bait protein. We found that SpaP interacts with SpaRFLAG through its residues V170 and L210 but not through V203 and A204 (Fig 4A) and that SpaR contacts SpaPFLAG via its residue N151 (Fig 4B). Using an autocleavage-deficient FLAG-tagged variant of the switch protein SpaS, we could further validate interactions between SpaS and SpaPV200X/SpaPV203X (Fig 4C). In summary, these crosslinking data indicate that, consistent with our previous report [12], 1 SpaQ, 1 SpaR, and 1 SpaS form a closely interconnected assembly that contacts SpaP at TM helix three (V170: SpaQ, SpaR) and TM helix four (V200/203: SpaQ, SpaS). The interaction of these four proteins seems to be integrated by SpaQ as this small protein makes contacts to all other three proteins (in vivo photocrosslinking-identified SpaS-SpaQ contacts communicated results of J. Monjarás Feria).
Previous results showed that SpaQ is critical for efficient formation of the needle complex base but due to technical limitations of the blue native PAGE approach used at the time, it was not clear whether assembly proceeds through a pre-assembled complex of all four minor export apparatus components before integration into the base or whether these components only interact upon base integration [9]. To examine the early events of the assembly of the T3SS export apparatus components, we probed the SpaP-SpaQ, SpaP-SpaS, and SpaR-SpaQ interactions identified by the crosslinking studies in strains deficient in the inner ring protein PrgK. These mutants are deffective for base assembly thus allowing to prove the requirement of a fully assembled base for the assembly of the export apparatus. Indeed, we detected SpaP-SpaQ and SpaP-SpaS interactions at SpaPX203 in the absence of PrgK (Fig 4D and 4E), and SpaR-SpaQ interactions at SpaRX209 (Fig 4F). SpaP-SpaP and SpaPV170X-SpaR crosslinks were also identified when plasmid-encoded SpaPQRS were expressed in E. coli BL21, lacking all other T3SS components (Fig 4G and 4H). Altogether, these results indicate that assembly of the export apparatus precedes and is independent of base assembly.
UV-irradiation of SpaPFLAG with pBpa at position L7 or SpaRFLAG with pBpa at positions F20, N44, and A45 resulted in an 8 kDa mobility shift of these proteins in SDS-PAGE (Fig 2C, 2D and 2E). Mass spectrometry analysis of the shifted bands identified PrgJ in both cases (S3 Fig, S2 Table). In an effort to characterize the extent of the SpaP-PrgJ interaction in more detail, we also noted the same mobility shift of SpaP after UV-irradiation of SpaPFLAG with pBpa at positions G2, N3, D4, I5, and S6, where crosslinked PrgJ was confirmed by immunodetection (Fig 5A). To rule out potential artifacts due to overexpression of the plasmid-borne constructs, we confirmed the crosslinks of SpaPG2XFLAG and SpaPS6XFLAG after expression from their native chromosomal context (Fig 5B). 2-dimensional blue native/SDS PAGE analysis of the crosslinks resulting from UV-irradiation of SpaRA45XFLAG showed that the observed SpaR-PrgJ interaction is only observed when SpaR is incorporated into the needle complex (Fig 5C). Furthermore, SpaP-PrgJ as well as SpaR-PrgJ interactions were not observed in an ATPase activity-deficient InvCK165E mutant, demonstrating that the detected interactions dependent on active type III secretion, which is consistent with the observation that incorporation of PrgJ into the needle complex and inner rod assembly require a functional type III secretion system (Fig 5D). Taken together, these results indicate that the periplasmic domains of SpaP and SpaR serve to anchor the inner rod protein PrgJ to the export apparatus, thus creating a continuous conduit for substrate translocation from the export apparatus to the needle filament.
The location of the SpaP5R1 complex at the center of the needle complex base, right underneath and connected to the filamentous conduit formed by the inner rod and needle proteins, suggests that this complex forms the T3SS’s substrate translocation pore in the bacterial inner membrane.
To obtain structural evidence for its putative pore-forming function, we analyzed the purified, negative-stained SpaPRFLAG complex by electron microscopy. 11202 individual particles were classified and aligned into 91 class averages (S4 Fig). A number of class averages showed a symmetric, donut-shaped complex with an iconic recession at its center (Fig 6A). The diameter of these particles was about 80 Å and the diameter of the recession was about 15 Å. Other class averages showed a more asymmetric shape with an extra density outside of the ring-structure or a mushroom-like shape (Fig 6A). Even though the sample analyzed consisted of a homogeneous population of SpaPRFLAG complexes, it cannot be ruled out that SpaP and SpaRFLAG partly dissociated during sample preparation so that a mixture of SpaP5 and SpaP5R1 complexes was imaged, explaining the diversity of observed classes. It is therefore possible that the donut-shaped particles represent SpaP5 complexes and the asymmetric extension the SpaP-bound SpaRFLAG. Overall, the particles’ shape and dimensions conformed well with the structure of the cup region of assembled bases reported previously (3).
We reasoned that the recession at the center of the observed particles might represent the protein translocation pore of the T3SS. To probe the conducting properties of the SpaPR complex, we assessed its ability to allow the access of biotin maleimide (BM, molecular mass = 500 Da) into the bacterial cytoplasm, an approach that has been used previously to test the gating of the Sec-translocon [29]. The maleimide moiety of BM can only react with and biotinylate free thiol groups of cysteine residues of cytoplasmic proteins if BM can penetrate the inner bacterial membrane through a sufficiently large pore. The extent of biotinylation can then be detected on a Western blot by utilizing streptavidin. Strong BM labeling of proteins was observed in whole cell lysates when SpaPR or SpaP alone were overexpressed from a medium copy plasmid (Fig 6B). Cell fractionation of the expression host showed that only cytoplasmic proteins were differentially labeled by BM upon expression of SpaPR and SpaP, labeling of periplasmic proteins, however, was almost indistinguishable in control and expressing bacteria (Fig 6B). General lysis of the expression host could be ruled out to cause the observed phenotype as neither the cytoplasmic protein RNA polymerase nor the periplasmic maltose binding protein were observed in the culture supernatant of SpaPR or SpaP expressing bacteria (S5A and S5B Fig). Formation of a sizable, ungated pore by these complexes was also indicated by the strong impact even modest overexpression of SpaP and SpaPR had on the viability of the expression host (S5C Fig). Altogether, these results suggest that BM accessed the cytoplasm of the expression host through a pore formed by the expressed proteins. Since SpaP expression alone led to BM labeling of cytoplasmic proteins, it is conceivable that SpaP alone is sufficient to form the actual substrate translocation pore. In line with this idea, overexpressed SpaPEPEA was observed to assemble into high molecular weight complexes when analyzed by blue native PAGE (Fig 6C), however, we were not able to isolate and investigate stable SpaP-only complexes. The access of 500 Da BM to the cytoplasm through the pore of the SpaP pentamer suggests a pore diameter of about 15 Å, which is consistent with the diameter of the recession observed by electron microscopy of the isolated SpaP5R1 complexes.
The export apparatus of bacterial T3SSs is its central unit that facilitates translocation of substrates across the bacterial inner membrane and likely the only gated barrier of these one-step secretion devices. While functions have been proposed for some export apparatus components, the components forming the actual substrate translocation pore in the bacterial inner membrane have not been defined.
In this study we present evidence that a homopentamer of the minor hydrophobic export apparatus component SpaP is a central component of the translocation pore in the inner membrane of the injectisome T3SS encoded by Salmonella pathogenicity island 1. We purified a stable complex of 5 SpaP and 1 SpaR that under electron microscopy exhibited a donut-like shape of about 80 Å in diameter and a 15 Å wide central recession. Expression of the components of this complex in E. coli rendered the bacterial cells permeable to 500 Da compounds, supporting the notion that it may work as translocation channel. Extensive mapping of protein-protein interactions of the TM domains of SpaP and SpaR by in vivo photocrosslinking revealed that SpaQ, SpaR, and SpaS form a compact assembly connected to the central pentamer formed by SpaP. We further demonstrated that assembly of this complex does not require its incorporation into the needle complex. We also detected crosslinks between SpaP and SpaR and the inner rod protein PrgJ showing that the inner rod makes direct contact with the export apparatus.
Previous analysis by blue native PAGE showed that SpaP and SpaR form stable complexes in an S. Typhimurium mutant unable to assemble the needle complex [9]. We now present evidence based on size-exclusion chromatography-multi angle laser light scattering and native mass spectrometry that this complex is composed of 5 SpaP and 1 SpaR. The stoichiometry of the isolated SpaP5R1 complex is consistent with the stoichiometry of SpaP and SpaR in the context of a fully assembled needle complex [12], which indicates that the isolated complex represents a relevant intermediate of needle complex assembly. This notion is further supported by the good match of the dimensions of the observed SpaPR complex with the dimensions of the cup substructure of the needle complex [30], which we previously showed to be composed of SpaP and SpaR [9]. Electron micrographs of the isolated SpaP5R1 complex and BM permeation experiments suggested a pore size of the substrate translocation channel of about 15 Å. Within the range of uncertainty, this diameter conforms with the 10 Å that were reported for the dimensions of the channel of an assembled S. Typhimurium SPI-1 needle complex containing a trapped translocation intermediate [5]. A tight seal during substrate translocation is expected to be important for T3SS to avoid leakage of ions through the open pore, so it is conceivable that the pore diameter closely resembles the dimensions of extended polypeptides or alpha helices. However, a larger pore diameter in its fully open state cannot be excluded given that the herein investigated isolated SpaP5R1 complex most certainly lacks the necessary elements for gating of the pore.
We detected extensive crosslinks of up to five consecutive SpaP at TM helix one and at the cytoplasmic face of TM helices three and four, suggesting that these regions form the major contact area between protomers of the SpaP pentamer. This notion was supported by results of a sequence co-variation-based prediction of residue-residue interactions of SpaP. The formation of these crosslinks was independent of the presence of other needle complex components, supporting the notion that the SpaP pentamer nucleates assembly of the needle complex. Interestingly, the presence of SpaP pentamer crosslinks at TM helices three and four correlated with secretion defects of the respective pBpa mutants, a phenomenon also seen for SpaPA12X and SpaPT15X. The secretion defect was not due to defects in their incorporation into assembled needle complexes, suggesting that these residues may play a critical role in protein translocation.
The recently reported stoichiometry of SpaP, SpaQ, SpaR, and SpaS of 5:1:1:1 [12] suggests that these export apparatus components form an asymmetric assembly within the needle complex. We show here that SpaQ, SpaR, and SpaS contact the SpaP pentamer at its TM helices three and four. We further demonstrate that SpaQ interacts with SpaP and SpaR. These observations, together with the observation that a fusion of SpaR and SpaS homologs retains function [31], suggest that SpaQ, SpaR, and SpaS are not wrapped around the SpaP pentamer but form a compact assembly at one side of SpaP, with SpaQ as the central component that makes contacts to all other components (Fig 7A and 7B). Besides SpaR’s contribution in anchoring the inner rod protein PrgJ, the assembly formed by SpaQ, SpaR, and SpaS might also facilitate gating of the SpaP pore, a critical aspect to prevent detrimental effects of nutrient and ion leakage across the bacterial inner membrane.
The assessment of the dependence of crosslinks between SpaP, SpaQ, SpaR, and SpaS on the presence of the inner ring protein PrgK allowed us to refine a model for the early steps of export apparatus assembly (Fig 7C). We propose that assembly starts with the formation of the SpaP pentamer. This initially unstable complex is stabilized upon binding SpaR. The high stability of the resulting SpaP5R1 intermediate suggests that this complex is the major nucleus of further needle complex assembly. Next, SpaQ and SpaS associate with the SpaP5R1 complex but presumably with weaker affinity since this complex could only be captured after in vivo crosslinking. InvA would then be recruited to the SpaPRQS complex although it is not clear whether its recruitment occurs prior or after this complex initiates the assembly of the needle complex rings. Subsequently, association of the outer membrane secretin InvG and the inner ring protein PrgH would lead to formation of the completed base-export apparatus holo-complex [21,23].
Beyond interactions among the export apparatus components, we also identified crosslinks between the periplasmic domains of SpaP and SpaR and the inner rod protein PrgJ. The close interaction of SpaP, SpaR, and PrgJ is likely to create a continuous conduit for substrate translocation, where PrgJ might serve as an adapter to connect the flat translocation pore of the inner membrane with the helical needle filament. Analysis of the needle complex by cryo-electron microscopy revealed a central juxtamembrane structure at the periplasmic interior of the base, which was termed socket [30], however, no protein could be assigned to contribute to this density. Our results suggest that the socket is composed of the periplasmic parts of SpaP and SpaR, together with the inner rod protein PrgJ. The mass of six PrgJ [12] and the periplasmic domains of five SpaP and one SpaR could well account for the observed density of the socket structure. Our observation now opens the door for further investigations of the relevance of the export apparatus-PrgJ interaction for needle length control, substrate specificity switching, and host cells sensing, functional roles that were suggested for PrgJ [32,33].
The positions of SpaP and SpaR that interact with PrgJ also help to consolidate the TM topology models of these two export apparatus proteins. SpaP is predicted to contain four TM helices (Fig 2A) and the presence of a cleavable signal sequence in flagellar homologs suggests an N-out/C-out TM orientation [34]. This model is supported by the interaction between the N-terminus of SpaP and the periplasmic inner rod detected in this study. Further support for this topology model comes from the presented sequence co-variation-based analysis of SpaP residue-residue interactions, which strongly predicted antiparallel interactions between TM 1 and 2, and between TM 3 and 4 (Fig 3D). The TM topology predictions of SpaR and its homologs are very uncertain, ranging from five to eight TM helices with mostly N-out orientation (Fig 2B, S6 Fig) [34,35]. A C-in orientation, on the other hand, was suggested based on the report of a functional protein fusion of the flagellar SpaR and SpaS homologs of Clostridium, given that the N-terminus of SpaS and its homologs is strongly predicted to reside in the cytoplasm [31,35,36]. Here we presented interactions of SpaR F20, N44, and A45 with the periplasmic protein PrgJ. These residues are predicted to be located within SpaR’s first two TM helices, however, our results rather suggest a periplasmic localization of this part of SpaR. This notion is supported by rather high ΔG values for membrane partitioning of the predicted TM helices one, two, and four (S6B Fig), so that a SpaR model comprising an N-out/C-in topology with only three TM helices is conceivable (S6C Fig).
In summary, we have presented evidence that a pentamer of SpaP forms the substrate translocation pore of T3SSs in the bacterial inner membrane. We show that this pentamer closely interacts with the export apparatus components SpaQ, SpaR, and SpaS in the plane of the membrane, an accessory assembly that may facilitate gating of the export pore. We further show that SpaP and SpaR intimately contact the periplasmic inner rod protein PrgJ and propose that the inner rod serves as an adapter to connect the flat export pore and the helical needle filament, thus creating a continuous conduit for substrate translocation from the bacterial cytoplasm into the host cell.
Chemicals were from Sigma-Aldrich unless otherwise specified. Detergent n-dodecyl-maltoside (DDM) was from Affimetrix-Anatrace. para-benzophenylalanine was from Bachem. SERVA Blue G and SERVAGel TG PRiME 8–16% precast gels were from Serva. NativePAGE Novex Bis-Tris 3–12% gels were from Life Technologies. Primers are listed in S5 Table and were synthetized by Eurofins and Integrated DNA Technologies. Polyclonal rabbit anti-MBP antibody were from New England Biolabs. Monoclonal mouse anti-RNApol antibody was from BioLegend. Monoclonal M2 anti-FLAG antibody, M2 anti-FLAG agarose beads, and 3xFLAG peptide were from Sigma-Aldrich. CaptureSelect-biotin, Streptavidin DyLight 800, and secondary antibodies goat anti-mouse IgG DyLight 800 conjugate and goat anti-rabbit IgG DyLight 680 conjugate were from Thermo-Fisher.
Bacterial strains and plasmids used in this study are listed in S4 Table. Primers for construction of strains and plasmids ere listed in S5 Table. The position and sequence of epitope tags introduced into SpaP, SpaR, and SpaS is shown in S7 Fig. All Salmonella strains were derived from S. Typhimurium strain SL1344 [37]. Bacterial cultures were supplemented as required with streptomycin (50 μg/mL), tetracycline (12.5 μg/mL), ampicillin (100μg/mL), kanamycin (25 μg/mL), or chloramphenicol (10 μg/mL).
The SpaP, and SpaPR complexes were expressed in E. coli BL21 (DE3) from rhamnose-inducible medium copy number plasmids encoding SpaPEPEA, SpaPQRFLAGS, or SpaPQRSTREP, respectively. Expression was autoinduced by over night growth at 37°C in TB medium. Bacterial cells were harvested, crude membranes purified as described previously [9], and membrane proteins were extracted with 1% DDM in PBS. After removal of unsolubilized material by ultracentrifugation for 30 min at 100.000 x g, complexes were immunoprecipitated according to the manufacturers instructions using CaptureSelect affinity gel for SpaPEPEA, M2 anti-FLAG agarose beads for SpaPRFLAG, and Strep-Tactin sepharose (IBA) for SpaPRSTREP. Complexes were natively eluted with 150 ng/ml SEPEA or 3xFLAG peptides, respectively, or with 2.5 mM desthiobiotin, each in PBS/0.04% DDM. The SpaPEPEA and the SpaPQRFLAGS, complexes were subsequently purified by anion exchange (Mono Q 5/50 GL, GE), while this step was omitted for the SpaPQRSTREP complex. Samples were further purified by size exclusion (Superdex 200 10/300 GL, GE) chromatography, and concentrated to 1 mg/ml using Amicon Ultra 100 k cutoff spin concentrators (Merck Millipore). Purified SpaP and SpaPR complexes were stored in liquid nitrogen until further use.
The detergent and polypeptide content of the purified SpaPRFLAG complex in PBS/0.04% DDM was determined by size exclusion chromatography—multi angle laser light scattering and analysis by the ASTRA software (Wyatt, Santa Barbara, CA) as previously described [38].
Purified SpaPRSTREP complex was concentrated to 20 μM in PBS/0.04% DDM, and buffer exchanged to 250 mM ammonium acetate, pH 7.5, complemented with 0.01% polyoxyethylene(9)dodecyl ether (C12E9) prior to native mass spectrometry analysis. Buffer exchange was carried out using Amicon Ultra 0.5 ml centrifugal filters with a 100-kDa cut-off (Millipore UK Ltd, Watford UK). Mass measurements were carried out on a Synapt G1 HDMS (Waters Corp., Manchester, UK) Q-ToF mass spectrometer [39]. The instrument was mass calibrated using a solution of 10 mg/ml cesium iodide in 250 mM ammonium acetate. 2.5 μL aliquots of samples were delivered to the mass spectrometer by means of nano-electrospray ionization via gold-coated capillaries, prepared in house [40]. Instrumental parameters were as follows: source pressure 6.0 mbar, capillary voltage 1.40 kV, cone voltage 150 V, trap energy 200 V, transfer energy 10 V, bias voltage 5 V, and trap pressure 1.63x10-2 mbar.
SpaP and SpaR TM topology was predicted using TOPCONS (http://topcons.cbr.su.se) [41]. The extent of the hydrophobic regions constituting TM helices was predicted using dGpred full portein scan (http://dgpred.cbr.su.se) [42] setting the minimal helix length to 18 and the maximal helix length to 31 aa. For visualization, the online tool PROTTER (http://wlab.ethz.ch/protter/start/) was used [43].
Analysis of type III-dependend secretion of proteins into the culture medium was carried out as described previously [20].
For protein detection, samples were subjected to SDS PAGE using SERVAGel TG PRiME 8–16% precast gels, transferred onto a PVDF membrane (Bio-Rad), and probed with primary antibodies anti-SipB, anti-InvJ, anti-PrgJ, anti-SpaP, anti-MBP, anti-RNApol, and M2 anti-FLAG. Secondary antibodies were goat anti-mouse IgG DyLight 800 conjugate and goat anti-rabbit IgG DyLight 680. EPEA-tagged SpaP was visualized using CaptureSelect-biotin anti C-Tag conjugate and Streptavidin DyLight 800. Scanning of the PVDF membrane and image analysis was performed with a Li-Cor Odyssey system and image Studio 2.1.10 (Li-Cor).
S. Typhimurium strains were grown at 37°C in LB broth supplemented with 0.3 M NaCl with low aeration to enhance expression of genes of SPI-1. For in vivo photocrosslinking of SpaPFLAG in Escherichia coli BL21 (DE3), bacteria were cultured at 37°C in LB broth. Cultures were supplemented with 500 μM rhamnose to induce expression of SpaPFLAG, SpaPFLAGQRS or SpaPQRFLAGS from low copy number pTACO10 plasmids [9]. To boost general SPI-1 expression, S. Typhimurium strains were transformed with pBAD24-hilA. Expression of the SPI-1 master regulator HilA was induced by addition of 0,05% arabinose to the cultures. Additionally the cultures were supplemented with the artificial amino acid para-benzoyl phenyl alanine (pBpa) to a final concentration of 1 mM and afterwards incubated for 5.5 h. 2 ODU of bacterial cells were harvested and washed once with 1 mL cold PBS. Cells were resuspended in 1 mL PBS and transferred into 6-well cell culture dishes. UV irradiation with λ = 365 nm was done on a UV transilluminator table (UVP) for 30 min.
10 OD units of bacterial lysates of S. Typhimurium or E. coli, respectively, were resuspended in 750 μl buffer K (50 mM triethanolamine, pH 7.5, 250 mM sucrose, 1 mM EDTA, 1 mM MgCl2, 10 μg/ml DNAse, 2 mg/mL lysozyme, 1:100 protease inhibitor cocktail), and incubated for 30 min on ice. Samples were bead milled and beads, unbroken cells and debris were removed by centrifugation for 10 min at 10.000 x g and 4°C. Crude membranes contained in the supernatant were precipitated by centrifugation for 45 min at 55,000 rpm and 4°C in a Beckman TLA 55 rotor. Pellets containing crude membranes were frozen until use.
1-dimensional blue native PAGE and 2-dimensional blue native/SDS PAGE of crude membranes was carried out as previously described [9].
S. Typhimurium ΔspaP or ΔspaPQRS mutants, respectively, transformed with pSUP, pSB3292, and pSB3398-based rhamnose-inducible low copy number plasmids containing SpaPFLAG amber mutants or SpaPQRS with SpaRFLAG amber mutants, respectively, were grown in 200 ml LB broth supplemented with 0.3M NaCl, 1 mM pBpa, 500 μM rhamnose, 0.02% arabinose, and appropriate antibiotics for 5 h at low aeration to express SPI-1 and assemble needle complexes. Purification of needle complexes was carried out as published previously [4,20,12] but LDAO was replaced by DDM (0.7% for lysis/extraction, 0.1% for maintenance) for lysis of cells and extraction of needle complexes throughout the protocol. Furthermore, an initial concentration of 35% (wt/vol) of CsCl was used to prepare the gradient. Purified needle complexes containing SpaPFLAG or SpaRFLAG with pBpa at desired positions were irradiated with UV light (365 nm) for 30 min to induce photocrosslinking to nearby proteins. Samples were subsequently analyzed by SDS PAGE, Western blotting, and immunodetection with M2 anti-FLAG antibodies. For MS analysis of crosslinked adducts, gel pieces at positions of observed crosslinks of pBpa-containing and control samples were cut out of Coomassie-stained SDS PAGE gels and subjected to in gel digestion.
For identification of crosslinked proteins, the area of a Coomassie-stained gel corresponding the position of the crosslinked band detected by Western blotting were excised and in-gel digested with trypsin [44]. For a better recovery, remaining proteins in the gel were again subjected to another tryptic digestion step. After each step extracted peptides were desalted using C18 StageTips [45]. Corresponding eluates were combined and subjected to LC-MS/MS analysis.
LC-MS/MS analyses were performed on an EasyLC II nano-HPLC (Proxeon Biosystems) coupled to an LTQ Orbitrap Elite mass spectrometer (Thermo Scientific) as decribed elsewhere [46] with slight modifications: The peptide mixtures were injected onto the column in HPLC solvent A (0.5% acetic acid) at a flow rate of 500 nl/min and subsequently eluted with a 106 min gradient of 5–33% HPLC solvent B (80% ACN in 0.5% acetic acid). During peptide elution the flow rate was kept constant at 200 nl/min. For proteome analysis, the 20 (Orbitrap Elite) most intense precursor ions were sequentially fragmented in each scan cycle using collision-induced dissociation (CID). In all measurements, sequenced precursor masses were excluded from further selection for 90 s. The target values for MS/MS fragmentation were 5000 charges and 106 charges for the MS scan.
The MS data were processed with MaxQuant software suite v.1.2.2.9 as described previously [47–49] with slight modifications. Database search was performed using the Andromeda search engine [48], which is part of MaxQuant. MS/MS spectra were searched against a target database consisting of 10,152 protein entries from S. Typhimurium and 248 commonly observed contaminants. In database search, full tryptic specificity was required and up to two missed cleavages were allowed. Carbamidomethylation of cysteine was set as fixed modification, protein N-terminal acetylation, and oxidation of methionine were set as variable modifications. Initial precursor mass tolerance was set to 6 parts per million (ppm) and at the fragment ion level 0.5 dalton (Da) was set for CID fragmentation. The MS data have been deposited to the ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org) via the PRIDE partner repository with the data set identifier PXD005028.
Sequence co-variation analysis was performed using EVcouplings [26–28] with pseudo-maximum likelihood approximation [50–52]. The multiple sequence alignment used as input for the model inference was created by jackhmmer 3.1 [53] (5 iterations) using the full sequence of Salmonella SpaP (UniProt: SPAP_SALTY, residues 1–224) as query against the November 2015 release of the UniProt Reference Cluster database (UniRef100) [54]. Sequences with more than 30% gaps are subsequently removed from the alignment. We then excluded alignment columns that contained 50% or more gaps from model inference and subsequent couplings predictions. Lastly, sequences were clustered at 80% sequence identity and then downweighted according to the cluster size to reduce redundancy. This resulted in an alignment of 5663 unique sequences with an effective number of 1080.4 non-redundant sequences (sequences/alignment length = 4.8) included in model inference and coupling prediction. The coupling scores of residue pairs were further normalized by estimating the background noise analogously to the procedure described in Hopf et al., 2014 [28]. Evaluation of the co-evolution prediction was done in the light of topology predictions obtained from deltaG, resulting in four predicted TM segments: (7, 38), (50, 75), (163, 193), (194, 211). Python (Python Software Foundation, http://www.python.org) and Ipython/Jupyter notebooks [55] were used for data analysis. The multiple sequence alignment, EC scores file, a contact map of the strongest couplings and an Ipython notebook of the analysis are available as supplement (S3 Table, S8 Fig, S2 and S3 Files).
Isolated SpaPRFLAG complexes were deposited on glow-discharged carbon coated copper-palladium grids and stained with 0.75% uranyl formate. Micrograph acquisition was performed on a FEI Tecnai F30 Polara at 300 kV, equipped with a Gatan Ultrascan 4000 UHS CCD (4k x 4k pixels, physical pixel size of 15 μm), using the LEGINON automated image acquisition system [56]. The corrected magnification was 71950x, resulting in a pixel size of 2.08 Å/pixel. 11202 particles were picked from the micrographs with EMAN2 boxer [57]. Particle images were first subjected to a maximum-likelihood classification and alignment (ML2D) in XMIPP [58] and then further processed in IMAGIC-5 (Image Science Software GmbH) through multi-reference alignment and classification by multi-variate statistical analysis.
SpaP or SpaPQRFLAG were moderately overexpressed in S. Typhimurium strain SB1770 (ΔprgHIJK, flhD::tet) from a rhamnose-inducible medium copy number plasmid by induction with 20 μM rhamnose. BM labeling was performed essentially as previously described [29], with minor modifications: After 3 h of induction, 0.2 ODU of bacterial cells were transferred to a fresh reaction tube and brought to the same volume by addition of fresh LB broth. Cells were labeled by addition of BM (EZ-link maleimide-PEG2-biotin, Thermo Pierce, final concentration 0.4 mM) for 30 min at room temperature with gentle agitation. The reaction was quenched by addition 2M β-mercaptoethanol to a final concentration of 10 mM. Cells were pelleted, re-suspended in SB buffer and incubated at 70°C for 10 min. BM labeling of proteins was analyzed by SDS PAGE, Western blotting, and detection of BM with streptavidin DyLight 800 dye (Thermo pierce). Scanning of the PVDF membrane and image analysis was performed with a Li-Cor Odyssey system and image Studio 2.1.10 (Li-Cor).
For subcellular fractionation, BM-labeled bacterial cells were pelleted by centrifugation. The culture supernatant was harvested and TCA precipitated. The bacterial cell pellet was resuspended and used to prepare the periplasmic and cytoplasmic fractions as described elsewhere. Briefly, pellets were resuspended by pipetting gently in ice-cold spheroplast buffer (40% sucrose, 33 mM Tris-HCl, pH 8.0) with freshly prepared lysozyme to a final concentration of 200 μg/ml, 50 μg/ml DNAse and 1.5 mM EDTA. The mixture was left on ice for 30 min with gentle stirring. Spheroplasts were stabilized by adding 20 mM MgCl2 and centrifuged at 3000 x g for 10 min at 4°C. The supernatant was transferred to ultracentrifugation tubes and centrifuged at 30 krpm for 30 min at 4°C in a Beckman TLA55 rotor to remove insoluble material. The supernatant (periplasmic fraction) was collected into fresh tube. The cytoplasmic fraction was prepared by resuspending the pellet of spheroplasts in 1 ml of 20 mM Tris-HCl, pH 8.0 and subsequent lysis by bead milling as described above. Lysates were transferred to ultracentrifugation tubes and centrifuged at 55 krpm for 45 min at 4°C in a Beckman TLA55 rotor. The supernatant (cytoplasmic fraction) was collected into fresh tubes.
|
10.1371/journal.ppat.1004059 | Caspase-1-Like Regulation of the proPO-System and Role of ppA and Caspase-1-Like Cleaved Peptides from proPO in Innate Immunity | Invertebrates rely on innate immunity to respond to the entry of foreign microorganisms. One of the important innate immune responses in arthropods is the activation of prophenoloxidase (proPO) by a proteolytic cascade finalized by the proPO-activating enzyme (ppA), which leads to melanization and the elimination of pathogens. Proteolytic cascades play a crucial role in innate immune reactions because they can be triggered more quickly than immune responses that require altered gene expression. Caspases are intracellular proteases involved in tightly regulated limited proteolysis of downstream processes and are also involved in inflammatory responses to infections for example by activation of interleukin 1ß. Here we show for the first time a link between caspase cleavage of proPO and release of this protein and the biological function of these fragments in response to bacterial infection in crayfish. Different fragments from the cleavage of proPO were studied to determine their roles in bacterial clearance and antimicrobial activity. These fragments include proPO-ppA, the N-terminal part of proPO cleaved by ppA, and proPO-casp1 and proPO-casp2, the fragments from the N-terminus after cleavage by caspase-1. The recombinant proteins corresponding to all three of these peptide fragments exhibited bacterial clearance activity in vivo, and proPO-ppA had antimicrobial activity, as evidenced by a drastic decrease in the number of Escherichia coli in vitro. The bacteria incubated with the proPO-ppA fragment were agglutinated and their cell morphology was altered. Our findings show an evolutionary conserved role for caspase cleavage in inflammation, and for the first time show a link between caspase induced inflammation and melanization. Further we give a more detailed understanding of how the proPO system is regulated in time and place and a role for the peptide generated by activation of proPO as well as for the peptides resulting from Caspase 1 proteolysis.
| Melanization is an important reaction in most multicellular organisms, both animals and plants. The initiation steps of this reaction in invertebrates are catalyzed by the prophenoloxidase (proPO) activating system a proteolytic enzyme cascade, which primary function is to recognize cell wall products from microorganisms and respond by activation of the system and generation of immune effector molecules. This cascade requires careful regulation to achieve spatial and temporal control to avoid dangerous side effects. We here show that a Caspase1-like enzyme can inactivate proPO when ppA is not activating the proPO to avoid deleterious effects and further we show for the first time that the N-terminal peptide from ppA cleavage of proPO (activation of proPO) has an important biological function as also the Caspase1 cleaved fragments. Our results also show that Caspase 1-induced inflammatory response is evolutionarily conserved and is linked to melanization.
| Melanization is the result of the oxidation of mono- and/or diphenols by a redox enzyme, often phenoloxidase, and it is an important reaction in most multicellular organisms, both animals and plants. Intruding microorganisms are frequently melanized in invertebrates, and during this process, low-molecular-weight phenolic substances are converted into polymeric melanin in a multi-step chain of reactions. The initiation steps of this reaction are catalyzed by the prophenoloxidase activating system (proPO system), and other steps occur spontaneously. The proPO system is a proteolytic enzyme cascade and its primary function is to recognize minuscule amounts (picograms per liter) of cell wall products from microorganisms (lipopolysaccharide (LPS), peptidoglycan (PGN) and glucans) and respond to the microorganism by activation of the system and the subsequent generation of immune factors. This cascade requires careful regulation to achieve spatial and temporal control to avoid dangerous side effects [1]. A number of regulatory factors are involved in controlling the activation of the proPO system, including proteinase inhibitors [2] and the melanization-inhibiting protein (in insects, crayfish and shrimp) [3], [4], [5].
The importance of melanization (proPO system) in controlling a number of specific host–pathogen encounters has been demonstrated over the past few years. One example of this is the bracovirus protein Egf1.0, which inhibits the prophenoloxidase (proPO)-activating proteinase in the insect Manduca sexta [6]. Two other recent examples are found in the parasitoid wasp Leptopilina boulardi, which targets the Drosophila phenoloxidase cascade by producing a specific serpin inhibitor [7], and in the bacterium Photorhabdus luminescens, which secretes a small organic molecule that acts as a negative regulator of PO activity [8]. In addition, a pathogenic Aeromonas hydrophila strain becomes highly virulent to crayfish when the PO transcript levels are experimentally reduced [9]. The proPO activation system, or melanization cascade, bears functional resemblance to the complement system, although the final reaction, melanization, is different. Intriguingly, recently we succeeded in showing that all of the steps in the proPO-cascade in Tenebrio molitor are shared with the proteinase cascade that leads to the activation of the Toll pathway for the production of antimicrobial peptides [10]. This shared cascade has been confirmed in several other insects [11], [12]. In the present study, we found that caspases are very important for the rapid degradation of proPO, which prevents oxidation in places where it is not appropriate.
Caspases, or cysteine-aspartic proteases, are a family of the cysteine proteases that are known for their function in apoptosis [13]. Some caspases are involved in the inflammatory system via the regulation of pro-inflammatory cytokines, and these caspases are necessary regulators of the unconventional secretion of leaderless proteins [14], [15]. In humans, caspase-1 is not only required for the activation of pro-interleukin (IL)-1β and pro-IL-18, but also functions as an activator of nuclear factor of the kappa-enhancer in B-cells (NF-κB) and p38 mitogen-activated protein kinase (MAPK) [16]. Interleukin-1β is produced as a cytosolic precursor and is dependent on caspase-1 cleavage for its activation and secretion [15], [17]. The proPO is also produced as a leaderless protein, most likely in the cytosol, and is secreted by an unknown mechanism. We, therefore, searched the sequence for caspase-1 cleavage sites and found two in the middle of the Cu-binding region. Therefore, we asked whether caspase-1-like cleavage of proPO is involved in the regulation of PO activity. We also asked whether the caspase-cleaved fragments have biological functions and whether these fragments are involved in immune functions even in the absence of PO activity. We also studied whether the peptide fragments generated by the cleavage of proPO into active PO by the prophenoloxidase activating enzyme (ppA), which gives a peptide of approximately 20 kD, might have some immune function during or immediately prior to melanization.
Our studies provide new information about the function of caspase-1-like activity in freshwater crayfish, in which it acts as a negative regulator of the proPO system. For the first time, we provide results showing that the fragments resulting from caspase or ppA cleavage have important biological functions.
ProPO, the inactive form of PO, is present in crayfish hemocytes, especially in the granular cells (GC). GCs are densely filled with granules, as indicated by their name. Upon activation by different environmental challenges such as microbes, exocytosis is induced, which causes the release of several proteins from the granules of the GCs and the release of proPO into the external milieu [18]. Immunostaining of proPO in GCs revealed that proPO is present in the cytoplasm but not in the granules, and not all GC cells express proPO (Figures 1A–C). ProPO is cleaved extracellularly to produce active PO upon activation by ppA. However, the mechanism by which proPO is released is still unknown. In beet armyworm and Drosophila, prostaglandins and JNK can stimulate cell lysis and subsequent proPO release [19], [20]. In mammal, there are many reports showing that inflammasomes and caspase-1 activation are involved in the secretion of proteins without signal peptides [21], [22]. Thus, we asked whether caspase-1 plays a role in proPO release and/or regulation. To answer this question, the presence of caspase-1 in the crayfish was examined. The transcriptome analysis of the freshwater crayfish P. leniusculus (unpublished data) revealed the presence of a translated amino acid sequence that has 36% identity to Drosophila caspase interleukin-1 beta converting enzyme (GenBank: NP524551). Additionally, by using an antibody against human caspase-1, two bands with sizes about 40 and 50 kDa were detected from crayfish hemocyte lysate (Figure 1D). These bands are probably two isoforms of the crayfish procaspase-1 like proteins. In comparison, in human six different isoforms of caspase-1 have been found. The 50 kDa procaspase-1 like protein was also detected in crayfish plasma and the level of this protein in plasma was decreased 1 h after an injection of E. coli or A. hydrophila compared to the control (0.15 M NaCl) (Figure 1E). Notably, when the 50 kDa band decreased, a 20 kDa band appeared in the plasma (Figure 1E). The 20 kDa band is similar in size to the p20 subunit of mammalian caspase-1, which is the active subunit of this protein. The active caspase-1 could only be detected in the supernatant and not in human keratinocyte lysates [15]. This is probably because it is rapidly degraded or released to the outside of cells and is therefore not present in cell lysates. Another explanation may be that active caspase-1 has a very short half-life, and therefore, its concentration under physiological conditions is very low [22], [23]. Caspase-1 activity was also found to be slightly increased in plasma at 1 h after injection with E.coli but no statistically significant difference could be observed (Figure 1F) and this activity could be decreased by incubation of the caspase-1 inhibitor, Z-YVAD-FMK.
Our results suggest that caspase-1-like activity is present in crayfish and that this activity can be activated during infection. As mentioned above, the activation of caspase-1 in vertebrates is involved in the secretion of several proteins, such as IL-1 β, but to date, no such mechanism has been identified in invertebrates, although the secretion of leaderless proteins such as proPO [18] and Pl-β-thymosins [24] has been observed. Therefore, the amino acid sequence of proPO was analyzed to determine whether there is a potential caspase-1 cleavage site in the proPO sequence. Two caspase-1 cleavage sites were found, after amino acids 363 and 389, which would give rise to two N-terminal proPO fragments with predicted sizes of approximately 42 kDa (proPO-casp1) and 45 kDa (proPO-casp2), respectively (Figure 2A). These two cleavage sites are located downstream of the cleavage site for ppA and would give rise to a small N-terminal fragment (20 kDa of proPO-ppA) and a C-terminal active PO [25]. Therefore, cleavage as the result of caspase-1-like activity has the potential to reduce the PO activity and act as a negative regulator of the proPO system.
To investigate whether any proPO-caspase fragment (proPO-casp) is released from hemocytes, plasma proteins from bacteria-infected crayfish were subjected to western blotting. The results in Figure 2B show that two caspase-cleaved proPO fragments were present in the plasma after bacterial infection and that the levels of both proPO-casps increased with time, whereas the plasma proPO level decreased. Moreover, the level of proPO-casp1 was higher than that of proPO-casp2 at all time points. This result suggests that proPO-casp2 may be degraded faster than proPO-casp1 in vivo. When non-virulent E. coli was injected, the released proPO was rapidly cleaved by a caspase, and high levels of the fragments were detected at 1 and 3 hours post injection. In contrast, the injection of a virulent A. hydrophila strain resulted in slower caspase cleavage, and a high level was reached after 3 hours (Figure 2B). Interestingly, when the levels of the C-terminal fragments of proPO in the samples were analyzed, neither active PO, nor proPO-casps could be detected, and only inactive proPO was found in the plasma. This result suggests that the C-terminal fragments of proPO produced by caspase-1-like cleavage, as well as ppA cleavage, are degraded rapidly.
Because cleavage by caspase-1 may reduce PO activity, the enzyme activity in the plasma was measured after bacterial infection (Figure 2C). The PO activity decreased after E. coli infection, but there was no significant difference, and interestingly, the plasma PO activity was significantly higher at 1 h after A. hydrophila infection, at time at which the plasma proPO-casp levels were low, and the enzyme activity markedly decreased by 3 h, when the levels of these fragments were higher. Notably, all animals died 4–6 hours after A. hydrophila infection.
Ca2+ has been reported to induce exocytosis in crayfish hemocytes [18] and regulates inflammasome activation and thus caspase-1 activation [17], [26]. Therefore, we investigated the effect of Ca2+ on proPO-casp release in in vitro experiments with isolated GCs. As shown in Figure 3A, the release of proPO and both proPO-casp fragments was Ca2+ and time dependent. When the antibody against the C-terminus of proPO was used, proPO but not active PO or proPO-casps could be detected. Again, this result suggests that the C-terminal fragments of proPO are rapidly degraded.
To confirm that the proPO-casp fragments are the result of a caspase-1-like cleavage, the effect of the caspase-1 inhibitors Z-YVAD-FMK or Ac-WEHD-FMK on the release of proPO-casp fragments was examined. The results presented in Figure 3B clearly show that the production of proPO-casps was markedly decreased in the presence of Z-YVAD (Figure 3B) or Z-WEHD-FMK (Figure S1). The amount of released proPO into the medium was also decreased in the presence of Z-YVAD-FMK, and a higher level of proPO in granular cell lysate was observed when the cells were incubated with caspase-1 inhibitors (Figure S1).
Furthermore, dsRNA caspase-1 treatment of granular hemocytes caused a complete reduction of the caspase-1 like transcript (Figure 3C), but no obvious reduction in protein level could be observed (data not shown). However, when the cells were treated with Ca2+ for 3 h to induce release of caspase-1 at 65 h of dsRNA treatment, the caspase-1 knockdown cells fail to produce new procaspase-1 like protein after another 24 h culture in L-15 (Figure 3D). The lower level of caspase-1 like protein resulted in a reduction of the levels of proPO-casp fragments both in cell lysate and medium. No change in total proPO protein could be observed after the RNAi treatment of granular cells (Figure 3D).
Because putative caspase-1 cleavage products were clearly detected outside GCs in vitro and in vivo, we decided to determine if these fragments possess biological functions. The N-terminal parts of proPO produced by cleavage at the putative caspase-1 cleavage site between Asp363 and Ala364 (proPO-casp1), and the fragment produced by cleavage between Asp389 and Asn390 (proPO-casp2) were produced as recombinant proteins with estimated sizes of 43 and 47 kDa, respectively. In addition, the N-terminal peptide fragment of proPO generated by cleavage by ppA between Arg176 and Thr177 [25] was produced (proPO-ppA).
To determine whether any of the proPO fragments are involved in the immune system, the bacterial clearance activities of these proteins were assessed. The fragments were mixed with bacteria and then injected into crayfish. Bacterial number in hemolymph was examined at 40 min and 3 h post injection. After 40 min, the E. coli titer was already significantly decreased in the proPO-ppA, proPO-casp1, and GFP treatment groups, whereas the injection of proPO-casp2 had no significant effect on the number of E. coli (Figure 4A). Because GFP also caused reduction of bacterial number at this time point, this might be a general protein effect. However, at 3 h post injection, the numbers of E. coli in the proPO-ppA, proPO-casp1, and proPO-casp2 injection groups were significantly lower than that in the non-protein injection groups and the GFP group (Figure 4B).The antimicrobial activities of all three fragments were then tested in vitro to determine if the bacterial clearance was caused by the proteins themselves or if other components were involved in the clearance process. The titer of E. coli decreased significantly after proPO-ppA treatment compared with the non-protein treatment, whereas the other fragments had no significant effect and did not exhibit any antibacterial activity (Figure 5A). When the treated bacteria were observed under the microscope, very strong agglutination was detected after treatment with the proPO-ppA peptide, whereas no signs of agglutination occurred with the proPO-casp fragments or GFP (Figure 5B). The minimal agglutinating concentration for proPO-ppA was the lowest for E. coli and Staphylococcus aureus, and the proPO-ppA fragment appeared to have the ability to agglutinate all of the tested bacterial species (Table 1). When the bacteria were observed by SEM after 15 and 40 min of incubation, we could see that proPO-ppA disrupted the E. coli cell morphology, causing the cell walls to shrink. After 15 min of incubation, the E. coli treated with proPO-ppA started to show signs of cell wall disruption, and a longitudinal line was observed (Figure 5C), in contrast to the GFP-treated bacteria (Figure 5D). After 40 min of incubation, the E. coli treated with proPO-ppA clearly formed clumps (Figure 5F), and the cells were flat (Figures 5H and 5I).
Then, a bacteria viability assay was performed to determine if the strong agglutination killed the bacteria. Fluorescence microscopy clearly revealed that the proPO-ppA fragment greatly decreased the cell viability compared with the control treatments as measured by the red staining of dead bacteria (Figure 6). A few agglutinated bacteria were stained with only SYTO9 and appeared green in the merged picture (live cells).
The proPO system is an important innate immune response and is composed of a cascade of proteinases that terminates with the activation of the proenzyme proPO. After proteolytic cleavage, proPO becomes an active redox enzyme, PO, which forms melanin and other antimicrobial products in the non-catalytic pathway from quinone to melanin [1], [27], [28]. Because the product of PO is highly toxic, it is necessary to keep the proPO system under strict control to avoid deleterious effects of an activated proPO system, principally the redox enzyme PO. Several factors that can control this system have been described, including a multitude of proteinase inhibitors [1], [28], [29]. Moreover, if PO is generated, the melanization inhibition protein (MIPs) can inhibit melanin formation [4], [5]. Another way to protect against the inappropriate activation of proPO is to keep this proenzyme and its activation cascade in separate subcellular compartments. Thus, all arthropod proPOs are produced as leaderless proteins and are presumably located in the cytoplasm, whereas the activation system (proPO-AS) is located in secretory granules in crustaceans. This arrangement is similar to that of IL-1β, which is formed as a precursor in the cytoplasm and is then released to the outside of the cell during or after activation. Because caspase-1 cleavage is necessary for this activation and release steps and because such cleavage has been shown to be of importance for the secretion of several leaderless proteins [5], we looked for caspase-1 cleavage sites in proPO. We identified a new important regulator of the proPO system, caspase-1-like activity, which can efficiently cleave proPO at two cleavage sites and make the enzyme catalytically inactive. Moreover, we found that the N-terminal products of this cleavage have effects on bacterial clearance. There are no previous reports of inflammasomes in invertebrates, and NOD-like receptors (NLRs), which are part of the vertebrate caspase-1-activating inflammasome complex, have not been found in invertebrate genomes except that of the sea urchin [17]. However, both vertebrates and invertebrates express several pattern recognition receptors, and there might be other still undiscovered inflammasome sensor molecules responsible for the activation of invertebrate caspase-1-like activity. Recently, the structure of a Drosophila apoptosome composed of the Apaf-1-like protein Dark was reported. After binding to Dark, the initiator caspase Dronc cleaves the caspase DrICE and initiates an intrinsic cell death pathway [30]. The NLR-inflammasome and different apoptosomes are all examples of the oligomerization of CARD domain proteins involved in caspase activation, and their roles in cell death and immune responses are only beginning to be understood. Our discovery may add the proPO system to the list of immune responses balanced by such caspase regulation.
The activation of proPO by ppA occurs via proteolytic cleavage near the N-terminal, and a peptide of approximately 20 kDa is released during this process. Once activated, the PO activity must be strictly localized to where melanization is needed. In analogy with the complement system in vertebrates, we asked whether the cleaved activation peptide also has biological activities, as C3-cleaved peptides do.
To study whether this peptide (proPO-ppA) and the fragments resulting from caspase cleavage are involved in bacterial clearance, we injected E. coli together with these different fragments separately into crayfish and measured the number of bacteria in the hemolymph after the injection. Interestingly, all three peptides had the ability to decrease the bacterial number in the hemolymph compared to the injection of a control protein. To further investigate the mechanisms of action of these three peptides, we incubated each peptide directly with E. coli to identify any putative antibacterial activity, and we observed clear antibacterial activity for the proPO-ppA peptide. The observation of antibacterial activity in vitro suggests that the observed antimicrobial activity was a result of this peptide itself, whereas the proPO-casp1 and proPO-casp2 seem to require other components in crayfish to promote bacterial clearance. After incubation with the proPO-ppA fragment, both Gram-negative and Gram-positive bacteria were found to be heavily agglutinated. Moreover, live/dead staining of proPO-ppA-incubated E. coli revealed that the agglutinated bacteria contained a high percentage of dead cells. The generation of a highly antibacterial peptide after cleavage is similar to the case of hemocyanin, for which proteolytic cleavage in the crayfish plasma produces astacidin 1 [31]. These findings correspond to the antimicrobial function of human eosinophil cationic protein (ECP). Incubation with ECP can cause bacterial agglutination and decreased viability. One further example is the C-terminal region of human extracellular superoxide dismutase (SOD), which also exhibits antimicrobial activity against Gram-negative and Gram-positive bacteria [32], [33].
The SEM study showed that the antibacterial effect of proPO-ppA seemed to be on the bacterial cell wall. The cell wall appeared to shrink in the presence of proPO-ppA and then the E. coli cells were flattened. This antimicrobial activity is similar to the activity of funnel web spider venom on Shigella sp. [34].
We present new findings that may explain how the proPO system is regulated and how its activity is localized (Figure 7). After ppA cleavage, the small N-terminal proPO-ppA peptide causes the agglutination of bacteria at the site of infection, and PO activity then may localize melanization to these bacterial aggregates. We also provide evidence that the release of proPO, a leaderless protein, from cells may involve caspase-1-like activity, similar to that regulating IL-1βrelease. Furthermore, caspase-1-like cleavage of proPO inactivates the enzyme and generates two N-terminal fragments with bacterial clearing activity. These findings show that proPO is a multifunctional protein, with a phenoloxidase in the C-terminal region and an agglutinating and antimicrobial peptide in the N-terminal region, as well as N-terminal proPO-casps peptides with distinct biological activities.
Freshwater crayfish (P. leniusculus) was purchased and reared in a closed system at 10°C. Only healthy animals were used for the experiments.
The antibody against human procaspase-1 and p20 subunit was purchased from Invitrogen. The antibody against kazal protease inhibitor (KPI) was from Santa Cruz Biotechnology (sc-46652). The ECL peroxidase-linked donkey anti-rabbit IgG (species-specific whole Ab) was purchased from GE Healthcare. The peroxidase-linked anti-goat IgG antibody (whole molecule) was purchased from Sigma. The FITC-conjugated goat anti-rabbit IgG (whole molecule, affinity isolated antigen-specific antibody) was purchased from Sigma.
To produce antibodies against the N-terminal and C-terminal peptides of proPO, proPO-N 1–76 and proPO-C 621–694 were cloned into the bacterial expression vector pGEX-4T-1 (GE Healthcare). Then, these plasmids were subsequently transformed into Escherichia coli cells (BL21), and a single colony was grown in LB medium containing 100 µg/ml ampicillin to OD600 = 0.5 and induced with 0.2 mM isopropyl β-d-thiogalactoside (IPTG) for 6 h at 20°C. Recombinant GST-fusion proteins were purified using GSTrap FF columns (GE Healthcare), and the GST tag was removed on the column by incubation with thrombin (GE Healthcare) at 4°C overnight. Then, the free recombinant peptides were eluted with PBS from the column. Two milligrams of recombinant proPO-N or proPO-C was used for the production of rabbit antiserum. The anti-proPO-N and anti-proPO-C antibodies were purified from the rabbit antiserum using GammaBind G-Sepharose (GE Healthcare) following the manufacturer's instructions.
GCs were separated using a 70% Percoll gradient in 0.15 M NaCl. The separated cells were resuspended in 0.15 M NaCl, seeded on coverslips, fixed and treated as described previously [35]. The immunostaining was performed using an antibody against the N-terminus of proPO (5 ng/µl, 1∶160) and a FITC-conjugated anti-rabbit antibody (1∶300). In addition, an antibody against a crayfish kazal protease inhibitor (KPI) was also used to counter stain hemocyte granules. The slides were mounted with VECTASHIELD mounting medium containing DAPI (Vector Laboratories). The stained cells were then observed under a fluorescence microscope.
The hemolymph was centrifuged at 1000× g for 5 min at 4°C, and the hemocyte pellet was collected and washed two times with PBS. Then, hemocytes were lysed in PBS containing 2% Triton X-100 [15] and 1× protease inhibitor cocktail (Complete, Mini, EDTA-free, Roche). The cell lysate was centrifuged at 15,000× g for 15 min at 4°C, and the supernatant was collected. The protein concentration was determined, and 20 µg of protein was mixed with Laemmli sample buffer (62.5 mM Tris-HCl, 2% SDS, 10% (v/v) glycerol, 0.1 M DTT, 0.01% bromophenol blue, pH 6.8).
The presence of caspase-1-like protein was first examined in crayfish hemocyte lysate (20 µg protein) by western blotting using an antibody against human caspase-1. The hemocyte lysate was prepared as described above. In addition, the amount of the caspase-1-like protein in crayfish plasma was also determined in bacterial injected crayfish since bacterial components have been reported to be activators of inflammasomes and caspase-1 [36], [37]. To perform this experiment, crayfish (N = 3 for each group) were injected with 0.15 M NaCl, non-virulent E. coli or highly virulent A. hydrophila B1 [38], and hemolymph was collected at 1 h after injection. Hemocytes were removed from the hemolymph by centrifugation at 1000× g for 5 min at 4°C. Then, 200 µg of plasma protein was loaded onto 12.5% SDS-PAGE gels and subjected to western blotting as previously described [35]. The presence of caspase was detected using a rabbit anti-caspase antibody (1∶3000) and ECL peroxidase-linked donkey anti-rabbit IgG (GE Healthcare) (1∶7500). The detection of actin was also performed as a loading control using a goat anti-actin antibody (1∶5000) and an anti-goat secondary antibody (1∶5000).
To determine caspase-1 activity, Ac-YVAD-pNA (Santa Cruz), a synthetic peptide and substrate for caspase-1, was used. Hemolymph was collected at 1 h after injection of 0.15 M NaCl, E. coli or A. hydrophila B1. Cell-free plasma samples were prepared as described above. Then 50 µl of plasma was mixed with 200 nM of Ac-YVAN-pNA in the presence or absence of the caspase-1 inhibitor, Z-YVAD-FMK (50 µM). The mixtures were incubated at 37°C for 1.5 h, and the absorbance was determined at 405 nM. The plasma without substrate was used as a negative control for each sample. The caspase-1 activity was reported as OD405/g plasma protein.
To determine if there are any potential caspase-1 cleavage sites in P. leniusculus proPO, the amino acid sequence of proPO (GenBank: CAA58471) was analyzed using the bioinformatics tool PeptideCutter (http://web.expasy.org/peptide_cutter/).
E. coli and A. hydrophila B1 were used to induce PO activity in vivo. Bacteria (1–3×107 CFU/100 µl) suspended in 0.15 M NaCl were injected into the crayfish (N = 5). The hemolymph was then collected before and 1 and 3 h after injection and centrifuged at 1000× g for 5 min to remove the hemocytes. Then, 30 µl of the cell-free plasma was used in a PO activity assay. The plasma was incubated for 30 min at room temperature (RT) with 20 µl of 3 mg/ml L-DOPA and 50 µl phosphate-buffered saline (PBS). The PO activity was determined by monitoring the absorbance at 490 nm, and a reaction mixture without substrate served as the baseline.
Cell-free plasma (250 µl) was centrifuged at 110,000× g at 4°C for 1.5 h to remove hemocyanin and 180 µl of the supernatant was subjected to acetone precipitation. Then, 1.6 µg of protein from each sample was loaded onto an SDS-PAGE gel, and the proPO-casps were detected by western blotting as described above using antibodies against the N-terminus or C-terminus of proPO.
Because proPO is highly expressed in GCs, separated GCs were used in these experiments. The GCs were resuspended in 0.15 M NaCl and seeded into 96-well plates. After attachment for 10 min, the cells were incubated at RT in 10 mM HEPES-0.2 M NaCl buffer (pH 6.8) containing different concentrations of CaCl2 (0, 1, or 10 mM). Then, 50 µl of buffer was collected from each well after 30 and 60 min and subjected to TCA precipitation. The protein pellets were dissolved in the same volume of Laemmli sample buffer and analyzed by western blotting using antibodies against the N-terminus (1∶3000) or C-terminus (1∶3000) of proPO as described above.
To investigate the effect of the caspase-1 inhibitors, Z-YVAD-FMK (Tocris Bioscience) and Ac-WEHD-FMK (Santa Cruz) on proPO-casp release, GCs were incubated with 75 µl of HEPES-NaCl containing the inhibitor (0, 1, 10, or 50 µM) for 30 min before the addition of 75 µl of HEPES-NaCl containing 2 mM CaCl2. Then, the buffer was collected at 30 and 60 min, and the samples were prepared and analyzed as described above.
To knockdown crayfish caspase-1, double-stranded RNA (dsRNA) was synthesized with MEGAScript Kit (Ambion) and the template was amplified using the following primers (T7 promoter sequence is in italic); for dsCaspase-1 5′-TAATACGACTCACTATAGGGACCTGTGGACCGACCTAGTGC-3′ and 5′-TAATACGACTCACTATAGGGGTCGGGCCTTTAGTTGGACACC-3′, dsGFP: 5′-TAATACGACTCACTATAGGGCGACGTAAACGGCCACAAGT-3′ and 5′-TAATACGACTCACTATAGGGTTCTTGTACAGCTCGTCCATGC-3′. The purified dsRNAs (1 µg/well) were then added to granular cells and maintained in 25% L-15 medium in 0.15 M NaCl at 16°C for 65 h (50% of medium was changed once at 48 h). The cells were then treated with 10 mM Ca2+ for 3 h to induce caspase-1 release. Next the cells were washed with 25% L-15 medium four times and then maintained in the medium for another 24 h to allow the cells to produce new caspase-1. Later, the cells were treated with 10 mM Ca2+ again, and same volume of medium was collected from each well at 60 min after incubation. In addition the granular cell lysate was also prepared from the 60-min Ca2+ treated cells. The samples were subjected to SDS-PAGE and the proPO-casp fragments and caspase-1 like protein were detected by western blotting.
ProPO-ppA was amplified from P. leniusculus hemocyte cDNA using the following primers: proPO32EcoRI-F, (5′ TTTTTTGAATTCCAGGTGACCCAGAAGTTGCTGAGGA 3′) and proPOppA-R, (5′ CGCCTCGAGCTACCTGTTCACTTCAACCTGCATGCTT 3′). The PCR product was visualized by agarose gel electrophoresis, extracted from the gel and purified before being ligated into the pET32a expression vector between the EcoRI and XhoI restriction sites. The protein was expressed in E. coli BL21(DE3)pLysS cells. After the IPTG induction, the proteins, which were expressed in inclusion bodies, were refolded and purified with Ni-affinity chromatography.
Further, proPO-casp1 and proPO-casp2 were amplified using proPO-F (5′ CATGCCATGGGCCATCATCATCATCATCATCAGGTGACCCAGAAGTTGCTGAGGA 3′) and proPO-casp-R1 (5′ CGCCTCGAGCTAATCTGCCTCAAACGCGTCTCCTAAG 3′) for proPO-casp1 and proPOcasp-R2 (5′ CGCCTCGAGCTAGTCGTGACAGAATGCCAGCAGCACA 3′) for proPO-casp2. Both PCR products were ligated into the pET28b expression vector between the NcoI and XhoI restriction sites and expressed in the E. coli expression system as described above. Bacterial cells were disrupted by sonication, and the recombinant proteins were purified with Ni-affinity chromatography and dialyzed against 20 mM Tris-HCl, pH 8.0, at 4°C.
Bacterial and protein injections were performed as follows. Briefly, wild-type E. coli were harvested at the mid-log phase, washed six times with 150 mM NaCl at 1200×g for 5 min and resuspended in 150 mM NaCl at 1×109 CFU/ml. Bacterial suspensions (100 µl) were mixed with 20 µg of recombinant protein or 20 mM Tris-HCl and injected into the crayfish at the base of a walking leg. At 40 min and 3 h after injection, hemolymph was collected, serially diluted with 150 mM NaCl and plated on LB agar. The plates were incubated at 37°C overnight. The number of bacterial colonies was counted, and the number of CFU per ml was calculated for each treatment.
To investigate whether the bacterial clearance activity was a direct effect of the protein fragments, we performed an in vitro bacterial clearance assay as follows. E. coli were prepared as described above, and 100 µl of resuspended E. coli was mixed with 20 µg of recombinant protein. Then, the volume was adjusted to 1 ml with 150 mM NaCl. The mixtures were incubated for 1 hour at room temperature with mild agitation, serially diluted and plated onto LB agar to calculate the CFU per ml. The plates were observed under a microscope.
The minimum protein concentration for bacterial agglutination was tested as previously described [39]. The bacteria used in the experiment were Staphylococcus aureus Cowan, Micrococcus luteus M III, E. coli D21, A. hydrophila B1, and Pseudomonas aeruginosa OT97. Overnight cultures of bacteria were collected and washed three times in 150 mM NaCl. Each bacterial species was resuspended, and the optical density was adjusted to 2. The recombinant proteins were twofold serially diluted, and 50 µl of each dilution was mixed with 50 µl of bacterial suspension and incubated at room temperature for 1 hour.
E. coli and protein mixtures were prepared as described for the in vitro bacterial clearance assay. After 5 min of incubation, SYTO9 (Invitrogen) was added to a final concentration of 50 nM, and propidium iodide (PI) was added to a final concentration of 1 µg/ml. Then, the samples were visualized with a fluorescence microscope.
E. coli at O.D. 0.5 (100 µl) was incubated with 20 µg of the proPO-ppA peptide for 40 min at room temperature with mild agitation. After incubation, the bacteria were harvested and fixed with glutaraldehyde following standard procedures for SEM.
|
10.1371/journal.ppat.1007885 | Peer pressure from a Proteus mirabilis self-recognition system controls participation in cooperative swarm motility | Colonies of the opportunistic pathogen Proteus mirabilis can distinguish self from non-self: in swarming colonies of two different strains, one strain excludes the other from the expanding colony edge. Predominant models characterize bacterial kin discrimination as immediate antagonism towards non-kin cells, typically through delivery of toxin effector molecules from one cell into its neighbor. Upon effector delivery, receiving cells must either neutralize it by presenting a cognate anti-toxin as would a clonal sibling, or suffer cell death or irreversible growth inhibition as would a non-kin cell. Here we expand this paradigm to explain the non-lethal Ids self-recognition system, which stops access to a social behavior in P. mirabilis by selectively and transiently inducing non-self cells into a growth-arrested lifestyle incompatible with cooperative swarming. This state is characterized by reduced expression of genes associated with protein synthesis, virulence, and motility, and also causes non-self cells to tolerate previously lethal concentrations of antibiotics. We show that temporary activation of the stringent response is necessary for entry into this state, ultimately resulting in the iterative exclusion of non-self cells as a swarm colony migrates outwards. These data clarify the intricate connection between non-lethal recognition and the lifecycle of P. mirabilis swarm colonies.
| A resident of animal intestines, Proteus mirabilis is a major cause of catheter-associated urinary tract infections and can cause recurrent, persistent infections. Swarming, which is a collective behavior that promotes centimeter-scale population migration, is implicated in colonization of bladders and kidneys. A regulatory factor of swarming is kin recognition, which involves the transfer of a self-identity protein from one cell into a physically adjacent neighboring cell. However, how kin recognition regulates swarming was previously unclear. We have now shown a mechanism linking kin recognition, swarm migration, and antibiotics tolerance: cells induce a transient antibiotics-tolerant, persister-like state in adjacent non-identical cells which in turn prevents non-identical cells from continuing to participate in collective swarming. These affected non-identical cells continue to exhibit large-scale gene expression suggesting an active shift into a different expression state. These data provide two key insights for the field. First, kin recognition can be a regulatory mechanism that acts with spatial and temporal precision. Second, induction into an antibiotics-tolerant state, instead of occurring stochastically, can be physically and spatially regulated by neighboring cells. These insights highlight the importance of further developing four-dimensional (time and X-, Y-, Z-axes) model systems for interrogating cell-cell signaling and control in microbial populations.
| Organisms rarely live in complete isolation. Living in a community can provide benefits to each individual. However, there is a constant balance between the interests of individuals and the maintenance of community-wide advantages. A stable evolutionary strategy is for individuals to preferentially direct advantages to close kin [1–3]. This behavior, known as kin discrimination, has been the subject of focused study.
Several examples of kin discrimination in bacteria have been elegantly described, including those mediated by Type IV [4], Type VI [5, 6], and Type VII [7] secretion system based effector exchange, contact-dependent inhibition (CDI) [8, 9], and the MafB toxins of the Neisseria [10]. One common thread between these systems is that they characterize discrimination as immediate antagonism towards cells or strains that are non-kin, typically through delivery of lethal toxin effector molecules. Upon effector delivery, receiving cells must either neutralize it by presenting a cognate anti-toxin or suffer immediate negative consequences, typically cell death [5] or permanent inhibition of growth [8]. Here we describe an expansion of these mechanisms: the Ids self-recognition system mediates kin discrimination in Proteus mirabilis by selectively inducing non-self cells into a growth-arrested lifestyle incompatible with social behavior, thereby controlling access to that behavior.
P. mirabilis, a major cause of recurrent complicated urinary tract infections [11], engages in several sophisticated social behaviors such as swarming on rigid surfaces. Swarms are formed by many elongated (~ 10–80 μm) “swarmer” cells moving cooperatively, allowing for colony expansion over centimeter-scale distances. Rounds of swarming are interspersed with periods of non-expansion termed “consolidation”. The oscillation between swarming and consolidation leads to a characteristic pattern of concentric rings on higher percentage agar plates [12]. Effective P. mirabilis swarming relies on the ability of swarmer cells to form large rafts that together move much more quickly than isolated individuals [13]. Rafts are fluid, transient collectives that cells frequently enter and exit. As such, an individual cell interacts with many different neighbors through the lifetime of a swarm. During swarming, P. mirabilis cells can communicate with each other by exchanging proteins through contact-dependent secretion systems [14, 15]. These signals in turn cause emergent changes in swarm behavior [16, 17].
P. mirabilis swarms exhibit the ability to recognize self in several ways. The oldest known example is Dienes line formation: two swarms of the same strain merge into a single swarm upon meeting, while two swarms of different strains instead form a human-visible boundary [18, 19]. More recently, the phenomenon of territorial exclusion was described: in a mixed swarm comprising two different strains, one strain is prevented from swarming outwards by the other [15]. Clonal swarms of P. mirabilis have a coherent self identity, minimally mediated by the Ids system encoded by six genes idsA-F. Deletion of the ids locus results in the mutant strain no longer recognizing its wild-type parent as self [20]. Several of the molecular mechanisms governing Ids-mediated self recognition have been described in detail. However, how the Ids system functions in local behaviors has remained elusive.
Briefly, two proteins, IdsD and IdsE, govern self identity. IdsD is transferred between cells in a Type VI secretion system (T6SS)-dependent fashion; disruption of the T6SS prevents all Ids signal transfer [17, 21]. A cell in a swarm is considered to be self if it produces an IdsE protein that can bind IdsD proteins sent from neighboring cells. Disruption of these IdsD-IdsE interactions, either through deletion of idsE or through production of non-self IdsD or IdsE variants, result in strains that display extreme attenuation in swarm expansion without loss of viability [16, 17]. Fig 1A shows cartoon representations of Ids-mediated self recognition, where endogenous IdsD and IdsE are represented by capital letters “D” and “E”, and a non-self variant of IdsD is represented by lowercase “d.” The IdsD protein must be incoming: endogenous IdsD and IdsE proteins produced within a single cell do not impact self-recognition behaviors [21]. We describe conditions that lead to non-self recognition as “Ids mismatch.”
Crucially, territorial exclusion by Ids does not affect viability. Excluded cells grow and divide at a comparable rate to non-excluded cells when isolated from swarms [17]. While Ids has been described as a toxin-antitoxin system [22, 23], this characterization is inconsistent with experimental data and is likely due to the reliance on the T6SS for transport of IdsD [17, 21]. The T6SS is often characterized as a lethal toxin delivery mechanism [24]. A mechanistic description of Ids-mediated recognition is needed to reconcile the data and would provide a model for other non-lethal mechanisms that might be attuned for surface-dwelling swarm migration.
Here we show that even though an Ids recognition signal transfer happens while cells are actively migrating as a swarm, the recognition response is delayed until cells have stopped moving. We show that recognition of non-self is at least partially mediated by ppGpp levels within the cell, and this contributes to a concerted shift of cells into a distinctive, antibiotic-tolerant state that is incompatible with continuation of swarming. We found that this Ids-induced state is short-acting; induction requires continuous cell-cell interactions. In the context of a swarm, the collective consequence is an iterative winnowing of the non-self cells from the swarm fronts during periods of no active migration. We posit that the cell-cell communication of these non-lethal factors therefore acts as a control system during swarm expansion by diverting non-self from developing into swarm-compatible cells and thus preventing non-self cells from taking part in cooperative swarm behavior.
We initially sought to determine the method through which Ids caused non-self cells to be territorially excluded from swarms. Given the lack of lethality and the stark attenuation of swarm colony expansion observed during Ids-mediated territorial exclusion [15, 17], we hypothesized that an Ids mismatch caused broad changes in gene expression of the recipient cell. Ids mismatch is defined here as transcellular communication of IdsD to a recipient cell lacking a cognate IdsE. Therefore, we performed RNA-Seq differential expression analysis using conditions that would produce either self or non-self interactions, all within genetically equivalent backgrounds. We isolated total RNA from cells undergoing consolidation, because consolidation is the swarm development stage most tightly connected to major transcriptional changes [25].
As a baseline, we compared transcriptional profiles between cells from clonal swarms of wildtype and independently, of a derived mutant strain lacking the ids genes (BB2000::idsΩCm [20], herein referred to as “Δids”). The swarm colonies of both strains expand equivalently and have no notable morphological differences [20]. A complete list of genes with differential transcript abundance is found in S1 Table. Overall, few differences were apparent between clonal swarms of wildtype and the Δids strain. 10 genes were found to be significant (fold-change > log2 1.5, p < 0.05), six of which were the ids genes deleted in the construction of the Δids strain (Fig 1B).
We next considered differences in strains experiencing Ids mismatch. We examined three different conditions: a clonal swarm in which every cell lacked the idsE gene (CCS06), a clonal swarm in which every cell is a chimera containing an IdsE protein unable to bind transferred IdsD proteins (CCS02), and cells of a Δids-derived strain constitutively producing Green Fluorescent Protein GFPmut2 (Δids-GFP) that were isolated from a 1:1 co-swarm with wildtype through fluorescent-activated cell sorting (FACS), henceforth termed “co-swarmed Δids". Strains CCS06 and CCS02 are the Δids background containing a plasmid expressing the ids operon under its native promoter; mutations to the ids genes were made in the plasmid-based allele. All strains have previously been verified and characterized [16, 17, 20, 21]. As the two clonal swarm colonies have attenuated expansion, we were only able to harvest whole colonies as visible consolidation phases were less distinct. RNA-Seq differential expression analysis was performed on cells from each condition as compared to appropriate control samples: co-swarmed Δids were compared with clonal Δids, and CCS02 and CCS06 swarms were compared to a clonal Δids-derived swarm in which cells expressed plasmid-encoded ids genes (CCS01, Δids pidsBB). Large-scale changes to relative transcript abundances were apparent for each condition: 231 genes in CCS06, 457 genes in CCS02, and 836 genes in co-swarmed Δids, which represents approximately 6%, 13%, and 23% of total genes, respectively (Fig 1, S2 Table, S3 Table, S4 Table).
General trends were apparent even though differences in relative abundance of transcripts were present in a diverse and widespread range of genes. We observed a concerted decrease in transcripts for class I, II, and III genes for flagellar synthesis such as flhDC, filA, and fliC. We also observed a decrease in transcripts for many genes associated with protein synthesis, such as the 50S ribosomal protein rplT and 30S ribosomal protein rpsP, along with ribosomal-associated elongation factors such as EF-Tu. Several genes involved in respiration, including those of the FoF1 ATP synthase, also had significantly fewer transcripts as did those transcripts for different virulence-associated protein families [26–31], such as hpmA, umoA, and zapD. Overall, fewer genes had an increased relative abundance as compared to the control strains: 109 genes in CCS06, 56 genes in CCS02, and 332 genes in co-swarmed Δids, respectively. Within these genes, several endogenous toxin-antitoxin systems displayed an increased relative abundances of transcripts, including genes that encode homologous proteins to Escherichia coli YfiA (also known as RaiA) [32] and to toxin/antitoxin pairs ParE/CC2985 [33] and Phd/Doc [34]. There was also an increased relative abundance for several fimbriae families, including genes encoding MR/P and P-like fimbria. A large proportion of differentially regulated genes were proteins of unknown function.
A subset of differentially abundant genes was shared among all three datasets; these were representative of families found in each Ids mismatch strain. Three genes had increased relative abundance in transcripts as compared to wildtype; 32 genes had decreased (Fig 1D). The 32 genes with decreased relative transcripts included those involved in motility, chemotaxis, ribosomal proteins, and metabolism (S5 Table). Of the three genes with increased relative transcripts, two encode the Phd/Doc endogenous toxin-antitoxin system. The third gene is rob, which has been associated with the induction of low-metabolism states in E. coli [35]. Thus, cells under the influence of incoming IdsD, and without a cognate IdsE protein present, enter a distinctive transcriptional state from either wild-type or Δids cells participating in a normal swarm cycle. Transcriptional shifts of these types are often associated with entry into altered states induced by a variety of environmental and temporal cues, including nutrient and membrane stress [36].
We hypothesized that the changes in Ids-excluded cells might also result in the secondary effect of increased antibiotic tolerance. We used susceptibility to antibiotics as a proxy for whether cells have entered an altered state. We conducted antibiotic tolerance assays on wildtype, the Δids strain, and a third strain containing an unmarked in-frame (nonpolar) deletion of the chromosome-encoded idsE within the wildtype background [21]. The Δids strain encodes a transgenic resistance gene to chloramphenicol; the deletion is stable in growth media without chloramphenicol. Neither the wildtype nor the ΔidsE strain encode transgenic genes for antibiotics resistance. All three strains are endogenously resistant to tetracycline.
We predicted that swarms of the ΔidsE strain would display increased antibiotic tolerance compared to wildtype and the Δids strain. We first tested the beta-lactam antibiotic ampicillin to which the wild-type BB2000 is susceptible. We used independent, clonal swarms of either wildtype, the Δids strain, or the ΔidsE strain, each grown in the absence of antibiotics. We harvested cells after the entry to the third swarm ring. Cells were resuspended in LB media and immediately subjected to 100 μg ml-1 ampicillin exposure. Samples were extracted for viability assays on fresh media lacking antibiotics at regular intervals until eight hours and again at 16 hours. No clear difference was observed between wildtype and the Δids strain at any timepoint (Fig 2A). Wildtype and the Δids strain exhibited a killing of approximately 105-fold after eight hours, while the ΔidsE strain experienced a killing of approximately 104-fold (Fig 2A). Under these conditions, the ΔidsE strains had an approximately 50-fold increase in survival as compared to wildtype, even after sixteen hours incubation in ampicillin (Fig 2A).
We repeated this assay with three additional antibiotics to explore whether the ΔidsE cells exhibited a broad tolerance to antibiotics. We utilized the fluoroquinolone ciprofloxacin and the aminoglycosides streptomycin and kanamycin. No difference between wildtype, the Δids strain, and the ΔidsE strain was observed in viable cell counts following 50 μg ml-1 streptomycin or 1 μg ml-1 ciprofloxacin exposure after 12 hours (Fig 2B). These antibiotics resulted in lower rates of cell killing as compared to ampicillin, likely reflecting a higher native resistance of P. mirabilis to these drugs [37, 38]. However, the ΔidsE strain showed an increased number of viable cells as compared to wildtype or the Δids strain after 60 μg ml-1 kanamycin incubation for 12 hours (Fig 2B). Therefore, the ΔidsE strain displays increased tolerance to both kanamycin and ampicillin under these conditions, indicating limited resistance to antibiotics.
Since Ids functions through cell-cell contact-dependent secretion of the identity marker IdsD [15, 17], we next tested whether IdsD secretion was required for antibiotic tolerance to emerge. Cells secrete IdsD into growth media [15]. We performed this assay using MJT01, which is an ΔidsE-derived strain containing a non-functional T6SS [17, 39] and so does not secrete IdsD. The deletion of idsE and the disruption of the T6SS-encoding gene tssB/vipA are unmarked and nonpolar on the chromosome. We observed no clear difference between MJT01 and wildtype over three biological repeats (S1 Fig). To examine whether social exchange was causative for antibiotics tolerance, we tested cells of the ΔidsE strain that were grown to stationary phase with shaking in liquid. IdsD transfer between cells is limited, if at all, during liquid growth. We observed no difference in antibiotics tolerance for the ΔidsE strains as compared to wildtype over three biological repeats (S2 Fig). Therefore, antibiotics tolerance was induced by an Ids mismatch and caused by the transfer of IdsD between neighboring cells without a cognate IdsE present.
We considered whether this Ids-mediated antibiotic tolerance might be due to entry into an irreversible state. To examine the dynamics of how cells exit from an Ids mismatch, we assayed co-swarms in which the GFP-producing Δids strain (Δids-GFP) was inoculated with an equal amount of a wild-type strain constitutively producing DsRed (wildtype-DsRed). We let the swarm progress to the third swarm ring and then harvested the swarms. Cells of each strain were immediately sorted with FACS. Analysis of sorted cells showed that Δids-GFP formed ~10% of the sample from the third swarm ring. Equal numbers of particles of each strain were inoculated in LB media without antibiotics, and growth at 37°C was measured for 24 hours through optical density at 600 nm. No differences between co-swarmed wildtype-DsRed and co-swarmed Δids-GFP were observed at any time-point (Fig 2C). We concluded that the effects induced by an Ids mismatch are transient outside of continual contact-mediated pressure. Consistent with this assertion, we found no differences between the growth of CCS02 (the chimera Ids mismatch strain) and strain CCS01 grown in liquid LB media at 37°C (S3 Fig). IdsD transfer does not occur, or is limited, during liquid growth. Therefore, an individual cell within a swarm is shifted into a distinct transcriptional state when it has received non-self IdsD from the surrounding cells. We found that this cell state caused by an Ids mismatch is temporary and reversible.
Entry into an antibiotic-tolerant state has been linked in other bacteria to the stringent response [36, 40], mediated by the alarmone messenger molecule (p)ppGpp. Although the stringent response has not been studied in P. mirabilis, the genome for wild-type BB2000 contains two canonical genes for production and degradation of ppGpp, relA and spoT [41]. We tested whether Ids mismatch was connected to the stringent response. Using quantitative high performance liquid chromatography (HPLC) [42], we directly measured total ppGpp quantities in cells independently harvested from swarms of clonal wildtype, clonal Δids, or clonal ΔidsE. Nucleotide samples were purified, separated by HPLC, and quantified by measuring UV absorbance spectra using established methods [42]. We performed three biological repeats of ppGpp measurements and found that the samples from the ΔidsE strain contained nearly twice the ppGpp levels as wildtype and the Δids strain (Fig 3A, S4 Fig). These results indicate that ppGpp levels and Ids mismatch are linked.
We next examined whether ppGpp is necessary for Ids mismatch, specifically focusing on the ΔidsE swarm deficiency. Deletions of relA or spoT can prevent ppGpp accumulation and as such, prevent cells from activating the stringent response in several bacteria as discussed in [43]. We constructed three ΔidsE-derived strains, each with an independent, unmarked, nonpolar chromosomal deletion of relA, spoT, or both. We quantified ppGpp levels in each strain using HPLC as described above and found that ppGpp levels in each strain were minimal as compared to wildtype (S5A Fig). We next assayed each strain for swarm colony expansion as compared to that of wildtype and the parent ΔidsE strain. We found that each newly constructed strain formed swarms of a diameter equivalent to wildtype and nearly twice that of the parent ΔidsE strain (Fig 3C). We also found that in 1:1 co-swarms with the wild-type strain, the ppGpp-deficient strains were excluded from the swarm colony edges, similar to the ΔidsE strain (S5B Fig). We interpret these results as interactions with the wild-type strain are not equivalent to interactions with clonal cells lacking ppGpp. Therefore, while ppGpp is required for Ids mismatch to have an effect in clonal swarms, it is likely not the only factor. Moreover, we posit that ppGpp might be a causative factor upstream of the observed transcriptional and physiological changes.
Having observed that a response to Ids mismatch is only present under consistent pressure from neighboring non-self cells (Fig 2), we reasoned then that the effects of Ids mismatch control might be spatially and/or temporally attuned. The small molecule ppGpp might allow for such a rapid and transient response in cells. To interrogate this model, we took advantage of the genes newly identified as being induced in the presence of non-self (Fig 1) to develop a fluorescent transcriptional reporter system. A gene encoding a variant of the fluorescent protein Venus [44] was engineered to be inserted immediately downstream of the gene BB2000_0531, resulting in Venus production being controlled by the upstream promoter. The BB2000_0531 gene, encoding a putative sigma-54 modulation protein, was chosen as it displayed increased expression under different Ids mismatch conditions (Fig 1B, S2 Table, S3 Table, S4 Table). The reporter construct was inserted unmarked and nonpolar in the chromosome of the Δids strain, resulting in strain MJT02; this strain had no apparent growth or swarm defects.
We performed fluorescence microscopy time-course experiments on mixed swarms to measure transcriptional changes associated with BB2000_0531 over the course of a swarm-consolidation cycle. Two co-swarm conditions were used. In the first, a mixed culture of 50% MJT02 and 50% the Δids strain was used to inoculate swarm-permissive agar; in the second, a mixed culture of 50% MJT02 and 50% wildtype-DsRed was used. Swarms were grown at 37°C until the first swarm expansion was visible. Venus fluorescence intensity was measured at 30-minute intervals thereafter in swarm areas, and the mean fluorescence was calculated. The fluorescence intensity for both co-swarm conditions was graphed (Fig 4A); representative images are in Fig 4B. We observed a temporal spike in fluorescence associated with BB2000_0531 correlated with the consolidation cycle; this increase was only apparent when Δids-derived cells were intermingled with wild-type cells (Fig 4). Therefore, the gene expression response to Ids mismatch occurs during consolidation.
In addition to causing territorial exclusion in mixed swarms, Ids mismatch determines boundary formation after collision between two clonal swarms [20]. Boundary formation following swarm contact is a complex process likely involving the contribution of several lethal and non-lethal systems [14, 15, 20, 45]. To test whether equivalent transcriptional shifts were observed during the initial stages of boundary formation, when cells of each strain are first in contact with one another, we measured fluorescence intensity associated with BB2000_0531 in MJT02 swarms following collision with wildtype and Δids swarms. We observed a mean increase in fluorescence intensity over several hours after MJT02 encountered wildtype, but not in encounters with the Δids strain (S6 Fig). The observed increase in fluorescence intensity occurred before a boundary was visually apparent. In fact, formation of a visible boundary between the two strains did not occur for a further 12–18 hours after the end of this experiment, which is consistent with previous observations [18]. Therefore, the Ids mismatch induces a response in the initial stages of self versus non-self recognition, after non-self cells have interacted. We reasoned that Ids mismatch control might be relevant for the formation and/or development of a swarm colony, which is consistent with our prior hypothesis that Ids impacts cooperative behavior [17].
Swarming is fundamentally a collective behavior. The spatial expansion of a wild-type swarm is connected to the oscillatory developmental cycle of outward migration and non-motile consolidation. To assess the hypothesis that the Ids system likely impacts local cell-cell interactions at the boundary and within an expanding swarm, we examined territorial exclusion in situ using epifluorescence microscopy and utilized co-swarms constructed with equal ratios of the Δids and wild-type strains. To allow visualisation of individual cells, 10% of Δids and wildtype cells in the starting mixture constitutively expressed GFP and DsRed fluorescent proteins, specifically strains Δids-GFP and wildtype-DsRed, respectively (Fig 5). The control experiment consisted of a co-swarm in which a starting inoculum consisted of wildtype doped with 10% wildtype constitutively producing GFPmut2 (strain wildtype-GFP) and 10% wildtype-DsRed (Fig 5B). Once swarmer cells emerged from the inoculum, the proportion of cells expressing each fluorophore was measured at half-hour intervals (Fig 5). The developmental stages of active outward motility versus no outward motility (i.e., consolidation) were noted by eye (S7 Fig).
We calculated the fluorophore ratios over time for both the Δids-GFP:wildtype-DsRed and wildtype-GFP:wildtype-DsRed co-swarms for three biological repeats. The GFP/DsRed ratios in the wildtype-GFP:wildtype-DsRed control experiment did not deviate over time, with approximately equal numbers of each strain observable in the swarm over the course of eight hours (Fig 5B). The Δids-GFP:wildtype-DsRed co-swarm did not show measurable changes until well after swarm emergence (Fig 5A), indicating that Δids-GFP cells arising from the inoculum were not excluded from swarm behavior. However, large decreases in the GFP/DsRed ratio were observed in consolidation periods between rounds of swarming, starting in the first consolidation phase (Fig 5A). Later swarm rings often contained no observable Δids-GFP cells.
To test whether the ratio changes observed over multiple swarm cycles were caused by lysis of the cells experiencing Ids mismatch, we monitored a single field of view of Δids-GFP:wildtype-DsRed or Δids-GFP:Δids co-swarms over one consolidation phase. Images were taken at 5-minute intervals from the start to end of the consolidation phase, with three biological repeats performed (S8 Fig). Cell lysis rates were less than 0.1% for both conditions, with no difference in cell lysis rates apparent at any point. We did not observe an absolute decrease in Δids-GFP cell numbers (S8A Fig). We also found that as the consolidation zone increased, the ratio of fluorescence associated with Δids-GFP decreased in co-swarms with wildtype and remained constant in co-swarms with Δids (S8B Fig). Observation of individual cells in consolidation areas suggests that the Δids-GFP strain has apparent differences in cell division dynamics based on whether in a co-swarm with wildtype (non-self) or with the Δids strain (no signal). Altogether, we found that Ids-mediated exclusion was increasingly effective over the course of the co-swarm with initial equal ratios of cells, eventually resulting in Δids-GFP cells being excluded from the leading edges of swarming colonies (Fig 5). Therefore, Ids mismatch results in cells unable to proceed through the swarm developmental cycle during the consolidation period.
As Ids-mediated exclusion was correlated with consolidation in P. mirabilis swarms, we generated “hyperswarming” wild-type and Δids strains (named “wildtype-HS” and “Δids-HS,” respectively) that continually swarm outwards without consolidation [28], which leads to rapid surface coverage through swarm colony expansion. We performed co-swarms to test whether hyperswarming protected the Δids-derived strains from exclusion by wildtype. We observed that neither Δids-HS:wildtype-HS nor Δids-HS:wildtype co-swarms resulted in the exclusion of the hyperswarming Δids strain (S9 Fig). However, Δids:wildtype-HS co-swarms, in which the Δids strain enters consolidation, did result in territorial exclusion of Δids-derived cells (S9 Fig). We concluded that outside of the consolidation phase, the Δids-derived cells that received non-self signals were not effectively excluded, indicating that Ids mismatch does not affect swarm performance in hyperswarming cells. We propose that Ids mismatch induces the recipient cell to experience a growth arrest in the swarmer cell developmental cycle, which then prevents cells from re-entry into swarm-compatible states.
Here we expand on models of kin discrimination [5, 8] by showing that the Ids system encompasses a complex and subtle recognition that is attuned to the challenges of rapid migration as a collective along a hard surface. Ids-mediated recognition controls the spatial location of non-self cells over the lifetime of a swarm. It appears that for this robustly swarming bacterium, access to a social behavior is impeded via a non-lethal mechanism: the Ids self-recognition system selectively induces non-self cells into a growth-arrested lifestyle incompatible with cooperative swarming. Intriguingly, Ids-like proteins are encoded within the genomes of other members of the Morganellaceae family, suggesting that this mechanism might be more broadly found. Further, these data suggest a model for Ids territorial exclusion in mixed swarms (Fig 6). IdsD is likely primarily transferred during active swarming when the secretion machinery is produced and abundantly visible [21, 25, 39]. During consolidation phase, the presence of IdsD with an absence of a cognate IdsE (resulting in unbound IdsD in recipient cells) causes a shift into a distinctive transcriptional state (Fig 1) that is partially due to activation of the stringent response via elevated ppGpp levels (Fig 3). This shifted state also causes a phenotype in affected cells that allows increased antibiotic tolerance (Fig 2). We propose that Ids mismatch functions by diverting cells, via growth arrest, from re-entry into P. mirabilis’ swarm-consolidation developmental cycle, which results in individual non-self cells being iteratively winnowed out of the migrating swarm front when initially present in equal ratios (Fig 6).
Several potential models could explain exactly how unbound IdsD affects the recipient cell. One model is that the presence of unbound IdsD in a recipient cell interrupts a checkpoint in the differentiation from a swarmer to a consolidated cell or vice-versa. While the transcriptomic data provides a reasonable starting point, the list of differences for each Ids mismatch condition as compared to wildtype is quite large. These changes do not resemble those previously described during swarm-consolidation transitions [25], suggesting that Ids mismatch induces entry into a novel expression state. It is also possible that IdsD might accumulate in the membrane over time, leading to a general stress response. However, several pieces of data contradict such a model. First, non-self cells are able to escape Ids-mediated territorial exclusion under laboratory conditions by overexpression of the master flagellar regulator flhDC, which abolishes consolidation to form hyperswarmer cells [31]. Hyperswarming Δids cells receiving a non-self signal are motile and able to swarm with wildtype (S9 Fig). Excluded Δids cells are still able to grow and divide in situ [17]. Further, deletion of relA and/or spoT, which reduces ppGpp levels, allows for increased swarm colony expansion of the clonal Ids mismatch strain, ΔidsE, thereby bypassing Ids mismatch control (Fig 3 and S5A Fig). The molecular mechanisms for how ppGpp levels might cause attenuated expansion remain to be uncovered. Although recognition signals need flagellar regulation and internal ppGpp levels to be effective, how these pathways intersect remains to be uncovered. Interpretation of our research is further complicated, because little is published about the stringent response and/or ppGpp activity in P. mirabilis.
There are several transcriptional changes in the Ids mismatch-induced state that are not readily explained by the ppGpp response. We anticipate that ongoing studies into the emergence of bacterial dormancy and related phenotypes [46–49] in other species might help to untangle the order and hierarchy of the Ids-induced changes described here. Several candidate pathways for further analysis are apparent from the transcriptomics datasets. The role of the signaling molecule c-di-GMP in regulating motile/sessile lifestyle changes in many bacteria is well-studied and an attractive target for future work [50]. The SOS response, mediated by RecA, has been implicated in persister formation in E. coli K12 [51] and may also play a role here. Ids gene regulation in general has been linked with the MrpJ transcriptional network important for P. mirabilis virulence [27]. The observation of increased MR/P fimbrial expression (Fig 1) suggests a potential link between Ids-induced changes, MrpJ, and changes in virulence.
More generally, we have presented evidence of a peer pressure system for recognition that iteratively winnows non-self cells from participating in the collective behavior of swarming—a social activity between cells that is observed among many bacterial species. Ids-mediated macroscale territorial behavior emerges from the sum of cell-cell contacts within a swarm [15, 17, 20, 52]. In this Ids model, cells do not receive any information about population composition and behavior other than that from their immediate neighbors, which is different when compared to other examples of bacterial collective behavior. For example, in bacterial quorum sensing, secretion of diffusible small molecules into the environment provides a global tracker accessible to all individuals in a group [53, 54]. Therefore, each individual cell has potentially equal access to the external signal, because the signal molecule can freely diffuse between/among cells. Each P. mirabilis cell, however, has access only to the signal of a physically adjacent cell. As such, in Ids mismatch-mediated exclusion, any information about the swarming population as a whole is decentralized and distributed among every member of the swarm. Access to that information is restricted to clusters of adjacent, neighboring cells. In these respects, Ids-mediated control represents an orthogonal model for collective behavior in bacteria that provides new opportunities to explore cell-cell communication, especially as regards to spatial coordination. Any theoretical model of Ids-mediated behavior will likely need to differ from those describing quorum sensing of diffusible molecules.
The Ids self-recognition system has distinct qualities from other contact-associated systems that have been described as bacterial communication. CDI systems [55], for example, have been described as lethal [56] or inducing permanent persister-like states [57]. However, broad spatial analysis has yet to be more generally pursued. P. mirabilis swarms could provide an excellent framework for directly analyzing individual cells before, during, and after Ids communication and for examining the global spatial consequences to these local interactions. Tracking individual cell fates through swarming and consolidation will help in this regard. Moreover, the ability of the Ids system to temporally and spatially control non-self cells by altering cell state raises the question of which other mechanisms for contact-mediated signaling in bacteria enable sophisticated interactions between individuals.
Finally, we find it unlikely that the Ids system is a specific adaptation to mitigate antibiotic pressure. This places it in contrast with Ghosh et al [57] who recently modelled CDI-mediated persistence in E. coli as a bet-hedging mechanism. We speculate instead that the antibiotic tolerance observed in this study represents an evolutionary "spandrel" of the type described by Gould and Lewontin [58, 59]. Ids-induced antibiotic tolerance in this case would be a by-product of its primary adaptive feature, regulating clonal swarm composition. Under this view, the antibiotic tolerance data shown in this paper should be regarded as a method of measuring Ids-induced phenotypic shifts in the context of territorial exclusion.
All strains used in this study are described in Table 1. P. mirabilis strains were maintained on LSW- agar [60]. CM55 blood agar base agar (Oxoid, Basingstoke UK) was used as a swarm-permissive agar. E. coli strains were maintained on Lennox lysogeny broth (LB) agar. All liquid cultures were grown in LB broth at 37°C with shaking. Swarm plates were grown either at room temperature or at 37°C. Antibiotics were added when appropriate at the following concentrations: kanamycin 35 μg ml-1, chloramphenicol 50 μg ml-1, carbenicillin 100 μg ml-1, ampicillin 100 μg ml-1, and tetracycline 15 μg ml-1.
All chromosomal mutations in BB2000 and the Δids strain were made as described in [17] with the following modifications for strains constructed de novo in this study: the suicide vector was pRE118 [61] and the conjugative E. coli strain was strain MFDpir [62]. The BB2000_0531 transcriptional reporter strain MJT02 includes a gene encoding the Venus fluorescent protein [44] immediately following the stop codon of gene BB2000_0531. All chromosomal mutations were confirmed by PCR amplification followed by Sanger sequencing of the amplified product (Genewiz, South Plainfield NJ) or by whole genome sequencing as described in [17].
Strains were grown overnight at 37°C in LB broth with appropriate antibiotics. Overnight cultures were diluted in LB broth to an optical density at 600 nm (OD600) of 1.0, then mixed to the desired experimental ratio and inoculated with an inoculation needle onto a CM55 swarm agar plate. Plates were incubated at 37°C for 18 hours, ensuring that the swarm had covered most of the agar plate. After incubation, swarm composition was measured by using a 48-pin multi-blot replicator to sample the swarm and replica plate on non-swarming LSW- agar with relevant antibiotics as described in [15].
Strains were grown on swarm-permissive agar plates with appropriate antibiotics at 37°C. For consolidating cell samples of wildtype, swarm colonies were left to progress overnight and confirmed to be in consolidation phase by light microscopy. Cells from the swarm edge were then harvested by scraping with a plastic loop into 1 ml of RNA Protect solution (Qiagen, Hilden, Germany). The ΔidsE and CCS02 samples were harvested after overnight incubation by scraping whole colonies into 1 ml RNA Protect solution. Total RNA was isolated using a RNeasy Mini kit (Qiagen, Hilden, Germany) according to the manufacturer’s instructions. RNA purity was measured using an Agilent 2200 Tapestation (Agilent, Santa Clara, CA). To enrich mRNA, rRNA was digested using terminator 5’ phosphate dependent exonuclease (Illumina, San Diego, CA) according to the manufacturer’s instructions. Enriched RNA samples were purified by phenol-chloroform extraction [63].
The cDNA libraries were prepared from mRNA-enriched RNA samples using an NEBNext Ultra RNA library prep kit (New England Biolabs, Ipswich, MA) according to the manufacturer’s instructions. Libraries were sequenced on an Illumina HiSeq 2500 instrument with 250 bp single-end reads, and base-calling was done with Illumina CASAVA 1.8 in the Harvard University Bauer Core Facility. Sequences were matched to BB2000 reference genome PMID: 24009111 (accession number CP004022) using TopHat2 using default arguments [64]. Differential expression data were generated using the Cufflinks RNA-Seq analysis suite [65] run on the Harvard Odyssey cluster. Specifically, the mRNA abundance data were generated using Cufflinks 2.1.1 with max-multiread-fraction 0.9 and -multi-read-correct. Samples were combined using cuffmerge with default arguments. Differential expression data were generated using Cuffdiff 2.1.1 with total-hits-norm. The data was analyzed using the CummeRbund package for R and Microsoft Excel. Gene functions were taken from the KEGG and COG databases [66, 67]. The data shown in this paper represent the combined analysis of two independent biological and are available at NCBI GEO accession number GSE131647.
Samples of excluded Δids cells were obtained through fluorescent-activated cell sorting (FACS). Fluorescent strains of wildtype-DsRed and Δids-GFP were grown in liquid at 37°C overnight and normalized to OD600 1.0. Cultures were then mixed to the desired experimental ratio and spotted on swarm agar. After the emergence of the third swarm ring, swarm colonies were harvested into 1X phosphate-buffered saline (PBS) and sorted using a BD FACSAria cell sorter (BD Biosciences, San Jose, CA) into RNA-Protect solution. cDNA samples for RNA-Seq were prepared from sorted samples as described above.
For experiments where strain ratio during swarming was examined, liquid cultures were grown overnight with shaking at 37°C. Cultures were normalized to OD600 1.0, mixed to the desired experimental ratio, then used to inoculate swarm-permissive agar plates, which were incubated at room temperature overnight at room temperature to allow inoculum development. Plates were then imaged at 30-minute intervals, incubating at 37°C between measurements. For experiments with the BB2000_0531 transcriptional reporter, 1 μl of mixed, normalized overnight culture were used to inoculate a 1-mm swarm agar pad, which was incubated at 37°C for four hours prior to imaging. Images were taken in GFP (150 ms exposure), RFP (500 ms exposure), and phase contrast channels using a Leica DM5500B microscope (Leica Microsystems, Buffalo Grove IL) and CoolSnap HQ CCD camera (Photometrics, Tucson AZ) cooled to -20°C. MetaMorph version 7.8.0.0 (Molecular Devices, Sunnyvale CA) was used for image acquisition, and FIJI [68] was used for image analysis. Raw images are available on Open Science Framework.
Strains were grown on swarm-permissive agar plates at 37°C until swarms reached the second round of consolidation (approximately six hours). Swarm colonies were harvested into LB broth and diluted in LB broth to OD600 1.0. Prior to antibiotic exposure, a sample was taken, serially diluted in LB broth, and plated on LSW- agar to count colony-forming units (CFUs/ml) in the sample. Antibiotics were added to the normalized culture at the following concentrations: ampicillin 100 μg ml-1, kanamycin 60 μg ml-1, streptomycin 50 μg ml-1, and ciprofloxacin 1 μg ml-1. Each mixture was incubated with shaking at 37°C. At the specified time-points, samples were taken, serially diluted in LB broth and plated on LSW- agar to measure CFUs/ml. LSW- agar plates were incubated for 16 hours or until visible colonies appeared. Colonies were counted using FIJI. Experiments were performed in triplicate.
Co-swarms of Δids-GFP and wildtype-DsRed were inoculated onto CM55 plates and allowed to swarm at 37°C. Samples were harvested from swarm plates and sorted via FACS as described above, except cells were sorted into PBS solution. Immediately after sorting, portions of sorted cell suspension for each strain, containing equal numbers of sorted particles, were used as inoculum for overnight cultures grown at 37°C with shaking in a Tecan Infinite 200 Pro microplate reader (Tecan, Männedorf, Switzerland). OD600 measurements were taken hourly. Experiments were performed in triplicate. For experiments with strains CCS01 and CCS02, cell sorting was unnecessary. Instead, clonal swarms were directly harvested into PBS. The resulting suspension was diluted to OD600 0.1 and used as inoculum for cultures containing relevant antibiotics.
A high-performance liquid chromatography (HPLC)-based method was used to quantify ppGpp levels, based on the work of Varik et al [42]. Swarm colonies of P. mirabilis were grown to the second swarm ring on CM55 agar. Samples for chromatography were obtained by harvesting cells in 1 ml 1M acetic acid and immediately flash-freezing in liquid nitrogen. Samples were thawed on ice for 1 hour 30 minutes with occasional vortexing, freeze-dried overnight and resuspended in 200 μl MQ-H2O, and then centrifuged at 4°C for 30 min to remove any insoluble fragments. Supernatants were run on a Spherisorb strong ion exchange chromatography column (80 Å, 4.6 by 150 mm, 5 μm, Waters, Milford MA). An isocratic program was used with flow rate 1.5 ml/min in running buffer consisting of 0.36 M ammonium dihydrogen phosphate, 2.5% acetonitrile (v/v), pH 3.6. Nucleotide concentrations were quantified by measuring UV absorbance at 252 nm, comparing peaks to those obtained from purified nucleotide and ppGpp samples (Trilink Biotechnologies, San Diego, CA).
|
10.1371/journal.ppat.1003437 | Interplay between Siderophores and Colibactin Genotoxin Biosynthetic Pathways in Escherichia coli | In Escherichia coli, the biosynthetic pathways of several small iron-scavenging molecules known as siderophores (enterobactin, salmochelins and yersiniabactin) and of a genotoxin (colibactin) are known to require a 4′-phosphopantetheinyl transferase (PPTase). Only two PPTases have been clearly identified: EntD and ClbA. The gene coding for EntD is part of the core genome of E. coli, whereas ClbA is encoded on the pks pathogenicity island which codes for colibactin. Interestingly, the pks island is physically associated with the high pathogenicity island (HPI) in a subset of highly virulent E. coli strains. The HPI carries the gene cluster required for yersiniabactin synthesis except for a gene coding its cognate PPTase. Here we investigated a potential interplay between the synthesis pathways leading to the production of siderophores and colibactin, through a functional interchangeability between EntD and ClbA. We demonstrated that ClbA could contribute to siderophores synthesis. Inactivation of both entD and clbA abolished the virulence of extra-intestinal pathogenic E. coli (ExPEC) in a mouse sepsis model, and the presence of either functional EntD or ClbA was required for the survival of ExPEC in vivo. This is the first report demonstrating a connection between multiple phosphopantetheinyl-requiring pathways leading to the biosynthesis of functionally distinct secondary metabolites in a given microorganism. Therefore, we hypothesize that the strict association of the pks island with HPI has been selected in highly virulent E. coli because ClbA is a promiscuous PPTase that can contribute to the synthesis of both the genotoxin and siderophores. The data highlight the complex regulatory interaction of various virulence features with different functions. The identification of key points of these networks is not only essential to the understanding of ExPEC virulence but also an attractive and promising target for the development of anti-virulence therapy strategies.
| The synthesis of numerous molecules involved in the virulence potential and fitness of pathogenic bacteria requires a particular enzyme family, i.e. phosphopantetheinyl transferases (PPTases). To date, the synthesis of a given bioactive metabolite was thought to require a specific PPTase. As PPTases are being investigated as promising targets for antibacterial development, we addressed the question of a possible functional interchangeability between PPTases in Escherichia coli. PPTases are known to be involved in the synthesis of low-molecular weight iron chelators (siderophores), and of a genotoxin named colibactin. Here we demonstrated interplay between the synthesis pathways leading to the production of siderophores and of colibactin. We showed that inactivation of both PPTases abolished the virulence of extra-intestinal pathogenic E. coli (ExPEC) in a mouse sepsis model. To our knowledge, this is the first demonstration of interplay between multiple PPTases-requiring pathways leading to the biosynthesis of functionally distinctive virulence factors, in a given microorganism. The extensive substrate specificity of PPTase ClbA could account for the co-selection and co-evolution of genomic islands encoding colibactin and yersiniabactin siderophore.
| Escherichia coli is a normal resident of the lower-gut of humans and animals. Although usually a commensal, E. coli can be also a pathogen, associated with diarrheal disease and extra-intestinal infections [1], [2]. The majority of E. coli strains can be assigned to one of five main phylogenetic groups: A, B1, B2, D and E [3]. Strains of the distinct phylogenetic groups differ in their phenotypic and genotypic characteristics [4]–[6]. Extra-intestinal pathogenic E. coli (ExPEC), which display enhanced ability to cause infection outside the intestinal tract, carry specific genetic determinants or virulence factors that are clustered on different pathogenicity islands [7]. These virulence factors associated with extra-intestinal infections are nonrandomly distributed, and strains of the E. coli phylogenetic group B2 harbor the greatest frequency and diversity of virulence traits [8], [9].
As iron bioavailability is limited in the host, ExPEC are known to synthesize up to four types of siderophores involved in iron uptake: enterobactin, salmochelins, yersiniabactin and aerobactin [10], [11]. The biosynthesis of the first three requires a 4′-phosphopantetheinyl transferase (PPTase). These enzymes activate polyketide synthases (PKSs) and nonribosomal peptide synthetases (NRPSs) by catalyzing the transfer of a phosphopantetheinyl (P-pant) moiety from coenzyme A to conserved serine residues on PKSs and NRPSs [12], [13]. In organisms containing multiple P-pant-requiring pathways, each pathway generally involves a dedicated cognate PPTase [12]. In E. coli, the EntD PPTase is involved in the synthesis of enterobactin [14] and salmochelins, which are glycosylated forms of enterobactin [15]. The IroA locus responsible for salmochelins production is located either on a chromosomal pathogenicity island or on a transmissible plasmid [16]. Contrary to enterobactin, salmochelins are able to evade the mammalian innate immune response protein lipocalin 2 (siderocalin) and are therefore more potent virulence factors [17]. The other siderophore necessitating a PPTase is yersiniabactin. This siderophore is encoded by the high-pathogenicity island (HPI) that was acquired through horizontal transfer [18]. The HPI core region was detected in more than 70% of ExPEC isolated from blood cultures, urine samples and cerebrospinal fluid [19]. While yersiniabactin production in Yersinia requires the YbtD PPTase encoded outside the HPI [20], no gene homologous to ybtD has been identified in the genome of E. coli strains producing yersiniabactin. The PPTase committed to the synthesis of yersiniabactin in E. coli remains unknown.
We have shown that a number of E. coli strains from phylogenetic group B2 display also the pks island, which codes for the production of colibactin, a polyketide-non ribosomal peptide genotoxin [21]. Colibactin is known to induce DNA double-strand breaks, cell cycle arrest in G2-phase and megalocytosis in infected eukaryotic cells [21]. E. coli strains harboring the pks island can induce DNA damage in enterocytes in vivo and trigger genomic instability in mammalian cells [22]. In a rodent model of colon inflammation, colibactin was demonstrated to potentiate the development of colon cancer [23]. Surprisingly, colibactin is also required for the colonic anti-inflammatory properties of the probiotic E. coli strain Nissle 1917 [24]. The synthesis of colibactin requires a PPTase encoded by the clbA gene located on the pks island [21]. Epidemiological studies revealed that the majority (73.1%) of the colibactin-positive E. coli strains was clinical ExPEC and that the pks island was significantly associated with a highly virulent subset of ExPEC isolates [25]. Strikingly, an analysis of the prevalence of the colibactin island among Enterobacteriaceae revealed that the pks island was constantly associated with the yersiniabactin gene cluster [26].
In this work we investigated a potential interplay between the biosynthetic pathways leading to the production of siderophores and of the colibactin genotoxin, through a possible functional interchangeability between PPTases in E. coli. We demonstrated that ClbA can contribute to the synthesis of siderophores both in vitro and in vivo. We proved in a mouse model of sepsis that the presence of either functional EntD or ClbA is required to maintain full virulence of ExPEC. This evidenced the interconnection between pathways leading to the synthesis of distinct secondary metabolites, via the PPTase ClbA. Therefore, the strict association of the pks island with HPI could have been selected in highly virulent E. coli isolates because ClbA can contribute to the synthesis of both the genotoxin and yersiniabactin.
Because colibactin and siderophores belong to the same family of chemical compounds, we investigated first whether the pks island could not only allow the production of a genotoxin, but also of a siderophore. The entE gene, that encodes the ligase component of synthase multienzyme complex necessary for the enterobactin biosynthesis, was inactivated in the enterobactin producer E. coli strain MG1655. The resulting MG1655 entE mutant strain was shown not to produce any siderophore, as detected on CAS plate (Fig. 1A). The wild type (WT) and entE derivative of strain MG1655 were transformed with the bacterial artificial chromosome (BAC) harboring the entire pks island (BAC pks+). Both strains MG1655+BAC pks+ and MG1655 entE+BAC pks+ were shown to produce the genotoxin, as evidenced by the induction of double-strand breaks in eukaryotic cells (data not shown). The production of siderophore was qualitatively investigated in the resulting strains by plating on CAS plates (Fig. 1A). A yellow halo was not observed surrounding the bacterial colonies of strain MG1655 entE+BAC pks+. This showed that the pks island did not code for the biosynthesis of a siderophore.
In order to test whether the ClbA PPTase was functionally capable of participating to the biosynthesis of enterobactin, the entD gene was disrupted in E. coli strain MG1655. The resulting MG1655 entD mutant strain was subsequently transformed with BAC pks+ and with the BAC harboring the entire pks island where the clbA gene was deleted (BAC pksΔclbA). The production of siderophore was investigated by plating the resulting strains on CAS medium (Fig. 1B). This revealed that disruption of the entD gene in strain MG1655 resulted in the abrogation of the production of enterobactin (Fig. 1B). The introduction of the intact pks island in strain MG1655 entD restored the production of yellow pigmentation surrounding the colonies. This was not observed upon the introduction of the pks island disrupted for the clbA gene (Fig. 1B). Introduction of a functional plasmidic clbA gene in strain MG1655 entD+BAC pksΔclbA and in strain MG1655 entD restored the production of enterobactin (data not shown).
These data evidenced that the ClbA PPTase can contribute to the enterobactin siderophore synthesis in vitro.
Yersiniabactin is a siderophore the biosynthesis of which requires the PPTase YbtD in Yersinia pestis [20]. Although numerous E. coli strains were shown to produce yersiniabactin, an in silico analysis of the genome of all the E. coli strains available to date did not reveal any gene homologous to the ybtD gene.
In order to test whether the ClbA PPTase was functionally proficient to participate to the biosynthesis of yersiniabactin, we analyzed the enterobactin and yersiniabactin producer E. coli strain SE15. The entD gene was disrupted in E. coli strain SE15. The resulting SE15 entD mutant was subsequently transformed with plasmids carrying wild type entD gene or clbA gene. The production of total siderophores was qualitatively (Fig. 2A) and quantitatively (Fig. 2B) investigated using the CAS assay. This revealed that disruption of the entD gene resulted in the abrogation of the production of siderophores in strain SE15 (Fig. 2A and 2B). As expected, complementation with entD gene restored the production of siderophores. Remarkably, complementation with clbA gene also resulted in the synthesis of siderophores (Fig. 2A and 2B). The synthesis of yersiniabactin was specifically quantified in the different SE15 derivatives (Fig. 2C). This revealed that in the entD mutant, the yersiniabactin biosynthesis was abolished. The introduction of entD or clbA genes in SE15 entD mutant strain resulted in the restoration of yersiniabactin production.
These data showed that in E. coli strain SE15, EntD is the PPTase dedicated to the synthesis of yersiniabactin. Moreover, the EntD function can be substituted by ClbA. This suggests that both EntD and ClbA are involved in the synthesis of yersiniabactin in E. coli strains producing endogenously EntD and ClbA.
As our data demonstrated that ClbA could complement EntD for the synthesis of enterobactin and yersiniabactin, we investigated whether EntD could rescue a clbA mutant for the production of colibactin. The entD gene was disrupted alone or in combination with the clbA gene in the colibactin producing E. coli strain M1/5. The M1/5 entD clbA double mutant was transformed with multicopy plasmids harboring wild type entD or clbA genes. The production of colibactin was quantified in the resulting strains through the quantification of megalocytic cells (Fig. 3A) and phosphorylation of H2AX histone (Fig. 3B) which correlate with DNA double strand breaks resulting from the genotoxic effect of colibactin [21], [22].
HeLa cells were infected with the different strains for 4 hours, fixed and stained with methylene blue in order to quantify the megalocytosis effect, as previously described [21]. This revealed that the megalocytosis effect observed with the M1/5 entD mutant strain was similar to the effect measured with the wild type M1/5 strain (Fig. 3A). Inactivation of the clbA gene in the M1/5 entD mutant abrogated the colibactin effect (Fig. 3A). Transformation of the M1/5 entD clbA mutant strain with plasmids carrying the functional wild type clbA gene resulted in the restoration of the megalocytosis. A partial complementation of the double mutation was observed with plasmid p-clbA (1) whereas the double mutant was fully complemented with p-clbA (2). The different copy number of the plasmids can account for the quantitative differences observed below. A complementation was not observed when the wild type entD gene was expressed from a multicopy plasmid in the double mutant (Fig. 3A).
Genotoxicity of colibactin [21] was also examined in HeLa cells using H2AX assay based on indirect DNA double strand break detection using In Cell Western (ICW) with infrared fluorescence for H2AX phosphorylation (γ-H2AX) quantification [27]. HeLa cells were infected with strains M1/5, M1/5 entD, M1/5 entD clbA or M1/5 entD clbA complemented with entD. Following the quantification of the γ-H2AX (green) and the DNA (red) signals (Fig. 3B), respectively, the fold induction of γ-H2AX per cell was calculated. This revealed a genotoxic dose–response depending on the multiplicity of infection (MOI, Fig. 3B). No difference of γ-H2AX per cell was observed between WT and entD mutant strains. Infection of HeLa cells with mutant M1/5 entD clbA did not induce phosphorylation of H2AX. Moreover, the introduction of the functional entD gene did not result in the generation of DNA double strand breaks in strain M1/5 entD clbA (Fig. 3B).
Altogether, these data evidenced that EntD does not contribute to the colibactin synthesis, even when highly expressed on a multicopy plasmid.
We then investigated whether other PPTases, originated from other bacterial species, could rescue a clbA mutant for the production of colibactin. The clbA gene was disrupted in E. coli strain M1/5. The M1/5 clbA mutant was transformed with plasmids harboring wild type ybtD gene that encodes the YbtD PPTase in Yersinia pestis, pptT gene the PptT PPTase in Mycobacterium tuberculosis, sfp gene the Sfp PPTase in Bacillus subtilis, and clbA gene. PptT is involved in biosynthesis of the mycobactin siderophore [28] and is essential for mycobacterial viability [29]. Sfp is required for production of the peptide antibiotic surfactin [30]. The production of colibactin was quantified in the resulting strains through the quantification of megalocytic cells (Fig. 4A) and phosphorylation of H2AX histone (Fig. 4B). This revealed that both the megalocytosis and the H2AX phosphorylation were restored in the clbA mutant upon introduction of ybtD, pptT and sfp genes.
These data evidenced that ClbA can be xeno-complemented for the colibactin synthesis.
In order to confirm that EntD and ClbA have narrow and broad substrate-specificity, respectively, we investigated whether EntD and ClbA had the capacity to activate the carrier protein involved in a reporter biosynthetic pathway. When activated by a PPTase, the single-module non-ribosomal peptide synthetase BpsA from Streptomyces lavendulae synthesizes a colored product (indigoidine), from a single substrate (L-glutamine) [31]. Plasmid p-BpsA that encodes BspA was transformed into strain MG1655 entD. The resulting MG1655 ΔentD+p-BspA strain was subsequently transformed with plasmids carrying ybtD, pptT, sfp, clbA, or entD genes. In addition, E. coli strain MG1655 BAC pks+ and MG1655 BAC pksΔclbA were transformed with p-BpsA. The resulting strains that carry both the NRPS and a functional PPTase were grown in auto-induction medium, as previously described [32]. A blue coloration was detectable in cultures after overnight incubation for all strains but strain MG1655 ΔentD+p-BspA+p-entD (Fig. 5A). A quantification of the indigoidine production was determined for all the strains (Fig. 5B). This confirmed that contrary to EntD, the PPTases YbtD, PptT, Sfp and ClbA were able to participate to the synthesis of the blue pigment.
This strengthens the fact that ClbA is more promiscuous in its substrate specificity than EntD in E. coli.
In order to address the consequences, on the virulence of E. coli, of the cross talk between the synthesis pathways of colibactin and siderophores demonstrated in vitro, we investigated E. coli strain SP15, an extra-intestinal pathogenic E. coli strain (ExPEC) of serotype O18:K1:H7 isolated from neonatal meningitis, in a mouse model of sepsis. E. coli strain SP15 produces colibactin and four different siderophores (aerobactin, yersiniabactin, enterobactin and salmochelin). The entD or clbA genes were disrupted individually and in combination. The strains were injected individually into the mice footpad; and the mice survival was monitored (Fig. 6A). This revealed that all the strains but SP15 entD clbA induced 70% mortality within 40 hours after injection. In contrast, virulence of strain SP15 entD clbA was completely attenuated in this mouse model of sepsis (Fig. 6A).
The bacterial dissemination in the mice was analyzed (Fig. 6B). Mice were sacrificed 18 hours post injection with PBS, WT strain, single or double mutants. Spleens and blood samples were collected, and bacteria were quantified by plating on selective medium (Fig. 6B). We observed that in both spleen and blood of infected animals the bacterial loads were similar with all the strains, but strain SP15 entD clbA. No bacteria were recovered from spleen or blood of mice injected with the double mutant SP15 entD clbA (Fig. 6B).
This demonstrated that both EntD and ClbA must be inactivated to abolish virulence of ExPEC in a mouse model of sepsis.
In order to investigate the relative importance of EntD and ClbA in the virulence of E. coli, the SP15 entD clbA mutant strain was transformed with plasmids harboring clbA or entD functional genes. The resulting complemented strains were injected in mice (Fig. 7). This showed that complementation of strain SP15 entD clbA with either clbA or entD totally restored the virulence of the strain (Fig. 7A). A slight but statistically significant delay in survival kinetics was observed when strain SP15 entD clbA complemented with the clbA gene was used for the injections (Fig. 7A). The quantification of bacteria in spleen and blood of the infected animals was determined (Fig. 7B). This revealed that complementation with clbA or entD allowed the survival of strain SP15 entD clbA in vivo, in a statistically significant manner at least in blood (Fig. 7B).
This evidenced that the presence of either functional EntD or ClbA is required to maintain full virulence of ExPEC in a mouse model of sepsis.
Our work demonstrates the interplay between the biosynthetic pathways of a genotoxin and multiple siderophores. We have shown that ClbA, encoded by the pks island, is a promiscuous PPTase which promotes the synthesis of colibactin, yersiniabactin, enterobactin and consequently salmochelins. Although we demonstrated that ClbA could substitute for an entD mutation, the reciprocity was not observed. EntD seems to be specific for the synthesis of siderophores, which is consistent with other published reports [32]. In contrast, YbtD, the PPTase involved in yersiniabactin production in Yersinia was shown to substitute for a clbA mutation and allowed the production of colibactin. Attempts to relate conserved motifs of the group II subfamily of PPTases [12] with substrate specificity did not allow us to understand the functional promiscuity evidenced among certain PPTases, since type II PPTases usually have very remote primary sequences. Unfortunately, it is not possible to compare either the 3D structure of these PPTases because only the structure of Sfp is available [33]. Type II PPTases are predicted to have a similar folding and very similar secondary structures [33]. However it is difficult to draw conclusions on the folding of proteins and to correlate it with substrate specificity. Only the comparison of 3D high-resolution structures would provide information about the structure/function relationship of PPTases. Our work provides novel evidence that make PPTases promising targets for antibacterial development [34], because these enzymes are crucial for the biosynthesis of a multitude of a pathogen's collection of essential metabolites and virulence factors [35].
Iron is an essential element for survival of E. coli. Therefore, E. coli strains have evolved a strategy for iron acquisition which uses multiple siderophores with high-affinity for ferric iron. These include enterobactin, salmochelins, aerobactin and yersiniabactin [11]. Each siderophore has specific affinity for iron and may be differentially regulated to provide different advantages, potentially allowing extra-intestinal pathogenic E. coli (ExPEC) to adapt to different environmental conditions or to overcome host innate immunity [10], [36], [37]. In our model of sepsis, the ExPEC mutant that produced only aerobactin as a siderophore (strain SP15 entD clbA) was completely attenuated. This suggests that aerobactin plays a minor role in the iron uptake in this sepsis model; but the importance of each siderophore can be host and strain dependent [38]. Interestingly, either ClbA or EntD were able to restore the virulence of strain SP15 entD clbA. However, we have shown that colibactin synthesis cannot be sustained by EntD. This suggests that not colibactin, but the siderophore systems (alone or in combination) are critical during the first step of the infection in this mouse model of sepsis. Indeed, the bacterial loads in both spleen and blood were similar in animal infected with SP15 entD clbA mutant complemented either with ClbA or EntD. Analysis of bacteria present in the popliteal lymph node confirmed this analysis (data not shown). Since the carriage of the pks island is correlated with successful long-term gut colonization in humans [39], colibactin could be important for the commensal lifestyle of ExPEC. Moreover, our unpublished data suggest that the genotoxin colibactin could also play a role in natural sepsis since lymphocytes are susceptible to the genotoxin.
Phylogroup B2, which includes the majority of ExPEC isolates, is considered to represent the evolutionary eldest lineage within the species [40]. Interestingly, the pks island found in B2 isolates is highly conserved, and is physically associated to a highly conserved High-Pathogenicity Island. This might even point towards a recent emergence of a distinct subgroup within phylogroup B2. In fact, epidemiological knowledge allows defining specific clonal lineages with high ExPEC virulence potential [41]. We believe that the most virulent and also the best colonizer of human gut resulted from a step-by-step acquisition and selection of different mobile elements. We propose here a scenario with the sequential integration of at least two pathogenicity islands and the cross talk via two PPTases (Fig. 8). At first, all E. coli strains produce at least one siderophore i.e. enterobactin. The entD gene and the other genes of the enterobactin system are part of the core genome and have been identified in all the E. coli strains isolated so far [42]. In contrast, the HPI encoding the yersiniabactin siderophore system devoid of any PPTase gene was acquired by horizontal gene transfer. Almost all E. coli HPIs appear to result from a single ancestor, which entered the E. coli species rather recently [43]. All strains of the phylogenetic group B2 and almost all of group D carry the HPI, whereas strains of groups A and B1 were found to be only occasionally HPI positive (Fig. 8, [19]). The spread of the HPI must have occurred in a dramatically fast fashion, which may indicate a strong selective pressure. We have shown in this study that EntD is actually the PPTase that mediates the synthesis of a functional yersiniabactin. E. coli strains that contain the HPI were demonstrated to be more virulent than isolates that lack the island [18]. Moreover, yersiniabactin is frequently associated with urinary tract infections [44], [45]. The pks island is known to be confined to the phylogenetic group B2. Besides, the pks island is highly represented within an especially highly virulent subset of B2 strains that exhibit extremely elevated virulence scores and an increased likelihood of causing bacteremia [25]. It has been previously demonstrated that all the E. coli strains that acquired the pks island encoding the colibactin through horizontal transfer, also displayed the HPI locus, with an integration site in tRNA asnW gene and asnT gene, respectively (Fig. 8; [26]). The pks island appears to be highly conserved (or even identical) in terms of nucleotide sequence in different E. coli isolates [26]. This may be a hint to a more recent acquisition of the pks island, compared to the HPI, which displays about 1–2% sequence divergence among the E. coli isolates (Schubert, unpublished data). We hypothesize that the association of the pks island with HPI has been selected in the highly virulent E. coli isolates because ClbA can contribute to the synthesis of both the genotoxin and yersiniabactin (and also enterobactin and consequently salmochelins). This deadly association is not confined in E. coli. Similar events also occurred in other pathogenic Enterobacteriacae since the pks island was also detected in Klebsiella pneumoniae, Enterobacter aerogenes, and Citrobacter koseri isolates where the island is also physically associated on the chromosome with the HPI locus [26].
Bacterial strains used in this study are listed in Table 1. E. coli SE15 (O150:H5) is a human commensal bacterium isolated from feces of a healthy adult and classified into E. coli phylogenetic group B2 [46]. Strain SE15 is devoid of the pks island. E. coli M1/5 is a human commensal bacterium isolated from feces of a healthy adult and classified into E. coli phylogenetic group B2. Strain M1/5 harbors of the pks island. Strain SP15 is an extra-intestinal pathogenic E. coli strain (ExPEC) of serotype O18:K1:H7 isolated from neonatal meningitis. Strain SP15 harbors the pks island. The repertoire of siderophores the E. coli strains possess is indicated in Table 1. Gene inactivations were engineered by using the lambda Red recombinase method [47] using primers listed in Table 2. For complementation, the clbA gene was cloned into plasmid pASK75, a cloning vector that harbors a pBR322 origin of replication and therefore is low copy number plasmid (p-clbA (1), table 1) or PCR-Script, a cloning vector that harbors a ColE1 origin of replication and therefore is high copy number plasmid (p-clbA (2), table 1). For complementation, the entD gene was cloned into PCR-Script (p-entD, table 1).
Before injection to mice, all E. coli strains were grown overnight in LB broth supplemented with antibiotics if required, at 37°C with shaking. These cultures were diluted 1∶100 in LB broth with antibiotics when necessary and grown for 3 h at 37°C with shaking. Bacterial cells were resuspended in sterile PBS to the appropriate concentration (2×109 CFU/mL). All the strains were shown to display similar growth kinetics in vitro in LB broth (data not shown).
Chrome azurol S (CAS) assay was used to detect siderophores produced by E. coli. The CAS solution was prepared according to Schwyn and Neilands [48]. E. coli strains were grown on CAS agar plates and incubated at 37°C overnight in the dark. The colonies with orange zones were siderophore-producing strains [48].
To quantify siderophore synthesis, 500 µL of CAS indicator solution containing 4 mM sulfosalicylic acid was mixed with the same volume of supernatant. The reaction mixtures were incubated for 60 min at room temperature to allow complex formation, and the siderophore-dependent color change was determined at OD630 nm. For quantification, the iron chelating agent 8 hydroxyquinoline (8HQ, sigma-aldrich) was used as the standard.
The expression of the fyuA gene encoding the yersiniabactin receptor (FyuA) is known to be up-regulated in the presence of extracellular yersiniabactin [49]. Thus, yersiniabactin-dependent up-regulation of fyuA expression can be monitored by means of a fyuA-reporter fusion in the indicator strain [50].
Bacterial strains were cultivated in NBD medium, i.e. Nutrient Broth (NB) medium supplemented with 200 µM α,α'-dipyridyl (Sigma), for 24 h at 37°C. Bacteria were pelleted by centrifugation and the supernatant was added to the indicator strain WR1542 carrying plasmid pACYC5.3L (kind gift of W. Rabsch, Wernigerode). The plasmid encodes all genes necessary for yersiniabactin uptake; i.e. irp6, irp7, irp8, fyuA and ybtA. Additionally, the fyuA promoter region fused to the luciferase reporter gene is included on pACYC5.3L. After further 24 h of incubation at 37°C the indicator strain was pelleted and resuspended in bacterial lysis buffer (100 mM potassium phosphate buffer [pH 7.8], 2 mM EDTA, 1% [wt/vol] Triton X-100, 5 mg/ml bovine serum albumin, 1 mM dithiothreitol, 5 mg/ml lysozyme). Complete lysis was performed by incubation at room temperature for 20 min and repeated mixing. The samples were centrifuged and supernatants of lysates were analyzed by addition of luciferase reagent (20 mM Tricine-HCl (pH 7.8), 1.07 mM (MgCO3)4 Mg(OH)2, 100 µM EDTA, 470 µM D(-) luciferin, 33.3 mM dithiothreitol, 270 µM Li3 coenzyme A, 530 µM Mg-ATP). Luciferase activities were determined in triplicates using the multimode reader Berthold Tristar LB 941. Values were corrected by relating luciferase activity to the OD600 of bacterial cultures grown 24 h in NBD medium. K12 E. coli strain DH5α served as negative control. The experiments were repeated at least three times.
After overnight cultures in LB broth supplemented with the appropriate antibiotics, bacteria were diluted 1∶10 in M9 minimal medium supplemented with 100 mM L-glutamine and 1 mM IPTG, and cultivated 16 h at 18–20°C under shaking [31], [51]. Bacteria were then collected by centrifugation at 900× g for 5 min. At this speed, bacterial cells were pelleted while indigoidine still remained in the supernatant [51]. Indigoidine production was quantified by measuring the absorbance of blue-colored supernatant (OD612 nm). The bacterial pellet was resuspended in PBS, and biomass was quantified by measuring the absorbance (OD450 nm). Finally, the indigoidine production was normalized with the ratio Indigoidine/Biomass (e.g. ratio OD612 nm/OD450 nm).
HeLa cells were maintained by serial passage in DMEM supplemented with 10% FCS, non-essential amino acids and 50 µg/mL gentamicin. HeLa cells were dispensed in 96-well cell culture plate (5×103 cells/well). For bacterial infections, overnight LB broth cultures of E. coli were diluted in interaction medium (DMEM, 5% FCS, 25 mM HEPES) and cell cultures (∼70% confluent) were infected with a multiplicity of infection (number of bacteria per HeLa cell at the onset of infection) of 3 to 400. Four hours post-inoculation, cells were washed 3 times with HBSS and incubated in cell culture medium 72 h with 200 µg/mL gentamicin before protein staining with methylene blue (1% w/v in Tris-HCl 0.01M). The methylene blue was extracted with HCl 0.1N. The quantification of staining was measured at OD660 nm.
The In Cell Western procedure was performed as described previously [27]. Briefly, HeLa cells were dispensed in 96-well cell culture plate (1.5×105 cells/200 µL/wells). Twenty four hours later, cells were infected with E. coli strains for 4 h. Eight hours post-infection the cells were directly fixed in the plate with 4% paraformaldehyde. Paraformaldehyde was neutralized, and cells were permeabilized as previously described [27]. Cells were blocked with MAXblock Blocking medium (Active Motif, Belgium) supplemented with phosphatase inhibitor PHOSTOP (Roche), followed by overnight incubation with rabbit monoclonal anti γ-H2AX (Cell Signaling) (1∶200). An infrared fluorescent secondary antibody absorbing at 800 nm (IRDyeTM 800CW, Rockland) was then applied (dilution 1∶500). For DNA labeling, RedDot2 (Biotium) was used (dilution 1∶500) together with the secondary antibody. The DNA and the γ-H2AX were simultaneously visualized using an Odyssey Infrared Imaging Scanner (Li-Cor ScienceTec, Les Ulis, France) with the 680 nm fluorophore (red color) and the 800 nm fluorophore (green dye). Relative fluorescence units from the scanning allowed a quantitative analysis. Relative fluorescent units for γ-H2AX per cell (as determined by γ-H2AX divided by DNA content) were divided by vehicle controls to determine percent change in phosphorylation of H2AX levels relative to control. All experiments were carried out in triplicate.
Animal experimentations were carried out in accordance with the European directive for the protection of animals used for scientific purposes. The protocols were validated by the local ethics committee on animal experiment “Comité d'éthique Midi Pyrénées pour l'expérimentation animale” which is affiliated to “Comité National de Réflexion Ethique sur l'Expérimentation Animale” and linked to the french ministry of research (Referenced protocols: PX-ANI-A2-94, 95, 96, 99, 100, and 101). Nine week old female C57BL/6J mice (JANVIER) were injected into the footpad with 108 ExPEC WT, clbA mutants, entD mutant, entD clbA mutant and entD clbA mutant complemented with clbA (entD clbA+p-clbA(2)) and entD (entD clbA+p-entD), together with intraperitoneal injection of 100 µL of carbenicillin (1.6 mg/mL) or PBS. When required, mice were sacrificed by lethal anaesthesia (rompun/ketamine in 0.9% NaCl) 18 h post injection. The abdominal cavity of anesthetized mouse was opened. The widest part of the posterior vena cava was localized and sectioned. Blood was collected by aspiration from the abdominal cavity. Spleens were surgically removed. Bacteria located in spleen cells were isolated from the mechanical dissociation of the splenic tissue using Precellys tissue homogenizer. Bacteria were quantified by plating of serial dilutions of blood and dissociated spleen on appropriate selective MacConkey agar. The antibiotics used to supplement the medium correspond to the resistance displayed by the different strains and are indicated in Table 1.
Statistical analyses were conducted using GraphPad Prism 5.0d. The mean ± standard deviation (SD) is shown in figures, and P values were calculated using a one-way or two-way ANOVA followed by a Bonferroni post-test unless otherwise stated. For bacterial quantification, CFU by mg of spleen or mL of blood were log transformed for the analysis. A P value of less than 0.05 was considered statistically significant and is denoted by *. P<0.01 is denoted by ** and P<0.001 by ***.
|
10.1371/journal.pntd.0000986 | Correlation of Clinical Trachoma and Infection in Aboriginal
Communities | Trachoma is the leading infectious cause of blindness due to conjunctival
infection with Chlamydia trachomatis. The presence of
active trachoma and evidence of infection are poorly correlated and a strong
immunologically-mediated inflammatory response means that clinical signs
last much longer than infection. This population-based study in five
Aboriginal communities endemic for trachoma in northern Australia compared a
fine grading of clinical trachoma with diagnostic positivity and organism
load.
A consensus fine grading of trachoma, based on clinical assessment and
photograding, was compared to PCR, a lipopolysacharide (LPS)-based
point-of-care (POC) and a 16S RNA-based nucleic acid amplification test
(NAAT). Organism load was measured in PCR positive samples.
A total of 1282 residents, or 85.2% of the study population, was
examined. Taking the findings of both eyes, the prevalence of trachomatous
inflammation-follicular (TF) in children aged 1–9 years was
25.1% (96/383) of whom 13 (13.7%) were PCR positive on the
left eye. When clinical data were limited to the left eye as this was tested
for PCR, the prevalence of TF decreased to 21.4% (82/383). The 301 TF
negative children, 13 (4.3%) were PCR positive. The fine grading of
active trachoma strongly correlated with organism load and disease severity
(rs = 0.498,
P = 0.0004). Overall, 53% of
clinical activity (TF1 or TF2) and 59% of PCR
positivity was found in those with disease scores less than the WHO
simplified grade of TF.
Detailed studies of the pathogenesis, distribution and natural history of
trachoma should use finer grading schemes for the more precise
identification of clinical status. In low prevalence areas, the LPS-based
POC test lacks the sensitivity to detect active ocular infection and nucleic
acid amplification tests such as PCR or the 16S-RNA based NAAT performed
better. Trachoma in the Aboriginal communities requires specific control
measures.
| Repeated episodes of C. trachomatis infection lead to active
trachoma clinically characterised by an often intense inflammatory response to
chlamydial antigens with later scarring and distortion of the eyelid leading to
blindness. However, the clinical signs of trachoma do not correlate well with
laboratory tests to detect the presence of Chlamydia. The WHO simplified
clinical grading scheme currently used for assessment of trachoma has a poor
correlation with C. trachomatis genomic test findings, even
though the detection of bacterial genome is strongly correlated with the
prevalence and severity of active trachoma. A detailed assessment of the
clinical signs using a finer grading system was studied in a population-based
survey in five Australian Aboriginal communities. Much clinical activity and
infection was found in those with clinical signs below the threshold used in the
current WHO grading scheme. Future studies of the distribution of infection and
pathogenesis should use finer grading methods than the current WHO scheme. The
prevalence of trachoma in these communities confirms that trachoma remains of
public health importance and sustained interventions to control trachoma are
warranted.
| Trachoma is the leading infectious cause of blindness [1], [2], [3], and results from repeated
episodes of conjunctival infection by Chlamydia trachomatis (CT)
serovars A, B, Ba and C. It is a major public health problem associated with poverty
in environments with inadequate sanitation, poor personal hygiene and poor water
supply and is now largely confined to developing countries, particularly in
Sub-Saharan Africa [2], [4], [5], [6]. Nucleic acid amplification tests (NAATs) require
appropriate facilities and skilled staff, but a assay designed for use in
resource-limited settings may offer some advantages for the diagnosis of infection
over clinical assessment [7], [8]. In general, irrespective of the diagnostic methodology,
there is a relatively poor correlation between clinically active trachoma and
biological evidence of infection, in part because signs of the disease are induced
by a strong immunologically-mediated inflammatory response that resolves much more
slowly than the infection [4], [5], [9], [10], [11], [12], [13]. It is further compounded by the occurrence of repeated
episodes of infection. Also important is the relative lack of precision in assessing
clinical status with the WHO simplified trachoma grading system [14], [15], which was
designed to be learnt and used by local health workers and generally has a high
level of reproducibility [16].
We sought to compare a fine consensus grading of trachoma combining clinical and
photographic grading [15], [17] with a commercially available polymerase chain reaction
(PCR), the CT/NG Amplicor test (Roche Diagnostic Corporation, IN, USA). We sought to
compare field performance of a previously described POC assay [7] and a sensitive in-house 16S-RNA
NAAT using an improved visual detection of nucleic acid by dipstick [18], [19] using the CT/NG
Amplicor assay targeting one sequence coding for ORF1 (Open Reading Frame 1) of the
Chlamydia cryptic plasmid as the reference test. Organism load
was quantified with real-time quantitative PCR (qPCR) in CT positive individuals
[20], [21].
Patients were recruited using the medical clinical list, the local council
housing list and local knowledge of the Aboriginal Health Workers from five
Aboriginal communities with endemic trachoma in the Katherine region of the
Northern Territory Australia during a five week period over July and August 2007
as described previously [15], [17]. These communities had not received any recent
azithromycin or other mass antibiotic treatment, although local health services
do prescribe and dispense a range of broad-spectrum antibiotics on an individual
patient basis that might indirectly have had an impact on trachoma and our
findings. However, without access to confidential patient history, the
recipients of these antibiotic prescriptions were not identifiable. In
principle, school children in remote communities receive one annual health check
that includes trachoma screening, but this process is patchy at best. Despite
recommendations, very little screening for trichiasis in elders has been
conducted in area where trachoma is endemic [15].
During the clinical assessment, both examiners wore two pairs of gloves. To limit
the risk of cross-contamination between specimens and to prevent the exposure to
transmittable diseases to consecutive subjects, utensils and surfaces [5], the outer
pair was removed between successive participants and the inner pair disinfected
and regularly disposed of. The examiner graded the clinical signs of trachoma
using a fine grading scheme (Table 1) [15], [17]. Digital photographs were taken of the left inverted
upper lid that were subsequently graded independently using the fine grading
scheme and any discrepancies were adjudicated to give the final consensus
grading [15],
[17]. The
clinical grading was also expressed in terms of the WHO simplified grading
system as “Trachomatous inflammation-follicular” (referred to as
TFWHO), “Trachomatous inflammation-intense”
(TIWHO) and “Trachomatous scarring”
(TSWHO) [14].
After the clinical assessment and photography, two ocular swabs from the left eye
were collected consecutively under stringent conditions to limit
cross-contamination [5], and rigorous photographic cataloguing and sample
labelling systems. Over 95% of samples were collected by the same swabber
throughout the study to minimise any sampling variability. Additionally, gloves,
surfaces, loupes, the camera and other utensils were swabbed twice a day to
detect possible cross-contamination.
We obtained approval for the study from the Human Research Ethics Committees of
the Royal Victorian Ear and Eye Hospital, the Australian National University,
the Northern Australian National University and the Northern Territory
Government Department of Health & Communities Services and Menzies School of
Health Research. Signed written consent was obtained from each person, with
consent for children under 18 years of age being provided by a parent or
guardian [15],
[17].
Clinical assessment can be difficult and inconsistent when conducted by poorly
trained or inexperienced staff [5], [22]. Taking digital photographs, was previously described
as an alternative method and compared to clinical assessment [15], [22]. Briefly, the
majority of the examinations and taking digital photographs was done by examiner
A (96%) to minimise inter-observer variability while the remaining
examination were performed by examiner B who also examined and graded
independently digital photographs without prior knowledge of the clinical
assessment. In a masked fashion, both examiners re-examined photographs and gave
an adjudicated score when either the clinical grade or the photographic grade
was 3 or greater. A total of 88, 29 and 93 photographs were re-examined for the
presence of follicles, inflammation and scarring, respectively. Weighted kappa
analysis was previously reported to determine the concordance between methods.
The data indicated that there was 79.7% agreement
(kappa = 0.40) between clinical assessment, clinical
grading and photographic assessment of trachomatous follicles
(TF1-TF4) and 96.1% agreement
(k = 0.71) when the fine score was translated to
TFWHO. The agreement for TIWHO, and TSWHO
was 89.3% (k = 0.67) and 92.7%
(k = 0.67), respectively [15]. Previous studies have shown
the advantages of using finer scales to enhance the sensitivity of clinical
measurement, although finer grading schemes may reduce the concordance, or
frequency of perfect agreement, between the grades assigned by pairs of
independent observations [23].
The POC test was performed on site using the first left-eye ocular swab collected
by only one experienced technician throughout the study. The assay detects
chlamydial lipopolysaccharide (LPS) as previously described [7] with the
following modifications for field use: 1) an alternative nitrocellulose membrane
was used as the manufacturer discontinued the membrane previously used, 2) the
ratio of lyophilised signal amplification system was modified for the test to
function at high ambient temperature and 3) increased length of the conjugate
tube which houses the dipstick to minimise the evaporation of the reagents
during wicking and to protect the membrane against dust.
In addition to the lyophilised signal amplification reagents consisting of a
biotin-labelled monoclonal antibody to chlamydial LPS and an anti-biotin
monoclonal antibody conjugated to colloidal gold particles as colour indicator
[7], the
nitrocellulose-based membranes are the heart of lateral or vertical flow assays.
The wicking rate, pore size, residual surfactants and detergents present on the
matrix affect the characteristics of nitrocellulose-based membranes and reaction
kinetics. Therefore, changing this porous substrate matrix and the addition of
some features (i.e. shape of the conjugate tube) require a systematic adjustment
of ratio of the lyophilized signal amplification reagents.
The anti-biotin monoclonal antibodies (clone BII-10A12A9A1, Diagnostic
Development Unit, University of Cambridge, Cambridge, UK) conjugated to
colloidal gold (British Biocell International, Cardiff, UK) by passive
adsorption specifically bind to the lyophilised signal amplification reagents,
consisting of a biotinylated monoclonal antibody to chlamydial LPS detection
antibody (clone CTIII-10B9A10A4D28, Diagnostic Development Unit) biotinylated
with the BAC-Sulfo-NHS-LC-biotin reagent (Sigma, St Louis, MO, USA) at a ratio
of nine biotins per antibody molecule.
For LPS-POC testing, ocular swabs were placed in the sample extraction tube with
a tapered bottom to facilitate extraction of the swab and a cap that allows it
to also function as a dropper. The lysis reagent and analyte stabiliser were
added sequentially as previously described [7]. Briefly, the lysis reagent
(400 µL, Diagnostic Development Unit) and analyte stabiliser (300
µL, Diagnostic Development Unit) were added sequentially and mixed by
gently dipping the swab to the bottom of the extraction tube three times after
addition of each reagent. Two hundred microlitres of the above extract were
immediately transferred to 800 µL of pre-dispensed Amplicor sample
dilution buffer (Roche) for PCR testing. Thereafter, the signal enhancer reagent
(33 µL, Diagnostic Development Unit) was added to each extract. This
allows the release of chlamydial-LPS for detection. Five drops of the resulting
extract (100 µL) were transferred to the detection tube into which the
dipstick is placed. Two hundred microlitres of the above extract were
immediately transferred to 800 µL of pre-dispensed Amplicor sample
dilution buffer (Roche) for PCR testing. Thereafter, the signal enhancer reagent
was added to each extract. This allows the release of chlamydial-LPS for
detection. Five drops of the resulting extract were transferred to the detection
tube into which the dipstick is placed. The detection tube contains lyophilised
reagents of the signal amplification system consisting of biotinylated
monoclonal antibody to chlamydial LPS and anti-biotin monoclonal antibodies
conjugated to colloidal gold particles as the colour indicator. The dipstick
contains a nitrocellulose membrane, lined with another monoclonal antibody to
chlamydial-LPS (clone CVII-105A5A8, Diagnostic Development Unit) at the capture
zone, which captures the immune complex formed between the chlamydial-LPS and
signal amplification system reagents, if present. The accumulation of coloured
conjugate at the capture line of the dipstick generates a visible colour change
as previously described [7]. To generate a visual signal on the parallel to and
above the capture zone, the dipstick was lined with the anti-biotin antibody
described above, which served as the procedural control zone. All antibodies
were produced in-house and purified by affinity chromatography to more than
95% purity before use.
For PCR testing, 200 µL of the POC extract, obtained before adding
6% H2O2, were mixed with 800 µL of Amplicor
sample dilution buffer (Roche) and placed at 4 °C within 1 hr, and frozen at
–20 °C within 2 days until transport to Cambridge, UK in dry-ice.
These samples were stored at –80 °C until blind-tested by
Amplicor.
The second matched swab was stored dry on cold packs, frozen at –20 °C
within 2 days of collection and transported to Cambridge, UK in dry ice, and
stored at –80 °C until tested to minimise any target degradation. For
chlamydial and internal control testing, they were placed overnight in the
Amplicor M4RT-transport medium (3 mL, Roche) and tested by one experienced
technician according to the manufacturer's instruction (Roche).
All samples yielding a positive PCR result were quantified by previously
described ethanol precipitation and qPCR methods [20], [21]. Briefly, homogenized
M4RT-media (500 µL) from an Amplicor CT/NG Specimen Collection tube
(Roche) containing the ocular swab were aliquoted into a DNase/RNase free
siliconized tube (BioQuote, North Yorkshire, UK). Specimens were incubated at
room temperature for 10 min prior to centrifugation at 17,860 g
(max speed: 15,000 rpm) for 15 min at 25 °C (1.0R Megafuge). Supernatants
obtained from diluted M4RT-media were decanted with sterile filter tips and the
resulting pellets were re-suspended in 1 ml of cell culture grade
Dulbecco's phosphate-buffered saline (DPBS) lacking Ca2+
and Mg2+ (BioWhittaker, Walkersville, MD) by vortexing. The
re-suspended pellets were re-centrifuged as indicated above and re-suspended in
100 µL of 2M solution of ammonium hydroxide (obtained from a diluted 5N
ammonium hydroxide volumetric standard, Sigma-Aldrich, St. Louis, MO, USA).
Specimens were vortexed vigorously, incubated at room temperature for 10 minutes
and vortexed again. If the pellet had not dissolved, it was solubilized by
repeat pipetting and continuous cycle of vortexing until dissolved. Each
specimen was placed into a heating block and heat-treated at 95–100 °C
for 1 hour, or until the ammonia had evaporated (dry tubes). Dried specimens
were re-suspended in 500 µL of molecular reagent-grade water and,
vigorously vortexed and incubated at room temperature for ≥30 minutes to
ensure that any precipitate had re-dissolved. The extracts were stored at 4
°C and tested within 24 h. The above extracted samples and standard curves
were prepared and amplified in duplicate on two different days (4 data points)
by Real-time qPCR.
Real-time qPCR was performed using a previously described method [20], [21]
targeting one sequence coding for ORF1 of the Chlamydia cryptic
plasmid [20],
[21].
This method was previously demonstrated highly reproducible
(R2 = 0.998) and with analytical sensitivity of
<10 copies per amplification [21]. The previously
described reproducibility was established against eleven standard curves
constructed for the EB standard on different days. Each curve was generated from
seven serial 10-fold dilutions of the pCTL12A plasmid amplified in duplicate. In
addition, previously published data showed that 7.72±0.68 (mean ±
SD) plasmid copies corresponded to one elementary body of C.
trachomatis (serovar L1), consistent with previously obtained
values [20].
Analysis of genital clinical specimens revealed a strong correlation
(R2 = 0.929) between
elementary body counts determined by a quantitative ligase chain reaction
(LCR)–based Chlamydia trachomatis LCx Assay (Abbott
Laboratories) which targets a conserved region of the cryptic plasmid and those
determined by the current qPCR method [20]. Although most of the
infected patients were likely to harbour C. trachomatis
serovars A, B, Ba and C, the primer sets for both Amplicor and qPCR assays
correspond to conserved regions of the C. trachomatis cryptic
plasmid and are therefore able to detect all C. trachomatis
serovars. Through the present analysis, the organism load was expressed in
number of plasmid per swab and not in EB per swab even though, to the knowledge
of the authors, it has not been reported that the number of cryptic plasmid
significantly varies between serovars. In addition, the second swab from
patients identified as Amplicor-positive and 50 randomly selected
Amplicor-negative samples with or without clinical signs were tested in
duplicate with the 16S-RNA assay. The second swabs were tested in a masked
fashion (randomised order) by Amplicor and the 16S-RNA assay. Sample that
yielded a positive result on the first swab, but was negative on the second
swab, was re-tested in a chessboard manner in presence of known positive and
negative samples and, positive and negative controls.
Amplification of RNA extracted samples (total RNA RNeasy Mini Kit, QIAGEN Inc.
Valencia, CA, USA) were performed by isothermal amplification and amplified
products detected visually on a dipstick as described previously [18], [19], [24], [25]. The test
designated as SAMBA (Simple AMplification-Based Assay) is based on a proprietary
technology [18],
[19]. Primer
and probe target conserved sequences for all the 16sRNA Chlamydia
trachomatis serovars obtained from the American Type Culture
Collection (ATCC; MD, USA). Regions are conserved for all Chlamydia
trachomatis serovars and were selected as previously described for
the diagnosis of 2009 pandemic influenza (H1N1) [25] with sequences obtained from
the National Center for Biotechnology Information (http://www.ncbi.nlm.nih.gov) and analysed with Jalview 2.3
(University of Dundee, UK). Detector and capture probes [24] were also designed to target
similarly these specific regions. The primers and probes were compared with the
Nucleotide Collection database at NCBI with the use of the Basic Local Alignment
Search Tool (BLAST). Specificity was established against a panel of
microorganisms commonly associated with human eye and skin (e.g. Staphylococcus,
Pseudomonas, Streptococcus, Escherichia,
Proteus and Candida, obtained from ATCC). The SAMBA
Chlamydia in a closed device to prevent amplicon
contamination has the same unique characteristics of the previously described
SAMBA HIV-1 test chemistry render it suitable for near-patient testing in both
developed and developing countries because the test uses thermostable reagents
and a simplified protocol with minimum sample processing [24].
In brief, after amplification, the amplicon was incubated in a 2 mL
microcentrifuge tube at 41 °C on a heating block. 20 µL of
amplification product were added to a proprietary detection mixture and the
dipstick was inserted in the reaction mixture. The test results were examined
after 25 min of incubation and signal on the dipstick scored by an experienced
operator according to the in-house scoring chart [25].
To assess the quality of the sampling procedure, human genomic DNA was quantified
with Double-Dye Taqman kit according to the manufacturer's instruction
(Primer Design, Southampton, UK) in all PCR positive samples and 41 randomly
selected PCR-negative samples. The primers of human genomic DNA kit detect a
single copy region of non-transcribed DNA.
Statistical analysis was performed with SAS v9.1 software. Confidence intervals
(CI) were calculated as exact binomials. The geometric mean of organism load as
well as its respective standard deviation (SD) and 95% confidence
interval (CI) for the ocular swabs were calculated from the natural log
transformation of the organism load obtained for each swab. The organism load of
the ocular samples was compared between the first and second grades of the
clinical signs by the Student's t-test, unequal variance
t-test Satterthwaite and equal variance pooled t-test. The
correlation between organism load and the fine grading scheme, the load first
swabs and the organism load of seconds swabs, and organism load between the
different population was obtained using Spearman Rho (R)
coefficients and paired Wilcoxon rank tests. Reliability of PCR positivity of
both swabs or with 16S-RNA positivity was assessed with the kappa coefficient
and its 95% confidence intervals. A p-value of <0.05
was considered statistically significant.
We examined 1316 of 1545 potential participants, giving an overall examination
rate of 85.2% (Figure
1). A total of 1282 participants were eligible for this analysis with
a median age of 17.1 years (range: 0.1–95). Each participant was assessed
for clinical signs of trachoma using a fine grading scheme (Table 1, [15]), by PCR
(Roche) and by the LPS-based POC assay.
On the basis of clinical examination of both eyes of each subject, 135
participants had active trachoma (10.5%; 95% CI: 8.9–12.2),
130 with TFWHO and five had TIWHO without
TFWHO. Taking the findings of both eyes the highest age-specific
prevalence of TF was in children 2–4 year-old (33/121; 27.3%)
followed by 5–9 year-old (51/229; 22.3%) and those with less than 2
years of age (13/61; 21.3%). The overall prevalence of TFWHO
in children aged 1–9 was 25.1% (96/383, 95% CI:
20.7–29.4) and the prevalence of the five communities ranged from
9.8% to 38.5% (4/41 [9.8%], 5/49
[10.2%], 23/115 [20.0%], 27/82
[32.9%] and 37/96 [38.5%], respectively).
In contrast, when clinical data were limited to the left eye in order to directly
compare with the PCR and POC testing results, the frequency of clinical signs of
active trachoma decreased to 8.6% (110/1282), 108 with TFWHO
and two had TIWHO without TFWHO (Table 2). The resulting prevalence in the
five communities in children aged 1–9 was 0/41 (0%), 4/49
(8.2%), 21/115 (18.3%), 21/82 (25.6%) and 36/96
(37.5%).
The PCR (Amplicor) positivity rate in the population was 3.6% (46/1282,
95% CI: 2.6–4.6) and, in children aged 1–9, 6.8%
(26/383, 95% CI: 4.3–9.3). Of the PCR positive participants, the
highest rate was in children 5–9 year-old (17/46; 37%) followed by
2–4 year-old (9/46; 19.6%) and 10–14 year-old (8/46;
17.4%, Figure 2). Of
the 46 people for whom the first swab from the left eye was PCR-positive, on
testing of the second swab, 43 (93.5%, 95%CI: 86.3–100) were
PCR-positive and 44 (95.7%) were 16S-RNA-positive. Two of the three
PCR-negative second swabs were 16S-RNA-positive and one of those had the lowest
organism load on the first swab. There was a good agreement between the first
and second swabs tested with PCR (kappa coefficient 0.97; 95% CI:
0.93–1.00) and between the first swab PCR and the 16S-RNA result (kappa
coefficient 0.98; 95% CI: 0.95–1.0).
The CT organism load was analysed by qPCR in all PCR-positive swabs (Figure 3). The geometric mean
organism load was 55,585 (95%CI: 801–3,811,754) pCTL12A plasmid per
swab for the first swab and 4,355 (95%CI: 98–193,602) for the
second. The mean organism load for the second swab was 12.8 times lower
(95%CI: 0.79–566.3) than for the first (paired Wilcoxon rank sum
test, P<0.0001). The organism load in the first swab was
strongly correlated with the load in the second swab (Spearman
Rho = 0.74,
P<0.0001).
To confirm the adequacy of specimen collection, human genomic DNA was quantified
for the 46 PCR-positive (1.25±0.69 µg of genomic DNA/swab) and 41
random PCR-negative (including nine participants presenting signs of
TFWHO – 1.33±0.70 µg of genomic DNA/swab). The
amount of genomic DNA was not significantly different between positive and
negative ocular samples (two-tailed P = 0.6), nor was there
a correlation between the organism load and the quantity of genomic DNA/swab.
All of the 32 control swabs of potential formites were negative by PCR.
A significant correlation was observed between PCR-positivity (Amplicor) and
TFWHO (Wilcoxon rank sum tests P<0.0001) and,
between PCR-positivity and the fine grading scheme (Spearman
Rho = 0.98 and
P = 0.0004) (Table 2). A higher proportion of people were
PCR positive as clinical disease, as assessed by the fine grading, became more
severe. However, it should be noted that 59% (27/46) of PCR positive
results occurred in people with TFWHO, although only 13%
(6/46) occurred in people with TF0.
As result, the agreement between PCR and TFWHO was poor for children
of ≤9 year of age (k = 0.15; 95%
CI: 0.01–0.25) and still poor
(k = 0.23; 95% CI: 0.08–0.37)
for older participants. Figure
4 describes the age-specific prevalence of the left eye fine grading
of TF1 (Figure
4A), TF2 (Figure
4B), TF3 (Figure 4C) and TF4 (Figure 4D) versus PCR positivity.
Of particular interest were six PCR positive participants who had not have active
follicular disease and were graded as TF0. All were female whose ages
were 9, 16, 21, 57, 66 and 73 years. Two lived in houses with children who were
PCR positive. Another two lived in houses in which three or more children had
TF3. The fifth woman was aged 73 and had TI1 and
TS3 with 30,918 plasmid/swab. She shared a house with two men,
one aged 28 who had TF2 and TI1 and the other aged 58 with
TF1 and TI1. The sixth was a 9 year-old girl with a
normal exam and 37,074 plasmid/swab whose house number was missing so her
household contacts could not be identified. Therefore, with the exception of the
last girl, a plausible case can be made for exposure to infection and four of
five had some signs of inflammation (TI of some degree).
The fine grading of TF0–4 (Figure 5A), TI0–4 in
presence of TFWHO (Figure 5B), TI0–4 in absence of TFWHO
(Figure 5C) and
TS0–4 in presence of TFWHO was positively
correlated with the organism load whereas there was no correlation for
TS0–4 in absence of TFWHO (Spearman Rho
(R) = 0.498 and
P = 0.0004,
R = 0.473 and
P = 0.0009,
R = 0.438 and
P = 0.0023,
R = 0.449 and
P = 0.0017 and
R = −0.039 and
P = 0.7946, respectively). The mean
organism loads for the WHO grades were: TFWHO present 133,252
plasmid/swab (95% CI: 2,173–8.2×106),
TFWHO absent 26,903 plasmid/swab (95% CI:
534–1.4×106), TIWHO present 400,312
plasmid/swab (95% CI: 38,101–4.2×106) and
TIWHO absent 40,135 plasmid/swab (95% CI:
655–2.5×106).
The LPS-based POC assay yielded 18 positive individuals, 14 were PCR positive and
4 were young adults who were PCR negative and without clinical disease
(TF0 and TI0). Using PCR as the comparator test, the
sensitivity and specificity of the POC was 30.4% (14/46, 95% CI:
17.1–43.7) and 99.7% (1232/1236, 95% CI: 99.4–100),
respectively.
The poor correlation between the prevalence of clinically active trachoma and
evidence of infection is not new [12], [13], especially in low prevalence communities [5]. As with any
infectious disease, there is an initial incubation period (4–8 days) between
inoculation and the development of clinical disease [5]. This is followed by frank
disease when both bacterium and clinical signs co-exist, and a later stage when the
infection is no longer present or cannot be detected by diagnostic tests, yet the
clinical signs persist as disease slowly resolve [5], [26]. In humans, the lag period
between the last detectable bacterial shedding and the resolution of the active
disease may take up to 9 months or so [5], [27]. As previously observed [7], the prevalence
of active trachoma varies when one or both eyes are considered, in this study from
8.6% to 10.5%, respectively [7], [28]. For practical and economic
reasons, swabs for PCR, LPS-based and 16S-RNA testing, and photographs were only
collected from the left eye. Therefore, clinical/laboratory diagnostic comparisons
were made only for the left eye using the consensus grading based on clinical and
photographic data.
To reduce the likelihood of over-grading of clinical disease, the assessment was made
both in the field using frequent reference to the WHO grading card and by
independent photo-grading. Although over-grading can still occur, this combined
approach reduces the risk.
A comparison of the performance data of the LPS-based POC with an analytical
sensitivity of 2,500 chlamydial elementary bodies [7] from Tanzania and the current
study in Australia is interesting. Although the prevalence rates of TFWHO
in 1 to 9 year olds are roughly comparable; 28% and 21% respectively,
the intensity of disease (the proportion of those with TF and/or TI who have TI) was
nearly three times higher in Tanzania (25% compared to 9.5%) as was
the mean organism load (147,267 compared to 55,585 plasmids/swab). The unequal
variance t-test Satterwaite using load
(P = 0.0057) and equal variance pooled t-test
using natural log transformation of the organism load indicated
(P = 0.0333) that organism load difference
between those samples collected in Aboriginal communities and those samples
collected in the Masai communities were significantly different. The lower intensity
of disease in Australia is reflected in the two to three times lower rates of PCR
positivity in both those with TFWHO (16% in Australia and
44% in Tanzania) and those without TFWHO (4.3% and
9.7%). Similarly the reduced performance of the LPS-based POC assay in
Australia may in part reflect the lower organism load and in part the modification
of the test. The reduced disease severity and infectious load observed in Australian
Aboriginal communities may reflect the dramatic differences in medical, environment
and living conditions between the Masai and Aboriginal people.
Even though the detection of infection by PCR is a poor predictor of the presence of
clinical disease and equally clinical disease was poorly correlated with infection,
organism load was strongly correlated with the prevalence and severity of active
trachoma as graded by the finer grading scheme and that 46% of infection was
found in people who did not have the WHO grade of TF but who still had some milder
clinical changes (TF1 or TF2). As mentioned, organism load
also correlated with the fine grading of trachoma. Similar findings come from an
earlier study that used a roughly similar finer grading scheme and that used both
tissue culture and direct fluorescent antibody cytology to detect infection [13]. That study
also found the load of infection was higher in those with more severe disease (WHO
grade TF) than in those with less severe clinical disease.
In that study with a less sensitive assessment of infection 12% of infection
was in those who did not have the simplified WHO grade of TF or TI. A rapid, simple
and affordable POC test capable of accurate identification of active infection would
nevertheless be a useful tool in trachoma control. The 16S-RNA test, a closed-system
device based on visual detection of nucleic acid on a dipstick, offers several
advantages and with its inherent sensitivity and thermostability and therefore has
the potential to meet this need [29], [30]. Because of the sealed containment of amplified sample a
dedicated laboratory is not require. Assay reagents have been converted into
thermostable formulas (stability tested at 55 °C for one month and 37 °C for
12 months, data not shown) and the test uses a simplified protocol with minimum
operation and sample processing made possible by a disposable modular cartridge
previously described by Lee et al
[24]. This
technology, Simple Amplification-Based Assay (SAMBA) has a substantial advantage
over currently available NATs, in that it is able to provide results quickly and
on-site, thereby facilitating appropriate clinical support. The simplicity of SAMBA
tests will allow their use in resource-poor settings in developing countries and for
near-patient testing in the developed world [24]. These features may offer
advantages over the LPS-based POC, although further evaluation of the 16S-RNA
Chlamydia SAMBA test in areas of varying trachoma prevalence is required to better
assess its performance and cost effectiveness.
Of considerable interest is the small number of individuals who had a positive PCR in
the absence of a detectable follicular response (TF0). Some were older
people with scarred conjunctiva who may be incapable of mounting such an immune
response. Others may have had a transitory infection or may have been in an
incubatory phase of infection. As demonstrated in this study, a rational explanation
for a possible source of infection was identified in the household in all but one
case. In this case an elderly woman who was PCR positive shared a house with two
other adults. A treatment program that only targeted households with children would
miss treating households such as this. However, a significant problem in these
communities is the frequent sharing of houses and it is not unusual for children to
sleep in two or three different houses each week [5]. We were not able to track this
sort of movement to identify extended families. More precise data of people's
movements and behaviour would be needed to explore this in further detail.
The strengths of this study include firstly the careful documentation and
quantification of the two outcomes, the presence of infection and clinical trachoma.
Careful specimen collection and handling limited sampling cross-contamination
critical when highly sensitive NAATs capable of detecting low level of targets are
used. Poor sampling methods and risk of cross-contamination in the field have caused
concern about the validity of some early NAAT findings [5], [31]. We observed no
significant difference between the quantity of human genomic DNA contained in
negative and positive PCR swabs that makes inadequate sample collection unlikely.
The repeat negative control swabs of fomites suggest that the field precautions
taken to prevent cross-contamination were successful. Further, the positioning of
the samples in the test plate was verified to ensure that neither positive specimens
nor positive controls were adjacent on repeat testing. Quantitative PCR was used to
assess the infectious load in all positive specimens.
Secondly, this study used a fine or semi-quantitative scale to grade the clinical
severity of trachoma and developed a consensus-based grade that used both clinical
and photographic findings. This allowed for a much more detailed assessment of the
presence of clinical signs and the recognition of less advanced disease than the
more frequently used WHO simplified trachoma grading system [14]. The use of a finer grading
scheme is especially important in detailed research studies looking for correlation
between clinical disease, environmental or genetic risk factors and the distribution
of infection in a community when attempts are made to separate those with
“disease” from those who are “asymptomatic”. A finer grading
scheme allows the identification of subjects with clinical signs below the threshold
set by the simplified WHO grading scheme because many people with less advanced
disease do not fit in the clear-cut definitions used in the WHO simplified trachoma
grading system.
Finally, this study was a population-based study with high community coverage that
included people of all ages.
Potential weaknesses include the inability to precisely link the exposure to
infection of individual participants as in these communities children in particular
may move frequently from one house to house another. In addition, there were
inevitably some missing data for both people and house numbers. The study did not
assess the sampling, testing, clinical-grading and photo-grading agreement between
swabbers, laboratories and examiners because the same swabber, laboratory and
clinical-grading or photo-grading examiners were used throughout the study. The
inter-operator agreement and intra-operator reproducibility of the LPS-based test
[7] and
clinical- versus photo-grading agreement [15] have been described
elsewhere.
However, the key messages of this study are that the use of a binary grading system
like the simplified WHO grading system will miss many of the more subtle issues
around the identification of chlamydial infection and clinical status [13]. Specifically,
53% of the clinical disease (TF1 or TF2) and 59%
of PCR positivity occurred in people with disease less severe than the WHO grade of
TF. These data suggest that detailed studies of the pathogenesis, distribution and
natural history of trachoma should make use of both a finer grading scheme and also
quantify the infectious load. The fine grading also revealed a high proportion of
people with clinical disease who were below the WHO threshold for TF. Most of the
active trachoma was seen in the younger children who also had the more severe
disease. Both the severity and prevalence of TF decreased with age. However, the
finding of a significant proportion of people with few follicles, TF1 is
noteworthy. Although some of this will reflect the waxing and waning of active
trachoma, it suggests that in some people at least a few follicles, once formed, may
persist for a long period of time or indefinitely. It is unclear whether these
“persistent” follicles reflect occasional exposure to chlamydial
antigens or infection or if they are a permanent tissue change.
Organism load varies in areas with different levels of endemicity and intensity of
disease. In areas with a lower prevalence or intensity, laboratory tests may be of
limited used for community-based assessment. However, it is in these situations that
these tests would be of most use for the confirmation of sporadic cases with
clinical disease. In addition, when the upper tarsal conjunctiva was swabbed
transversally to collect an appropriate specimen for testing, the organism load of
the consecutive swab was dramatically decreased even though the amount of collected
cells was similar. This great disparity in organism load raises concerns about the
use of consecutive specimens in preference to split specimens.
The second generation NAATs capable of detecting high multiple-copy of targets such
as ribosomal rRNA should enhance analytical sensitivity [32]. This may enable to detect
some low level infection previously missed by PCR [32], [33] or other less sensitive
methods such as Real-time qPCR targeting a single copy genomic sequence such as the
major outer membrane protein gene (omp1) or the outer membrane
complex B protein gene (omcB) instead of the multiple-copy
sequences (Chlamydia cryptic plasmid), LPS-based rapid test [7], culture and
direct immuno-fluorescence (DFA), and so extend the period of detectable infection.
A point-of-care nucleic acid amplification test based on targets with multiple
copies such as 16S-RNA would be a more appropriate tool to detect low level of
infection previously missed by LPS-based rapid test [7].
Finally, the current prevalence of both active trachoma and trichiasis are roughly
similar to that reported 30 year ago in this region by National Trachoma Eye Health
Program [15] and
both of which exceed the thresholds set by WHO to define blinding trachoma as a
public health problem indicate the need for appropriate interventions to control
trachoma and prevent blindness in these five Aboriginal communities.
|
10.1371/journal.pgen.1004897 | A Discrete Transition Zone Organizes the Topological and Regulatory Autonomy of the Adjacent Tfap2c and Bmp7 Genes | Despite the well-documented role of remote enhancers in controlling developmental gene expression, the mechanisms that allocate enhancers to genes are poorly characterized. Here, we investigate the cis-regulatory organization of the locus containing the Tfap2c and Bmp7 genes in vivo, using a series of engineered chromosomal rearrangements. While these genes lie adjacent to one another, we demonstrate that they are independently regulated by distinct sets of enhancers, which in turn define non-overlapping regulatory domains. Chromosome conformation capture experiments reveal a corresponding partition of the locus in two distinct structural entities, demarcated by a discrete transition zone. The impact of engineered chromosomal rearrangements on the topology of the locus and the resultant gene expression changes indicate that this transition zone functionally organizes the structural partition of the locus, thereby defining enhancer-target gene allocation. This partition is, however, not absolute: we show that it allows competing interactions across it that may be non-productive for the competing gene, but modulate expression of the competed one. Altogether, these data highlight the prime role of the topological organization of the genome in long-distance regulation of gene expression.
| The specificity of enhancer-gene interactions is fundamental to the execution of gene regulatory programs underpinning embryonic development and cell differentiation. However, our understanding of the mechanisms conferring specificity to enhancers and target gene interactions is limited. In this study, we characterize the cis-regulatory organization of a large genomic locus consisting of two developmental genes, Tfap2c and Bmp7. We show that this locus is structurally partitioned into two distinct domains by the constitutive action of a discrete transition zone located between the two genes. This separation restricts selectively the functional action of enhancers to the genes present within the same domain. Interestingly, the effects of this region as a boundary are relative, as it allows some competing interactions to take place across domains. We show that these interactions modulate the functional output of a brain enhancer on its primary target gene resulting in the spatial restriction of its expression domain. These results support a functional link between topological chromatin domains and allocation of enhancers to genes. They further show that a precise adjustment of chromatin interaction levels fine-tunes gene regulation by long-range enhancers.
| Differential regulation of gene expression transforms shared genomic information into the cell type-specific programs underlying organismal development and homeostasis. In vertebrates, it is not uncommon to find gene regulatory elements, in particular enhancers, hundreds of kilobases away from their target gene (reviewed in [1], [2]). The mere scale of this genomic distance raises the question of how enhancers and promoters can find each other, and how enhancers distinguish between their specific target and other neighboring genes, which may even lie much closer. Understanding the molecular basis of such specific interactions is essential as their impairment can lead to mis-expression of the normal target gene [3], [4] or to inappropriate activation of neighboring genes [5]–[8], with often severe phenotypic consequences [7], [9]–[12].
Enhancers can typically activate transcription from different promoters, a property that is part of their initial definition [13] and which has been amply used to assess enhancer activity. Many enhancers act pervasively across their endogenous genomic surroundings [14], [15], and enhancer sharing is not unusual between neighboring genes, particularly within multigenic clusters [16]–[22]. Noteworthy, this can also occur between genes with no functional relationship except genomic proximity [9], [23]–[25]. Nonetheless, in many loci, adjacent genes exhibit distinct expression patterns, implying the existence of mechanisms that limit the promiscuous potential of enhancers.
Different mechanisms and genomic elements have been invoked to explain enhancer-target gene specificity. They can be divided in two main categories, depending on whether they may promote interactions (eg. nature of the promoter, tethering elements [26], [27]), or block them. Amongst the latter, insulators prevent contact of an enhancer with an adjacent promoter, when placed in between [28]–[30]. This capacity of insulators to organize the genome in separate regulatory compartments designate them as critical components in ensuring specificity of cis-regulatory interactions [31]. However, only a handful of insulator elements have been functionally assessed in their native genomic context, and therefore their mode(s) of action is still poorly understood. Contrary to earlier models, a growing body of evidence suggests that insulators do not function autonomously, but rather through higher-order 3D conformations [32].
The necessity to consider the genome's three-dimensional organization is further highlighted by genome-wide high-resolution interaction maps obtained by chromosomal conformation capture techniques [33]. These studies revealed that the genome is compartmentalized in topologically-associating domains (TADs) [34], [35]. TADs have been proposed to contribute to gene expression by limiting enhancer action [36], [37]. In support of this view, genes located within the same TAD tend to be expressed coordinately [35], [38], and TADs have been found to encompass the regulatory domains defined by long-range enhancer activities [15], [39]. Recent works have addressed the finer-scale structural organization of TADs, revealing a complex hierarchy of interactions, which may contribute to mediate long-distance interactions between enhancers and promoters [40], [41] and to subdivide them into distinct regulatory domains [15]. In most instances, the functionality of structural contacts is difficult to evaluate precisely and the causal relationship between structural conformation and gene regulation remains unclear.
To better understand the relationships between 3D structural properties of the genome and enhance-promoter allocations, we focused on a large interval of approximately 0.5 Mb containing two different developmental genes, Bmp7 and Tfap2c. These two genes, which encode a secreted signaling molecule and a nuclear transcription factor, respectively, are active in multiple tissues and organs during embryogenesis [42]–[48]. Both genes have promoter architectures compatible with tissue-specific and long-distance regulatory inputs [49]. Their expression overlaps in the limbs, forebrain and branchial arches of mid-gestation mouse embryos, while in other contexts, their expression is specific of one or the other and exclusive. Therefore this locus constitutes an ideal system to study the control of long-distance enhancer specificities.
To investigate the regulatory organization of this locus, we used a transposon/recombination-based chromosomal engineering approach [14]. We show here that the genomic interval consists of two largely independent regulatory domains, corresponding to each of the two genes. Analysis of the chromatin conformation of re-engineered genomic configurations identified a central transition zone (TZ) that defines different topological sub-domains. Importantly, the allocation of enhancers to one or the other gene is determined by this partition. Altogether, our data support the view that the topological organization of the genome restricts enhancers to specific domains, determining therefore their “specific” target gene choice. Interestingly, we found that the presence of Bmp7 in cis has a mild influence on the expression level of Tfap2c in the developing forebrain, indicating that the position of the two genes to different topological domains does not lead to an absolute insulation.
To determine the regulatory organization of the Tfap2c-Bmp7 locus, we adapted the GROMIT (Genome Regulatory Organization Mapping with Integrated Transposons) strategy [14]. Firstly, at the 3′ end of the endogenous Bmp7 gene, we inserted a transgene consisting of a Sleeping Beauty transposon comprising 1) a regulatory sensor gene (a LacZ reporter under the control of a short naïve synthetic promoter region derived from the human β-globin gene [14], [50]) and 2) a loxP site. After establishment of a mouse line carrying the correct insertion, we removed the selection marker used to identify candidate targeted ES clones, a step which left behind an additional loxP site, next to the Sleeping Beauty transposon. We designated this allele as SB-B(3end) (Fig. 1). By serial remobilisation of the transposon in vivo [14], we obtained several insertions located in this region of mouse chromosome 2 (S1 Table). Of these, seven insertions were distributed along the Tfap2c-Bmp7 locus (Fig. 1A): three very close to SB-B(3end) (within 23 kb), one (SB-B(up)) 20 kb upstream of Bmp7, and another one in the first intron of Bmp7 (SB-B(in)). The remaining two (SB-A1 and SB-A2) lie within the large intergenic region separating Tfap2c and Bmp7. In parallel, we established a mouse line (BA0758) from an ES clone carrying a βgeo gene-trap insertion in Tfap2c [51].
We analyzed the expression pattern of the regulatory sensor at different insertion sites in E10.5 to E12.5 mouse embryos, at stages when Tfap2c and Bmp7 show both shared and specific expression patterns (Fig. 1, S1 Fig.). The two insertions located between Tfap2c and Bmp7 (SB-A1 and -A2) showed very similar LacZ staining in the oro-facial region, the branchial arches, and in the forebrain (Fig. 1B, left). These three expression domains are strikingly consistent with reported expression patterns of Tfap2c [42] and particularly with the Tfap2c LacZ gene-trap allele (Fig. 1B- S1 Fig.). This overlap and agreement in expression suggested that SB-A1 and -A2 were included in the Tfap2c “regulatory domain” [15]. The expression of the reporters showed however different relative intensity between the lateral and medial part of the forebrain: while BA0758 and SB-A1 were preferably expressed in the lateral forebrain, with weaker expression in the medial region, SB-A2 showed the inverse pattern, with a stronger medial than lateral LacZ staining. Such position-effects (the promoter is the same for SB-A2 and SB-A2) are not uncommon within regulatory domains [15], [52]. They may reflect the presence in the locus of several forebrain enhancers with distinct medial/lateral activity and different range of action.
These forebrain expression domains were not observed with any of the four insertions located within the 23-kb region at the 3′end of Bmp7 (Fig. 1B, S1 Fig.), suggesting that the telomeric limit of Tfap2c regulatory domain is upstream of this region. More distant insertions in Bmp7 (SB-B(in); SB-B(up)) showed weak medial-only forebrain expression at E11.5, with no lateral expression detected, as also observed for Bmp7 [44]. None of the six telomeric insertions showed the characteristic oro-facial expression observed with the Tfap2c-associated insertions. In contrast, they shared several common expression domains not reported by the SB-A1 and –A2 insertions (Fig. 1B). The four insertions at the 3′end of Bmp7 and SB-B(in) showed all prominent staining for LacZ expression in the developing heart (from E10.5 to E12.5), and in the interdigital mesenchyme (at E12.5). SB-B(up) displayed only faint LacZ staining in the interdigital mesenchyme, and no staining in the heart. However, LacZ expression from this position overlapped characteristically with other SB-B insertions in the whiskers, nasal pits, and forebrain (S1 Fig.), defining collectively a regulatory domain distinct from the one associated with Tfap2c. This domain includes Bmp7, and accordingly, several of the reported activities overlap with known Bmp7 expression domains [47], [53]. Some regions of the Bmp7 expression domain were not reflected accurately in the activity of the SB reporters, being either missing or spatially expanded. These differences may arise from the limited range of action of some promoter-proximal enhancers [53], and/or from the different post-transcriptional stability and dynamics of LacZ compared to the endogenous Bmp7 transcripts.
Overall, the regulatory activities detected by the sensor differed significantly between the centromeric and telomeric part of the locus, and highlighted two distinct and non-overlapping regulatory domains, each defined by multiple distinct tissue-specific activities, one domain corresponding to Tfap2c and the other to Bmp7. We focused for subsequent analyses on the forebrain (medial and lateral) and heart, as representative markers of these two domains. In these two tissues, the expression pattern of the different genes is stable from E10.5 and E12, contrasting with the dynamic expression of these genes in the developing limbs and face. Also, for these two expression domains, it is technically possible to dissect from embryos the part where the gene or the enhancer is active, without the contribution of too many non-expressing cells.
To further characterize the functional relevance of these two domains and associated enhancers, we used in vivo Cre-mediated recombination to engineer chromosomal deletions removing either the telomeric half or the whole of the intergenic region (Fig. 2). Each deletion was produced using a combination of loxP sites in cis and trans [54] in order to keep the LacZ sensor at the deletion breakpoint (see Materials and Methods). With the TAMERE strategy, we also obtained a large duplication, reciprocal to del3 (S2 Fig.). All three deletions led to a complete loss of LacZ expression in the embryonic heart and forebrain (Fig. 2B) suggesting that the enhancers detected by SB-A1 and SB-B(3end) lie in the region encompassed by del1. Dup3-lacZ embryos showed LacZ expression in the heart similar to SB-B(3end), corroborating the presence of the heart enhancer(s) at the 3′ side of Bmp7 (S2 Fig.). These deletions also provided information on the locations of additional enhancers associated with other expression domains (S2 Fig.).
We next determined if the enhancers present in the del1 interval contributed to Tfap2c and Bmp7 expression by whole-mount in situ hybridization and RT-qPCR (Fig. 2C–D). In del1 homozygous embryos, Bmp7 expression was drastically reduced in the heart compared to wild-type littermates, while the very weak expression of Tfap2c in the heart was unaffected (Fig. 2C). In the forebrain, where both genes are expressed, we found an almost complete loss of Tfap2c expression in both the medial and lateral parts of del1 embryos. In contrast, Bmp7 expression was barely affected and showed only a slight reduction in the lateral forebrain (Fig. 2C).
These analyses demonstrated a critical role of elements located within the del1 segment for the specific expression of Tfap2c in the forebrain and of Bmp7 in the heart, respectively. Several peaks enriched for chromatin marks associated with active enhancers (H3K27ac, EP300) have been detected within this region in the forebrain and the heart of E11.5 embryos [55]–[57] (S3 Fig.). Interestingly, the distribution of these regions is coincident with the location of the two regulatory domains. Many forebrain H3K27ac peaks are located between Tfap2c and SB-A1/A2, while the only ones present around Bmp7 lie in the first intron of the gene. Conversely, heart H3K27ac-enriched elements cluster around the 3′ end of Bmp7. H3K27ac peaks were also identified outside of the del1 region around the locus. The forebrain H3K27ac peak adjacent to Bmp7 could account for its unaffected expression in del1; however, the role of the predicted forebrain and heart enhancers located respectively centromeric and telomeric to del1, respectively, remained unclear, as they were seemingly unable to confer significant activity to the reporter gene or to the endogenous genes in these tissues, in the absence of del1 sequences.
To confirm that del1 contained enhancers with the expected activities, we cloned FB1, an evolutionarily conserved element enriched for both H3K27ac and EP300 in the forebrain, upstream of the regulatory sensor construct. In this transgenic assay, FB1 drove specific and reproducible LacZ expression in the forebrain in E11–12 embryos (Fig. 2E), including the Tfap2c expression domain. However, FB1 appeared broadly and equally active in both medial and lateral forebrain, contrasting with the restricted expression detected by the same reporter gene than the one used in the transgenic assay when inserted in the endogenous locus on either side of FB1. In this context, it showed alternatively preferential expression in the lateral (SB-A1, like Tfap2c) or medial (SB-A2). These differences suggested that additional factors – possibly the other H3K27ac-region present in the vicinity (see below) – may modulate FB1 intrinsic activity in a position-dependent manner. Amongst the predicted heart enhancers, one of them (mm75) had been tested previously [58] and reported to have broad enhancer activity in the heart of E11.5 mouse embryos (Fig. 2F).
Taken together, these data demonstrated that the del1 region contained heart-specific and forebrain-specific regulatory element(s) critical for the expression of Bmp7 in the heart, and of Tfap2c in the forebrain, respectively. Importantly, these elements appeared to be dispensable for the regulation of one another's genes. These selective influences and the separate location of the different enhancers further confirmed the partition of this genomic interval into two distinct regulatory domains containing enhancers which act exclusively on one or the other gene (Fig. 2G).
We next investigated how the regulatory subdivision of the locus corresponded to its topological organization. Hi-C data available for mouse ES cells and cortex [34] suggests that the locus has a relatively loose topological structure, confined between two prominent topologically associating domains (Fig. 3A, S4 Fig.). To determine the pattern of physical contacts involving Tfap2c and Bmp7, we carried out circular chromatin conformation capture experiments followed by high-throughput sequencing (4C-Seq) using the promoters of these two genes as viewpoints (Fig. 3). We performed these 4C-Seq analyses on dissected samples where one and/or the other gene were expressed (E11.5 heart, medial and lateral forebrain) and whole body of E11.5 embryos (where most cells are non-expressing either of the two genes). We also included samples from E12.5 limbs, which comprised a majority of non-expressing cells.
For both viewpoints, the 4C profiles highlighted a large primary interaction domain characterized by high 4C read counts (Fig. 3B, C). We applied a segmentation algorithm [59] to delineate this primary domain in the different conditions (S2 Table). The calculated primary interaction domains for a given viewpoint were nearly identical across the different tissue samples. The 4C profiles were predominantly similar between samples, with the exception of a moderate increase of the 4C signals over the enhancers associated with each gene in the tissues in which they are active (for Tfap2c: FB1 and flanking H3K27ac-enriched regions in the brain samples; for Bmp7 mm75 and surrounding H3K27ac-enriched regions in the heart sample). We confirmed the increased interactions of Tfap2c with FB1 and of Bmp7 with mm75 in an independent 3C experiment (S5 Fig.). Importantly, the reciprocal 3C experiment with FB1 as a viewpoint showed that it contacted strongly Tfap2c in the forebrain, but not in the heart, and had much weaker/rarer contacts with Bmp7.
Noteworthy, the Tfap2c domain and the Bmp7 domain end shortly before the edges of the flanking TADs detected in mouse ES cells [34], consistent with the notion that these 4C primary domains corresponded to the structural conformation adopted by the locus. In all samples, the primary contact domains of one gene included the enhancer regions we found associated with it, but excluded the ones associated with the other gene. Nevertheless, we observed a consistent overlap between the two domains, demarcating a region of about 10- to 30-kb region, which we termed the transition zone (TZ). To further characterize this region, we used two additional viewpoints for 4C analysis (Fig. 3D–E). Contacts observed from a viewpoint located just before the centromeric end of the Bmp7 primary domain showed extensive overlap with the latter, extending broadly over Bmp7 but not stopping almost abruptly at the TZ (Fig. 3D). Similarly, FB1 showed only weak contact with positions located on the other side of the TZ (S5C Fig.). This asymmetry in the distribution of contacts suggested the TZ indeed corresponds to a conformational transition between two different conformations. Importantly, a viewpoint located in the TZ itself showed prominent contacts extending towards both genes (Fig. 3E), consistent with the strong 4C signals observed over the TZ in the reciprocal 4C experiments.
Next, we performed 4C analyses on del1 homozygous embryos, where the TZ region was deleted together with a larger part of the locus, including the different enhancers (S6 Fig.). In this context, we observed a wide extension of the contacts made by Tfap2c (resp. Bmp7) in the telomeric (resp. centromeric) region, over distances larger than the size of the deleted region. At the same time, the centromeric (resp. telomeric) profiles remained highly similar between WT and del1. Interestingly, the intervals with frequent contacts by Tfap2c and Bmp7 now largely overlapped, as if they “merged” into one domain only limited by the adjacent TADs (S6 Fig., S3 Table). These new extended contacts supported the notion that the TZ may contribute to delineate two distinct structural domains. However, as del1 also significantly reduced the linear distance between Tfap2c and Bmp7, we decided to use other types of alleles to challenge the structural and regulatory organization of the locus and to test the influence of the TZ on these.
We used insertions carrying loxP sites in the opposite orientation to the one left at the SB-B(3end) position in cis to engineer three balanced inversions by CRE-mediated recombination (Fig. 4A, S1 Table). In INV-L1 and -L2, the distance between Bmp7 and the heart enhancer increased to 5.7 and 1.1 Mb, respectively, whereas the relative order and distances between Tfap2c, the enhancers and the TZ region were unchanged (S7 Fig.). In INV-M, the heart enhancer was now equidistant from Bmp7 and Tfap2c (187 and 207 kb, compared to distances of 80 kb and 312 kb in the wild-type allele, with mm75 taken as reference). However, in this allele, the TZ was now located between Bmp7 and the heart enhancer(s). With each inversion, the LacZ reporter remained adjacent to the heart enhancer region and displayed its normal heart expression (Fig. 4B, S7 Fig.), demonstrating that these rearrangements did not disrupt heart enhancer activity. In the three inversions, Bmp7 expression was strongly reduced in the heart, comparable to levels observed with del1 (Fig. 4C). In contrast, Tfap2c expression was enhanced by a thousand-fold in the heart of INV-M animals (Fig. 4D), implying that in this genomic configuration, the heart enhancers now activated Tfap2c instead of Bmp7. This complete switch of the heart enhancer(s) from Bmp7 to Tfap2c coincided with the new relative position of the TZ. The importance of the position of the TZ was further supported by a lack of up-regulation of Tfap2c in INV-L1 and INV-L2 (Fig. 4D), where its location with regards to the TZ/heart enhancers remained unchanged. In INV-L1, we instead found an up-regulation of Ptgis (Fig. 4E), which was now located on the other side of the TZ, next to mm75. As Ptgis was closer to the heart enhancer (S7A Fig.) we were unable in this case to fully rule out a possible influence of distance on promoter choice. However, in INV-L2, Dok5, the new gene juxtaposed “next to” the heart enhancer(s) opposite to TZ was much further away than Tfap2c (1.1 Mb versus 0.3 Mb). In this context, neither Dok5 (Fig. 4F) nor Tfap2c were up-regulated in the heart, ruling out the possibility that the heart enhancer(s) act simply by default the nearest gene.
To examine at the consequences of these rearrangements on the structural conformation of the region, we performed 4C experiments on INV-M and INV-L2 embryos (Fig. 5, S8–S10 Figs.). In INV-M, as in WT controls, Tfap2c showed robust interactions over a domain extending up to the TZ. Due to the inversion, this domain now included the heart enhancer, which displayed much stronger interaction with Tfap2c than those observed in WT (S8A Fig., pink versus grey arrow), a result consistent with mm75 now activating Tfap2c. Conversely, the new primary interaction domain of Bmp7 stopped at the TZ, with a very reduced 4C signal over the heart enhancer in INV-M when compared to WT (S8D Fig., grey versus pink arrow). The viewpoint located between mm75 and TZ, which was part of the Bmp7 interaction domain in WT, showed in INV-M broad and extended contacts overlapping with the Tfap2c interaction domain, ending at the TZ region (Fig. 5B). Interestingly, the inversion had no effect on the 4C profile of the TZ-associated viewpoint, which extended on both sides in all configurations. Thus, in INV-M as in WT, the locus appeared structurally partitioned at the TZ: instead of maintaining their normal contacts and regulatory preferences, genes and regulatory elements established new interactions, depending on their respective position in relation to the TZ.
In INV-L2 embryos, the 4C profile of Tfap2c appeared generally unchanged and did not expand across the TZ into its new flanking region. The TZ-flanking viewpoint remained still limited by the TZ, but highlighted on the other side a broad domain of nearly 1 Mb in the Dok5-Cbln4 gene desert, which is now adjacent to it. The 4C signal was strongly diminished before reaching the promoter of Dok5, which may explain the lack of up-regulation of this gene in the heart of INV-L2 embryos (Fig. 4F). Again, the TZ itself contacted both flanking regions, the relocated Tfap2c domain, and the new Dok5-Cbln4. Importantly, in INV-L2, Bmp7 showed broad contacts over the region now present at its 3′end, extending for up to 0.5 Mb further in the Cbln4 locus, supporting the notion that the presence of the TZ limited the extent of the Bmp7 contact range (Fig. 5C, S3 Table).
Remarkably, the new distribution of 4C contacts in the different rearrangements appeared to follow quite strictly the relative position of the TZ. It did not appear to depend on the nature of the flanking sequences themselves. The directional bias of contacts made by the viewpoint flanking the TZ is the same in the different configurations (WT, INV-M and INV-L2) (S10 Fig., on the right), irrespectively of the flanking sequences.
The expression and structural changes observed in the heart suggested that the TZ behaved as a simple insulator region. In INV-L1 and INV-L2, the Tfap2c domain was fully maintained and unaffected by the genomic rearrangements. Therefore one would expect little impact on Tfap2c. However, we observed an up-regulation of Tfap2c in the medial telencephalon in both alleles (Fig. 6A–B). This up-regulation is unlikely to be caused by the juxtaposition of new forebrain enhancers, as the regulatory sensor did not detect any forebrain activity in L1 and L2 position, in either the inverted or non-inverted configurations (Fig. 6C). We noted that in INV-L1 and –L2, Bmp7, which is strongly expressed in the medial forebrain, was relocated away from Tfap2c and its forebrain enhancer. This rearrangement had no effect on Bmp7 expression in the forebrain, suggesting that it was the presence of Bmp7 in cis that negatively influenced Tfap2c. Supporting this hypothesis, we did not observe any up-regulation of Tfap2c in the medial forebrain of INV-M embryos (Fig. 6D), where Bmp7 remained adjacent to the Tfap2c. These observations prompted us to re-examine the 4C profiles. As stated before, the intensity of the 4C signals diminished strongly beyond the TZ region. However, we observed that the 4C contacts made by the Bmp7 promoter, albeit weak, were stronger over the Tfap2c domain than over the region located symmetrically from the viewpoint (S9 Fig., green boxes). Reciprocally, Tfap2c showed weak but consistent interactions with the Bmp7 region in WT and INV-M (S9 Fig., blue boxes), interactions which are not observed with a symmetrically located region, or with the region at the equivalent place in INV-L2. To further test if the INV-L1 and –L2 up-regulation of Tfap2c depended on the removal of Bmp7, we produced INV-Bmp7 which consists in a simple inversion of the gene itself. Consequently, Bmp7 remained adjacent to the Tfap2c domain, and separated from it by the TZ (S11A Fig.). In this configuration, we did not observe significant changes of Bmp7 or Tfap2c expression, with the exception of a small reduction of Bmp7 expression in the lateral forebrain. Altogether, these results supported that the simple presence of an active Bmp7 in cis, despite the presence of the TZ region, can affect Tfap2c expression in the medial forebrain.
We also noted that INV-M led to a significant reduction of Tfap2c expression in the lateral forebrain (Fig. 6D), even if the genomic region between Tfap2c and FB1 was unaffected. This reduction could result from the relocation to the other side of the TZ of two forebrain-specific H3K27ac-enriched regions included in INV-M. As we observed neither a concomitant up-regulation of Bmp7 (Fig. 6D) nor changes in the activity reported by the sensor (Fig. 6E), it is possible that these elements may not act autonomously but rather modulate the long-range action of FB1.
We show here that the neighboring Tfap2c and Bmp7 genes are controlled by distinct set of enhancers acting specifically on one or the other gene. Since we observed in a balanced genomic rearrangement a switch of enhancer-gene preferences, the specificity of these enhancers for one or the other gene cannot result exclusively from differences in their promoter structures, as proposed for other situations [49], [60]. In contrast, our results indicate that, for this locus, the regulatory interactions are in a large part determined by the relative position of the different elements, as reported for other complex regions [7], [61], [62].
Our 4C experiments showed that Bmp7 and Tfap2c lie in genomic domains that share limited physical contacts. These domains were only weakly demarcated in the available Hi-C data in ES cells [34]. Therefore, it is unclear if the Tfap2c and Bmp7 domains correspond to adjacent sub-TADs [41], or weak TADs in a rather unstructured region. However, the distinction between these different levels of spatial segregation of the genome may in part be semantic, based on arbitrary thresholds, which may not be pertinent for gene regulation. We showed here that the distinct enhancers that regulate each gene (this work, [53], [63]) reside and act within the corresponding conformational domain, further supporting the functional relevance of the structural partition we described in establishing distinct domains of regulation [15]. Furthermore, we showed that a balanced rearrangement exchanging the relative position of genes, enhancers and the TZ region led to a concomitant redistribution of physical and regulatory interactions. The switch of the heart enhancer from Bmp7 to Tfap2c and the patterns of contacts observed in this configuration demonstrate together that the topological separation in two distinct domains is key to allocate distant enhancers to one or the other gene.
We observed extensive similarities in the 4C profiles between the different cell tissues assayed, irrespective of the expression state of the corresponding genes. This indicates that the Tfap2c-Bmp7 locus adopts a rather generic conformation which undergoes limited changes in response to transcriptional activity. Such a constitutive folding has also been described for other loci [34], [40], [64]–[66]. It suggests that the structural partitioning of the locus into two domains pre-exists and guides regulatory interactions, instead of deriving from directed interactions between active genes and enhancers.
Our functional dissection of the locus highlights that the transition zone separating the two domains has an important role in organizing this topological subdivision. The fusion of the interaction profiles of the two promoters and the centromeric extension of the Bmp7 interaction domain upon removal of the TZ strongly argue in favor of the TZ preventing interactions between Bmp7 and Tfap2c. The different balanced inversions further demonstrate that the TZ organizes this topological separation irrespectively of the nature of its flanking sequences. Interestingly, the TZ region interacts robustly with both flanking regions, suggesting that the topological segregation between Tfap2c and Bmp7 may arise from its action as an interaction sink or decoy, not as a blocker or repulsive element. TAD “boundaries” often displayed strong interactions with regions flanking them on both side [34], suggesting that this behavior could be a rather general feature of topological transitions. The TZ does not appear to coincide with a region of constitutive transcription, contrarily to a large subset of typical TAD boundaries [34]. It is flanked by and includes several constitutive CTCF sites [38]. CTCF sites have been proposed to anchor long-range interactions and to act, together with cohesin and Mediator complexes, as master regulators of the chromosomal 3D conformation [67], [68]. However, as only a subset of CTCF sites act as insulators [15], [69], and as depletion of CTCF only mildly impacts chromosomal topologies [70] and long-range gene regulation [71], the precise role of these sequences – and of other regions of the TZ – would need to be directly assessed.
With regard to the allocation of the heart enhancer, the TZ behave similarly to a classical insulator (Fig. 7). However, the analysis of INV-L1 and –L2 indicates that the TZ does not provide complete shielding from external influences, as the presence, beyond the TZ, of an active Bmp7 promoter can interfere with the expression of Tfap2c in the medial forebrain. Although contacts between Bmp7 and Tfap2c and its associated forebrain enhancer(s) are limited and even insufficient to lead to productive interactions (i.e. activation of Bmp7), they are nonetheless present at higher than background level. Our data suggests that they may be frequent and/or strong enough to perturb the regulation of Tfap2c by its forebrain enhancer(s), most probably through promoter competition. Several studies have reported that promoters have a tendency to come into close proximity [40], [72], [73], particularly when they are co-active and linked. Our analysis indicates that the TZ appears to counteract this generic promoter clustering by limiting admixing of the two domains, but it does not however totally prevent the diffusion of regulatory influences between them. The functional impact of these influences underscores the difficulties of defining functional thresholds for the interaction data obtained with 4C or Hi-C. It also emphasizes that topological domains should not be considered as strict autarchic units: topological separation does not exclude neighborly relationships and semipermeable borders. Transformation of the intrinsically broad forebrain activity of FB1 into the graded expression pattern shown by Tfap2c may involve additional neighboring enhancer elements, as hinted to by the INV-M data. However, our observations suggest that the permeability of the TZ to active Bmp7 may also contribute to this fine-tuning (Fig. 7C). In operational terms, the TZ should be considered as a rheostatic controller rather than as a strict insulator.
Interestingly, a sequence orthologous to FB1 is present between Tfap2c and Bmp7 in the coelacanth, but not in teleosts or sharks (S12 Fig.). This indicates that the origin of FB1 can be traced back to the ancestor of the lobe-finned fishes. In contrast, the sequence of the TZ region is far less conserved, suggesting a more recent origin. Expression of Bmp7 in the forebrain is likely an ancestral feature, as it is shared amongst Bmp7 orthologues and paralogues [44]. Conversely, Tfap2c is the only member of its family expressed in the forebrain [42], [45], and the only one directly adjacent to a Bmp gene. The evolution of FB1 as a forebrain enhancer may have been favoured by the pre-existing expression of Bmp7 in this tissue, as suggested for other loci [74]–[76]. In this scenario, we suggest that Bmp7 may have initially been the primary target of this emerging enhancer. The evolution of a region with insulating-like activity would have make FB1 available to Tfap2c. Interestingly, the forebrain expression of Tfap2c regulates the formation of basal progenitors in the developing cortex in mammals [77] and variations of this expression levels, in space and time, have been proposed to account for the increased number of cortical neurons present in higher primates [77]. Changes in gene expression changes are usually attributed to evolution of enhancers or promoters [78]. Our results indicate that a simple change of the filtering capacity of the TZ may also provide evolution with means of modulating gene expression.
(See S1 Text for details)
The initial allele used to produce SB-B(3end) was obtained by homologous recombination in ES cells (E14). The targeting construct comprised: the SB8 transposon [79]; an additional loxP site outside of the transposon; a neomycin resistance gene under the control of the PGK promoter that are flanked by two FRT sequences. The homology arms (chr2:172686051–172689701 and chr2:172689702–172694528 (NCBI37/mm9)) were amplified by PCR and then attached to the targeting construct above. After transformation and selection in ES cells, correctly targeted clones were injected into donor C57BL/6J blastocyst. Germline transmission was obtained from one chimera. The FRT-flanked selection cassette was then removed by breeding with hACTB-FLPe mice, leaving only the transposon and the loxP sequence outside of it at the site (allele SB-B(3end)). The ES clone BA0758 was obtained from BayGenomics, verified by PCR genotyping, and injected to establish a Tfap2c-gene trap line.
The SB transposon was remobilised and new insertions were mapped as described before [14]. Alleles carrying the different deletions, duplications and inversions were produced by in vivo genomic engineering [18], [54], using the 129S1/Sv-Hprttm1(cre)Mnn/J CRE line [80]. Deletions del1 and del3 were obtained by recombination in cis between the static loxP site at the end of Bmp7 and the one moved along with the transposed insertion SB-A1 and SB-Sall4, respectively. To keep the regulatory sensor at the deletion breakpoint, we also produced another version of these deletions, del1-LacZ and del3-LacZ, by CRE-mediated recombination in trans [54], between the loxP site from SB-B(3end) and the one at SB-A1 and SB-Sall4, respectively. For the del2-lacZ allele, we used a recombination in trans, between SB-B(3end) and BA0758. Mice were genotyped by PCR (see Supplemental Experimental Procedures).
Mouse experiments were conducted in accordance with the principles and guidelines in place at European Molecular Biology Laboratory, as defined and overseen by its Institutional Animal Care and Use Committee, in accordance with the European Convention 18/3/1986 and Directives 86/609/EEC and 2010/63/EU.
LacZ staining and whole-mount in situ hybridization was carried out following standard protocols. For RT-qPCR, total RNA was extracted from the frozen tissues using RNeasy kit (QIAGEN), and then cDNA was synthesized using the ProtoScript II First Strand cDNA Synthesis Kit (New England Biolabs). The quantitative PCR was performed using StepOne Real-Time PCR System with SYBR green reagent (Applied Biosystems). Gapdh was used to normalize expression level for each sample. The extra-embryonic membranes were used for PCR-genotyping of the embryos.
We cloned the FB1 enhancer (chr2:172551998–172555000, NCBI37/mm9) upstream of the reporter gene used in SB8, in a lentiviral vector [81]. The transgenic provirus was produced in HEK293 cells as described elsewhere [81]. Briefly, the virus was micro-injected under the zona pellucida of one-cell embryos which were maintained in culture up to the blastocyst stage. Embryos were then reimplanted into foster mothers and, at stage E11.5 or E12.5, stained for LacZ activity and genotyped.
To prepare the 3C library we dissected out the heart and the lateral and medial forebrains from E11.5 C57BL/6 embryos. The cells were dissociated, fixed and then processed following the protocol in Splinter et al. [82]. The fixed genomic DNA was digested with NlaIII enzyme and subsequently self-ligated. To quantify the ligation products of interest, we conducted qPCR with TaqMan probes. qPCR was performed with four technical replicates, and for each value, mean and standard deviation were plotted.
For the 4C analyses, the 3C libraries were first prepared as described above from the respective tissues with NlaIII enzyme. They were then subjected to digestion by DpnII and ligation. After purification of the circularized DNA, inverse PCR was performed to obtain 4C libraries. Reading primers had 3–6 nucleotides of tag sequence, to allow for demultiplexing of the pooled libraries after sequencing. PCR products were purified, mixed altogether and sequenced on a HiSeq 2000 (Illumina). For data analysis, we first demultiplexed the FASTQ files of the 4C sequencing libraries and then aligned them to the mm9 reference genome using Bowtie version 1.0.0 [83]. To normalize with regard to library size, we divided the counts by the total number of counts on the viewpoint chromosome (chr2) for each library and multiplied these values by 1,000,000 (“RPM normalization”). We then smoothed the counts over adjacent fragments, using a window size of 11 fragments. Details are available in Supplementary Information. Sequencing data of the 4C libraries is deposited at ENA (Study Accession ERP005557)
|
10.1371/journal.ppat.1002189 | Alterations in the Aedes aegypti Transcriptome during Infection with West Nile, Dengue and Yellow Fever Viruses | West Nile (WNV), dengue (DENV) and yellow fever (YFV) viruses are (re)emerging, mosquito-borne flaviviruses that cause human disease and mortality worldwide. Alterations in mosquito gene expression common and unique to individual flaviviral infections are poorly understood. Here, we present a microarray analysis of the Aedes aegypti transcriptome over time during infection with DENV, WNV or YFV. We identified 203 mosquito genes that were ≥5-fold differentially up-regulated (DUR) and 202 genes that were ≥10-fold differentially down-regulated (DDR) during infection with one of the three flaviviruses. Comparative analysis revealed that the expression profile of 20 DUR genes and 15 DDR genes was quite similar between the three flaviviruses on D1 of infection, indicating a potentially conserved transcriptomic signature of flaviviral infection. Bioinformatics analysis revealed changes in expression of genes from diverse cellular processes, including ion binding, transport, metabolic processes and peptidase activity. We also demonstrate that virally-regulated gene expression is tissue-specific. The overexpression of several virally down-regulated genes decreased WNV infection in mosquito cells and Aedes aegypti mosquitoes. Among these, a pupal cuticle protein was shown to bind WNV envelope protein, leading to inhibition of infection in vitro and the prevention of lethal WNV encephalitis in mice. This work provides an extensive list of targets for controlling flaviviral infection in mosquitoes that may also be used to develop broad preventative and therapeutic measures for multiple flaviviruses.
| Dengue (DENV), West Nile (WNV) and Yellow Fever (YFV) viruses are responsible for severe human disease and mortality worldwide. There is no vaccine available for dengue or West Nile virus and no specific antiviral is available for any of these viral infections. These viruses are transmitted to humans through the bite of a mosquito vector. Understanding the effects of viral infection on gene expression in the mosquito is crucial to the development of effective antiviral treatments for mosquitoes and may enable researchers to interrupt the human-insect infection cycle. Here we investigate the alterations in gene expression across the entire Aedes aegypti genome during infection with DENV, YFV and WNV over time. We describe several genes that share a similar expression profile during infection with all three viruses. We also use a WNV mosquito cell, mosquito and mouse model to show that virally downregulated genes are inhibitory to infection when overexpressed and that viral regulation of mosquito genes is tissue-specific. Our results provide an extensive amount of data highlighting viral gene targets in the mosquito during infection. This data may also be used to develop broad-spectrum anti-flaviviral treatments in mosquitoes.
| West Nile (WNV), dengue (DENV) and yellow fever (YFV) viruses are globally important, re-emerging mosquito-borne flaviviruses that cause widespread human disease and mortality [1]. WNV can cause serious illness in man, resulting in encephalitis and death, and is soon expected to be endemic in most of the United States and South America [1], [2]. DENV is among the most important human infectious diseases globally. There are an estimated 100 million cases per year, with over 500,000 cases of potentially fatal dengue hemorrhagic fever [3], [4]. There is no specific treatment for either West Nile or dengue virus, and efforts to create an effective dengue vaccine have been hindered due to safety concerns and potential antibody-dependent enhancement [3], [5]. YFV is endemic to tropical regions of Africa and South America and causes a febrile illness often involving hemorrhagic manifestations with fatality rates up to 50% [6], [7]. There is a YFV vaccine available but it is underutilized in many countries with endemic YFV and no specific antiviral is available [8].
Flaviviruses typically replicate within a mosquito vector for 7–10 days before the vector can transmit virus to humans [1], [5]. Several recent studies have profiled gene expression during the course of flavivirus infection in mosquitoes [9], [10], [11], [12], [13], [14], [15], [16], [17]. Innate immune genes are the focus of many of these investigations, and the Toll, Janus kinase (JAK)-signal transducer and activator of transcription (STAT) pathways have emerged as important anti-flaviviral mechanisms in the mosquito [9], [10]. There is also evidence that DENV actively suppresses mosquito immune responses in vitro [18]. Serine proteases have been shown to be important for both blood digestion and viral propagation, though it is not clear whether they aid or impair viral infection in the mosquito [14], [19]. An RNA interference screen in Drosophila melanogaster cells identified many DENV insect host factors that were shown to be relevant for human cells as well as mosquitoes in vivo [11]. In addition, a recent transcriptomic analysis of Culex quinquefasciatus genes revealed many common and distinct pathogen-response genes to infection with WNV, Wuchereria bancrofti and non-native bacteria, including many genes involved in metabolism and transport [12].
Previous studies investigating gene expression in response to viral infection in both insect and human cells have primarily focused on one flavivirus and/or one gene family. Further investigation is necessary for a global picture of the impact on mosquito gene expression throughout the course of infection with individual flaviviruses. Here, we use comprehensive microarray analysis to identify key alterations in the Ae. aegypti transcriptome during infection with WNV, DENV or YFV. The Ae. aegypti mosquito is a major vector for numerous flaviviruses [20] and is an ideal model for viral infection studies since the genome has been sequenced and characterized [21]. In addition, Ae. aegypti are susceptible to flaviviral infection with WNV in the laboratory and pose a threat for WNV transmission in nature [22], [23]. The flaviviruses used in this study were chosen based on global prevalence and to represent both encephalitic and non-encephalitic flavivirus clades [24]. Analysis was performed on day 1, day 2 and day 7 (D1, D2 and D7, respectively) to observe changes in gene expression at both early and late stages of infection. We report on 405 differentially expressed genes during infection with the three flaviviruses over the three timepoints. This is the first study to our knowledge that compares and contrasts mosquito gene expression in response to multiple flavivirus infection over time.
We used an established protocol for mosquito inoculation to infect the Rockefeller strain of Aedes aegypti mosquitoes with one of three flaviviruses, WNV, DENV or YFV, or mock solution via intrathoracic injection [13]. This method was chosen to positively infect the mosquitoes as well as ensure an even distribution of virus between individual mosquito groups. On D1, D2 and D7, RNA was isolated from the mosquitoes and subjected to genome-wide microarray analysis to determine alterations in gene expression between the mock and infected groups. Analysis was done using a custom X4 NimbleGen array designed against the published Aedes aegypti genome [21]. The experiment schematic is shown in Figure 1A. The complete microarray data set can be found in Table S1 and at the NCBI Gene Expression Omnibus (GEO) #GSE28208. Cut-offs of ≥5-fold or ≥10-fold were applied to identify either differentially up-regulated genes (DURGs) or differentially down-regulated genes (DDRGs), respectively. We identified 405 differentially expressed genes (DEGs) during infection with the three flaviviruses. The highest number of DEGs was observed on D1 of infection with all three viruses and the fewest DEGs were found on D7 (Figure S1). DURG and DDRG expression was confirmed using quantitative (q)RT-PCR, analyzing the expression of 11 highly DURGs and 9 highly DDRGs on D1, D2, D7 and D14 in a separate group of Ae. aegypti mosquitoes infected with WNV and DENV (Figure 2). Table 1 lists these genes with predicted function in the mosquito as well as any homolog identified through a BLAST search using the Drosophila melanogaster genome. Overall, the extent of downregulation of gene transcripts was more dramatic than the upregulation during infection with these flaviviruses. The DDRGs analyzed by (q)RT-PCR showed up to 500-fold decrease in expression during virus infection compared to controls, likely due to the sensitivity of this method of analysis.
For a global perspective on gene regulation during flaviviral infection, hierarchical clustering analysis was done for all DURGs and DDRGs using the Pearson correlation of gene expressions to generate a heatmap (Figure 1B). Detailed heatmaps indicating levels of individual genes can be found in Figures S2A-G. Extensive clustering was found for the three timepoints during YFV infection, on D7 for WNV and DENV infection, and between D1 and D2 for DENV infection. Significant clustering was also evident for all flaviviruses at all timepoints versus mock infection. These results highlight both similarities and differences in gene regulation by individual flaviviral infections. Further analysis revealed 20 DURGs in common between the three viruses on D1 of infection (Figure 1C), including AAEL014440, a putative juvenile hormone-inducible protein. There were only four DURGs in common on D2 of infection with all three flaviviruses, including AAEL004861-RA, a peroxisomal integral membrane protein (Per8p) and three hypothetical proteins. WNV and DENV shared seven DURGs on D2 of infection, all hypothetical conserved proteins. There were no DURGs in common on D7 of infection with all three flaviviruses. YFV had the most unique DURGs, with 54 on D1, 19 on D2 and 86 on D7. We also noted common and unique DDRGs during infection (Figure 1D), with 15 shared between the three infections on D1, including AAEL011045, a pupal cuticle (PC) protein, and AAEL003012, a matrix metalloprotease (MMP), which were significantly down-regulated. The most unique DDRGs were found during DENV infection, with 64 on D1, including AAEL012402, which codes for an elongase protein, and 62 on D2, including AAEL014108, which codes for an aquaporin protein. DENV and YFV shared 11 DDRGs on D1 of infection, the most between any two flaviviruses at any timepoint, including AAEL004897, a brain chitinase, and two cytochrome P450 genes, AAEL012770 and AAEL000340. Looking at overlapping DEGs throughout infection with any given virus, we found considerable variation between timepoints (Figure 3). During DENV infection, no genes were differentially expressed at all timepoints, while YFV had seven DURGs and three DDRGs in common during infection at D1, D2 and D7. There were slightly more DEGs in common between D1 and D2 for each infection, with 8 DURGs and 31 DDRGs shared on D1 and D2 during infection with DENV (Figure 3).
The 405 genes that were differentially expressed during infection with one of the flaviviruses were classified into groups based on biological process (BP), molecular function (MF) and cellular component (CC) (Figure 4, Tables S2 and S3.) Genes that did not have any annotation were excluded from analysis. The largest proportions of DURGs with BP annotation were involved in DNA-dependent transcription regulation (3%) and protein amino acid phosphorylation (2%) (Figure 4A). The most abundant MFs among DURGs were zinc ion binding (9%), nucleic acid binding (5%), nucleotide binding (4%) and transcription factor binding (3%) (Figure 4B). The most abundant BP of the DDRGs were associated with chitin metabolism (12%), transport (7%) and proteolysis (7%) (Figure 3D) and the MFs were related to structural constituent of cuticle (23%) and serine-type endopeptidase activity (10%) (Figure 3E). A similar reduction in chitin-binding proteins and alteration in transport genes was found during infection of Aedes with Sindbis virus [25]. In addition, metabolism was previously found to be altered by DENV infection on day 10 post-infection [10]. Most of the DURGs were found to be intracellular (9%) and nuclear (7%) (Figure 4C) and most of the DDRGs were extracellular (10%) (Figure 4F). Functional clustering of DEGs revealed that the most significant BPs of DURGs were regulation of transcription (26%) and phosphate metabolic process (22%) and MF clustered at ion binding (55%) (Figure 5A and 5B.) The clustering of DURGs placed most of them in intracellular non-membrane-bound organelles at a surprisingly high 57%, with 43% at the plasma membrane (Figure 5C). For the DDRGs, only MF had significant functional clustering, with the majority involved in peptidase activity (31%), chitin binding (24%) and ion binding (24%).
During infection of the mosquito, flaviviruses must disseminate from the midgut (MG) through the body to the salivary gland (SG) and so likely alter gene expression in different organs at various times. To determine tissue-specific expression of the identified DEGs, we infected Ae. aegypti with WNV through blood feeding and dissected the MG, abdomen (AB) and SG on D1, D2, D7 and D14 p.i.. We performed qRT-PCR on select DDRGs at each timepoint to determine levels of gene expression in each tissue (Figure 6). On D1, all DDRGs tested were significantly downregulated in the MG, which is expected as this is where the virus localizes immediately after feeding. Surprisingly, AAEL011375 was also differentially downregulated in the SG on D1 and D2 p.i. when the virus is not expected to be present. It is possible that signalling molecules travel from the infected MG throughout the mosquito early in infection, affecting gene expression in other organs. By D7 p.i., many DDRGs are upregulated in the MG, possibly to compensate for previous downregulation. At D14 p.i., most DDRGs are downregulated in the SG, which may be indicative of the virus disseminating to this organ by this timepoint. An exception to this is AAEL009577, which is slightly upregulated in the SG and downregulated in the MG at D14 p.i. This could be due to a precise role the protein plays in the mosquito during infection that is not related to viral dissemination. The putative MMP (AAEL003012), found to be highly downregulated in the whole mosquito during infection with all three flaviviruses at all timepoints, was significantly underexpressed in the MG on D1 (12-fold) and then highly upregulated in the AB by D14 (19.5-fold) p.i. In fact, most of the genes that were highly downregulated in the whole mosquito during infection with all three flaviviruses are not significantly downregulated, and often slightly upregulated, in the AB at most timepoints tested. These results suggest a complex balance of gene regulation that occurs in the various organs during flaviviral infection.
Six highly DDRGs from different functional groups that shared similar expression profiles during all three flaviviral infections were selected for further characterization, testing the notion that these genes represent a conserved principal in flaviviral infection. We chose to use WNV to characterize these genes as both mosquito cells and live mosquitoes are highly susceptible to infection and a WNV mouse model of infection and disease has been well-established, with infected mice developing encephalitis that leads to death [26]. We hypothesized that virally down-regulated genes may be inhibitory to infection. First, the genes were expressed in Ae. aegypti cells and the susceptibility of the cells to WNV infection was examined using immunofluorescence microscopy with antibodies against WNV envelope protein. A representative image of WNV-infected cells can be found in Figure S3. The overexpression of four previously identified virally-DDRG genes, AAEL001704, AAEL011045, AAEL001022 and AAEL003012, caused a significant reduction in WNV infection of CCL-125 mosquito cells (266.8+/−5.1, 548.75+/−7, 247.28+/−5.9, 284.05+/−6, fold reduction, respectively) compared to GFP alone (Figure 7A). These genes were also shown to inhibit DENV infection of mosquito cells (data not shown). Next, the effect of these proteins on mosquito infection in vivo was investigated. It was previously reported that whole body transfection (WBT) of DNA plasmids into mosquitoes results in high expression of target genes [27]. This method was improved by adding a lipid-based transfection reagent to the protocol, which enhanced expression of transfected genes on D3 and D14 post-WBT (p.WBT) (Figure S4). Ae. aegypti were transfected by intra-thoracic inoculation with DNA plasmids encoding either GFP, AAEL001704, AAEL011045, or AAEL003012 and infected with WNV through blood feeding 6 days p.WBT. At day 10 p.i., the level of WNV in the mosquitoes was measured using qRT-PCR. The expression of each of the three genes significantly lowered WNV infection by approximately one million-fold (Figure 7B). These results strongly support the hypothesis that these proteins play an important role during virus infection of the mosquito.
Since the expression of the selected DDRGs lowered infection both in vitro and in vivo, we hypothesized that interaction of the mosquito proteins with WNV may be responsible for the inhibition. Recombinant protein was generated from two of the DDRGs that highly inhibited infection: AAEL011045, which codes for a pupal cuticle (PC) protein, and AAEL003012, which produces a putative matrix metalloprotease (MMP). We first investigated if the mosquito proteins interact with viral proteins using ELISA assays to detect potential binding of PC and MMP with DENV envelope (E), WNV envelope (E), DENV capsid (C) and WNV NS1 proteins. PC protein bound DENV E, WNV E and WNV NS1 but the MMP protein did not bind any viral protein (Figure 7C). Incubation of PC or MMP with each of the four viral proteins for one hour at 37°C, followed by Western blot analysis using antibodies against the viral proteins, showed altered protein profiles (Figure S5.) This indicates that PC directly binds viral proteins and that both PC and MMP contribute to increased oligomerization of WNV E, DENV E and WNV NS1. Since PC bound WNV E protein, which is found on the surface of the virion and is responsible for receptor-binding and virus entry, it suggested that the interaction may potentially inhibit WNV cell entry and subsequent infection. To investigate this, we evaluated the effects of both PC and MMP protein interactions with WNV on infection in vivo in mice. MMP or PC was incubated with WNV for an hour at 37°C and the solution was injected into mice intraperitoneally (i.p.). Mice that received WNV pre-incubated with PC protein had significantly higher survival rates, with approximately 60% surviving infection, than mice that received either 104 or 105 pfu of WNV, all of which died within 14 days p.i. (Figure 7D). None of the mice that received virus pre-incubated with MMP protein survived WNV infection, even though increased levels of both proteins reduced infection of mosquito cells and live Ae. aegypti. This indicates that PC and MMP likely use different mechanisms to inhibit infection in mosquitoes. The protein binding data, the in vitro WNV infection data, and the in vivo WNV mosquito and mouse infection data together suggested that the mosquito PC protein binds the E protein on the flavivirus surface to directly decrease the potential for infection.
To further investigate the mechanism that the PC protein uses to inhibit viral infection, we looked at the effects of both MMP and PC expression on WNV infection over time in CCL-125 Ae. aegypti mosquito cells. Insect expression plasmids encoding each protein were transfected into mosquito cells along with a plasmid expressing green fluorescent protein. The cells were sorted and infected with WNV. The expression of both proteins impaired the ability of WNV to infect the cells (Figure 7E), though at different times. At 3 hours p.i,, the expression of PC resulted in over ten-fold lower levels of viral RNA than the expression of MMP or in cells with no protein overexpression. The level of viral RNA with PC overexpression was below 10 at 3 hours p.i., while the level with MMP or with no increased protein is approximately 102. This indicates that PC likely affects viral entry and that MMP may not be interfering with this step of infection. By 6 hours p.i, both MMP and PC protein expression resulted in an almost 2-log difference in viral RNA levels when compared to cells with no protein expression. By 24 hours p.i., the cells expressing either MMP or PC were significantly less infected than the control cells, with an almost 4-log difference in viral RNA levels. These data agree with our findings that the MMP protein reduced WNV infection in both mosquito cells and mosquitoes in vivo yet is unable to directly inhibit WNV in solution from infecting mice, while the PC protein was able to prevent WNV from causing significant infection in each of those scenarios. This further indicates that PC may be inhibiting infection from the point of entry by binding to the E protein while MMP may indirectly inhibit infection at a later point.
Discovery of host factors regulated during viral infection of the mosquito may identify conserved protein families and pathways representing both mosquito anti-viral mechanisms as well as requirements for viral life cycles. Our analysis highlights many mosquito genes that are important for infection with three globally (re)emerging flaviviruses, West Nile, dengue and yellow fever. In our studies, we used Aedes aegypti as an in vivo model of flavivirus infection. Though the Ae. aegypti mosquito is a major vector for DENV and YFV transmission in nature, it is generally considered a secondary vector for the transmission of WNV. The major vector for WNV transmission is the Culex mosquito, though WNV has been detected in both Ae. aegypti and Ae. albopictus in nature [28], [29]. In addition, both Aedes and Culex mosquitoes are members of the Culicinae subfamily of mosquito vectors. Comparative analysis revealed that the expression profile of 20 significantly upregulated genes and 15 downregulated genes is quite similar between the three flaviviruses on D1 of infection, indicating a potentially conserved transcriptomic signature of flaviviral infection. Indeed, while Aedes is the major vector for both DENV and YFV and a secondary vector for WNV, we found a similar overlap of gene expression profiles between WNV and DENV as we found for YFV and DENV. This suggests that there are flavivirus-specific alterations in the mosquito transcriptome regardless of which flavivirus infects the mosquito as well as which mosquito is the major vector of the flavivirus used.
One of the genes significantly upregulated during all three infections was juvenile hormone-inducible protein (AAEL014440), which has a homolog in Drosophila melanogaster that is thought to regulate the expression of many other genes [30]. Another gene, core histone H3 (AAEL003685), was over 4-fold upregulated at all timepoints during infection with all three viruses. Several viral proteins target host chromatin and histone proteins to interfere with host gene expression and nucleosome assembly by various mechanisms and for diverse purposes [31]. Herpes simplex virus type 1 (HSV-1) is known to utilize histones for its own genome during lytic infection [32]. The importance of histone proteins in flaviviral infection, and infection in the mosquito vector in general, remains to be investigated. Many genes were found to be differentially regulated in mosquitoes infected with two of the three flaviviruses. For example, during infection with both DENV and WNV, AAEL009750 was over 20-fold downregulated on D1 of infection and significantly lower than mock infection at other timepoints but expression was not significantly altered during YFV infection. This gene codes for a member of the mosquito allergen proteins, which are known to bind human IgE [33] and could interfere with viral transmission to humans if produced in abundance during infection. An interesting point is whether flaviviral infection is altering gene expression directly or indirectly, for example through the enrichment of small interfering RNA molecules. To investigate this, we compared highly downregulated genes from our DENV infected mosquitoes to the recent publication by Hess et al regarding small RNA levels during DENV infection and found several correlations [34]. For example, AAEL010160 was downregulated 18.8, 59.3 and 7.45-fold on days 1, 2 and 7 of DENV infection in our analysis, respectively, and a corresponding sense sRNA was enriched with 2.15 log fold-change on day 2 of DENV infection from the dataset published by Hess et al. Another gene, AAEL001953, was downregulated 11.84, 19.05 and 3.84-fold on days 1, 2 and 7 during DENV infection in our analysis, respectively, and a corresponding sense sRNA was enriched with 3.65 log fold-change on day 2 of DENV infection in the previous study [34]. This indicates that flaviviruses likely alter gene expression through both direct and indirect mechanisms during infection of the mosquito.
The majority of highly altered mosquito transcripts were not canonical innate immune genes though we were able to correlate our analysis with previous data on viral infection and mosquito immunity. Previous studies suggest that depleting PIAS (protein inhibitor of activated STAT), a negative regulator of the Jak-STAT pathway, resulted in down-regulation of five antimicrobial genes (four Cecropin A-like genes and one defensin l-like gene) that were also downregulated by DENV infection [9]. All five of these genes were significantly downregulated in our analysis during infection with all three flaviviruses infections at all three timepoints. One of these genes, AAEL000611, was highly downregulated late in infection, with 15-fold, 27-fold and 38-fold lower expression on D7 of YFV, DENV and WNV, respectively. Another Cecropin A-like gene, AAEL000627, was also highly downregulated on D7 of infection, with expression 14-fold, 14-fold and 41-fold lower during infection with YFV, DENV and WNV, respectively. This indicates that the mosquito Jak-STAT pathway is likely involved throughout infection with all three flaviviruses. This also suggests that YFV, DENV and WNV may have evolved a conserved mechanism to suppress this antiviral pathway during infection. The Toll pathway has also been previously implicated in anti-flaviviral defense by the mosquito [10]. One gene shown to be involved in the mosquito Toll pathway and downregulated by DENV, AAEL001929, was 2.5-fold lower on D1 of infection with YFV and 2.5-fold lower on D2 of DENV infection in our study. Another Toll gene, AAEL003507, was only significantly downregulated during YFV infection, with 2.3-fold and 2.6-fold lower expression on D1 and D7, respectively. Our analysis also found a serine protease gene (AAEL006568) to be downregulated during infection with all three flaviviruses. Previous studies show that some midgut serine proteases limit DENV-2 infection in Aedes aegypti [19]. The Drosophila homolog of this gene, Spatzle-processing enzyme (CG16705), is known to be a positive regulator of the Toll pathway [35]. This implies that serine proteases may contribute to the innate immune response to viruses in mosquitoes.
We also demonstrate that virally-regulated gene expression is tissue-specific. Since flaviviruses must travel from the site of entry and infection, the midgut, throughout the mosquito body before reaching the salivary glands, it was likely that the expression of many genes would be differentially altered in various organs at different timepoints. Genes which are highly upregulated early in infection are likely important for flaviviral colonization of the midgut as well as the start of dissemination out of the midgut. Alternatively, these genes could represent the innate immune response of the mosquito to viral replication in the midgut. We saw several of the flavivirally-down regulated genes in the whole mosquito also downregulated by WNV infection in the midgut on D1 of infection, including AAEL011375, which was 17-fold lower than the mock group. This gene encodes a protein in the trypsin family, and trypsin silencing has been shown to increase DENV infection in Aedes [19]. One gene that was highly downregulated in the whole mosquito in response to infection with all three flaviviruses, AAEL003012, encodes the putative matrix metalloprotease (MMP) protein. This gene was also downregulated in the midgut, salivary gland and abdomen on D1 of WNV infection, suggesting that the protein may play a role in controlling the initial viral infection in the mosquito. Genes downregulated late in infection are possibly inhibitory to establishment of infection in the salivary glands or to transmission of virus to a new host. By D14 p.i, many of the identified virally-downregulated genes have lower expression in the salivary glands, which might be expected as the virus should be concentrated in this organ by this timepoint. Interestingly, this is the only organ in which the MMP protein is still downregulated at D14 p.i., which again indicates that this protein may be involved in controlling viral replication and/or infection.
Bioinformatics analysis revealed significant changes in the expression of genes from diverse cellular processes, including ion binding, ion transport, metabolic processes and peptidase activity. In a previous study investigating gene expression in midguts of WNV-infected Culex mosquitoes on D10 p.i., almost 5% of the highly upregulated genes were related to ion transport. This alteration in ion transport molecules was hypothesized to aid in viral spread through polarized cells by maintaining proper cell polarity and stable solute transport functions [12]. A study on the infection of Aedes with Sindbis virus (SINV) found genes involved in ion transport upregulated D4 p.i. [25]. The same group also saw a decrease in genes related to chitin binding on D1 of SINV infection and our analysis revealed a significant reduction in genes involved with chitin binding and the structural constituent of cuticle. Metabolism and oxidoreductive processes were previously found to be major functional groups altered by DENV infection in Aedes on D10 p.i. [10]. In agreement, our analysis showed that genes involved with oxidation reduction and zinc ion binding were highly upregulated and metabolic process and chitin metabolic process were highly downregulated. We also found that genes encoding serine protease inhibitors were significantly downregulated by infection with all three flaviviruses. In a study investigating mammalian genes important in WNV, it was found that silencing serine peptidase inhibitors significantly increased infection, indicating that a reduction in these proteins favors viral infection [36]. This highlights the likely overlap between mosquito and mammalian flaviviral host factors. We also looked at the cellular location of the DEGs found during infection. In our analysis, the majority of virally-upregulated genes produce intracellular and nuclear proteins and most of the virally-downregulated genes encoded proteins found in the extracellular region.
Several of the proteins encoded by the most significantly virally-downregulated genes are shown to be inhibitory to WNV infection both in mosquito cells and in mosquitoes in vivo. This is direct evidence that viral infection in the mosquito decreased the expression of proteins that are likely to impede viral replication or infection of new cells. Two of these proteins, a pupal cuticle protein and a matrix metalloprotease, were shown to alter the protein profiles of WNV E and NS1. In addition, the PC protein was able to directly bind WNV E protein and inhibit viral infection in mosquito cells as early as 3 hours post-infection. This suggests that PC is acting at the step of viral entry, likely by directly binding the E protein on the virus surface. The MMP protein was also able to inhibit WNV infection in cells, though at a later timepoint than PC. When mixed with live WNV, PC protein enhanced the survival of injected mice, indicating direct action on the virus to impede infection. These results provide strong evidence that the virally-downregulated genes identified in this study likely represent proteins that are inhibitory to flaviviral infection. To further ensure that our findings are relevant for WNV infection in nature, we aligned the pupal cuticle protein from Ae. aegypti with the corresponding Cx. quinquefasciatus protein and found 92% sequence identity. In addition, we performed a BLAST of the Ae. aegypti DDRGs that we use in the WNV studies against the Cx. quinquefasciatus genome and found very high sequence identity (85–95%). This indicates that the genes identified as inhibitory to WNV infection in Aedes mosquitoes are likely also inhibitory to infection in Culex mosquitoes.
This investigation uncovered many previously unknown host factors differentially regulated by flaviviral infection. This is also the first study, to our knowledge, to compare infection with three flaviviruses in the same mosquito at the same timepoints. Understanding the effects of infection on the mosquito, both common and unique to individual flaviviruses, will aid in developing broadly applicable methods to treat and prevent infection.
The CCL-125 Aedes aegypti cell line (ATCC, VA) was used for transfection and infection studies. The cells were grown at 30°C and 5% CO2 in DMEM supplemented with 10% heat-inactivated fetal bovine serum (Gemini, CA), 1% penicillin-streptomycin and 1% tryptose phosphate broth (Sigma, MO). Flavivirus was grown in C6/36 Aedes albopictus cell line using the same media. Strains used were: WNV 2741, DENV-2 New Guinea C, YFV Asibi strain. Cells were infected at an m.o.i. of 1.0, virus was allowed to propagate for 6–8 days, supernatant was removed, spun down and virus stock was stored at −80°C until use.
The Rockefeller strain of Ae. aegypti mosquitoes were either infected by intra-thoracic inoculation or blood-feeding, as indicated in the text and figure legends. For thoracic injections, virus was used at 6.5 logs per mL and 0.5 µL were injected per mosquito. For blood feeding, 100 µL of virus was added to 1 mL serum-inactivated blood from c57L/B6 mice and fed to mosquitoes for 20 minutes at room temperature using a hemotek feeder. Mosquitoes were maintained in groups of 10 at 30°C, 80% humidity. Mosquitoes were supplied raisins as a source of dietary sugar.
RNA was isolated from WNV, DENV type 2 or YFV infected Ae. aegypti mosquitoes on days 1, 2 and 7. RNA was purified using the Rneasy kit (Qiagen, CA) and hybridized with Nimblegen X4 microarray chips using 81-mer probes designed from 18,000 open reading frames (ORF) found in the Ae. aegypti genome, with 2 different probes per ORF. The gene expression data was normalized using quantile normalization [37]. Partek Genomic Suite v. 6.4 (http://www.partek.com) was used for the statistical data analysis and ANOVA was applied to identify differentially expressed genes between each infection versus mock for each timepoint. False discovery rate was used to adjust the p-value for multiple testing corrections [38].
The functional annotation and clustering of DEGs was performed using the DAVID Bioinformatics Resource 6.7 [39], [40]. Briefly, IDs of DDRGs and DURGs were uploaded separately to the DAVID web interface and converted to unique DAVID IDs. The background gene list for the functional clustering consisted of the IDs of all Ae. Aegypti transcripts represented on the Nimblegen X4 microarray chips. The functional annotation and clustering were then performed using the DAVID default parameters.
RNA was isolated from infected Ae. aegypti mosquitoes on days 1, 2, 7 and 14 and purified using RNeasy kit (Qiagen, CA) according to manufacturer's instructions. cDNA was made from the RNA using a SuperscriptIII kit (Invitrogen, CA). cDNA from 1 µg RNA was used in each quantitative (q)RT-PCR reaction along with SYBR green chemistry. Fold change in gene expression was calculated using the CT value differences normalized to actin expression. Oligos can be found in Table S4.
CCL-125 Aedes aegypti cells were infected with WNV at an MOI of 0.1. 24 hours post-infection, cells were fixed in 4% paraformaldehyde for 20 min at RT, washed with PBS(-) and then stained for infection using an antibody against recombinant WNV E protein conjugated with TRITC. The antibody was diluted in 1% BSA at 1/250 and cells were incubated for 20 minutes at RT. Infection was visualized using fluorescent microscopy.
GST-tagged protein was made from two mosquito genes: AAEL011045, which codes for a pupal cuticle protein, and AAEL003012, which produces a putative matrix metalloprotease. Protein was produced in E.coli and batch purified using glutathione sepharose (GS) (GE, NJ) with centrifugation. Briefly, pelleted bacteria from 1L culture were lysed and mixed with 2 mL GS resin with end-over-end mixing for 1 h at RT. The resin was spun down at 500 rpm, washed with PBS(-) and protein was eluted with buffer (50 mM Tris-HCl, 10 mM reduced glutathione, pH 8.0). The GST tags were removed using PreScission Protease enzyme (GE, NJ).
Binding between the Ae. aegypti PC or MMP proteins with flaviviral proteins was investigated using ELISA analysis. Briefly, 5 µg of mosquito or GST control protein was coated onto a 96-well ELISA plate (Thermo Fisher Sci, MA) and incubated overnight at 4°C. The plate was blocked with 1% BSA in PBS(-) and incubated with 1 µg of flaviviral protein (either WNV E, WNV NS1, DENV E or DENV C) or BSA control for an hour at RT. The proteins were washed off, secondary-HRP was added for 30 min at RT, washed off and TMB substrate was added for 20 min at RT. Stop solution was added and the O.D. of the wells read at 450 nm. WNV NS1 protein was a kind gift from Dr. Michael Diamond (Washington University, MO), WNV and DENV E were kind gifts from L2 (CT) and recombinant DENV C was produced in our laboratory.
The Ae. aegypti PC or MMP proteins were incubated with each of four viral proteins (WNV E, WNV NS1, DENV E, DENV C) for one hour at 37°C. The solution was run on a 4-12% SDS-PAGE gel for 1.5 h at 15 milliamps per gel. The proteins were then transferred to nitrocellulose membrane. The nitrocellulose was blocked with 5% milk in 1% TBST for 1 h at RT and then incubated with the appropriate primary antibody overnight at 4°C. The nitrocellulose was washed and then incubated with the appropriate horseradish peroxidase secondary antibody for 1 h at RT. The protein blots were incubated with ECL substrates (Amersham, NJ) for 5 min at RT and then detected on Kodak film. Antibodies used: anti-WNV envelope (L2, CT), anti-dengue envelope (L2, CT), anti-WNV NS1 (gift from Dr. Michael Diamond, Washington University School of Medicine, MO) and mouse immune serum against recombinant DENV-2 capsid protein made in our laboratory.
Nine week old female C57BL/6 mice were infected with WNV (with and without mosquito proteins) intraperitoneally (i.p.) at a dose of 103 plaque forming units (pfu) per mouse. All animal experimental protocols were approved by the Institutional Animal Care and Use Committee of Yale University and experiments were done in a Biosafety Level 3 animal facility according to the regulations of Yale University.
All plasmids were transfected into CCL-125 cells using Effectene (Qiagen, CA) according to manufacturer's instructions. Briefly, for a 10 cm2 plate, 10 µg of DNA was mixed with 500 µL buffer EC and 32 µL enhancer was added. This was allowed to incubate for 5 min on the benchtop. Then, 30 µL Effectene reagent was added and the solution vortexed briefly. After 10 min incubation, the solution was added to the cells. Expression was observed 24 h post-transfection and peaked at 48 h.
The following GENBANK accession numbers were referenced in the manuscript text: AAEL014440, AAEL004861-RA, AAEL011045, AAEL003012, AAEL012402, AAEL014108, AAEL004897, AAEL012770, AAEL000340, AAEL011375, AAEL009577, AAEL001704, AAEL001022, AAEL003685, AAEL009750, AAEL010160, AAEL001953, AAEL000611, AAEL000627, AAEL001929, AAEL003507, AAEL006568, AAEL011375.
Our study was carried out in strict accordance with the recommendations Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All animal experimental protocols were approved by the Institutional Animal Care and Use Committee of Yale University (Protocol Permit Number: 2008-07941) and experiments were done in a Biosafety Level 3 animal facility according to the regulations of Yale University. All efforts were made to minimize suffering.
|
10.1371/journal.ppat.1005917 | Variant Exported Blood-Stage Proteins Encoded by Plasmodium Multigene Families Are Expressed in Liver Stages Where They Are Exported into the Parasitophorous Vacuole | Many variant proteins encoded by Plasmodium-specific multigene families are exported into red blood cells (RBC). P. falciparum-specific variant proteins encoded by the var, stevor and rifin multigene families are exported onto the surface of infected red blood cells (iRBC) and mediate interactions between iRBC and host cells resulting in tissue sequestration and rosetting. However, the precise function of most other Plasmodium multigene families encoding exported proteins is unknown. To understand the role of RBC-exported proteins of rodent malaria parasites (RMP) we analysed the expression and cellular location by fluorescent-tagging of members of the pir, fam-a and fam-b multigene families. Furthermore, we performed phylogenetic analyses of the fam-a and fam-b multigene families, which indicate that both families have a history of functional differentiation unique to RMP. We demonstrate for all three families that expression of family members in iRBC is not mutually exclusive. Most tagged proteins were transported into the iRBC cytoplasm but not onto the iRBC plasma membrane, indicating that they are unlikely to play a direct role in iRBC-host cell interactions. Unexpectedly, most family members are also expressed during the liver stage, where they are transported into the parasitophorous vacuole. This suggests that these protein families promote parasite development in both the liver and blood, either by supporting parasite development within hepatocytes and erythrocytes and/or by manipulating the host immune response. Indeed, in the case of Fam-A, which have a steroidogenic acute regulatory-related lipid transfer (START) domain, we found that several family members can transfer phosphatidylcholine in vitro. These observations indicate that these proteins may transport (host) phosphatidylcholine for membrane synthesis. This is the first demonstration of a biological function of any exported variant protein family of rodent malaria parasites.
| Malaria-parasites invade and multiply in hepatocytes and erythrocytes. The human parasite P. falciparum transports proteins encoded by multigene families onto the surface of erythrocytes, mediating interactions between infected red blood cells (iRBCs) and other host-cells and are thought to play a key role in parasite survival during blood-stage development. The function of other exported Plasmodium protein families remains largely unknown. We provide novel insights into expression and cellular location of proteins encoded by three large multigene families of rodent malaria parasites (Fam-a, Fam-b and PIR). Multiple members of the same family are expressed in a single iRBC, unlike P. falciparum PfEMP1 proteins where individual iRBCs express only a single member. Most proteins we examined are located in the RBC cytoplasm and are not transported onto the iRBC surface membrane, indicating that these proteins are unlikely to mediate interactions between iRBCs and host-cells. Unexpectedly, liver stages also express many of these proteins, where they locate to the vacuole surrounding the parasite inside the hepatocyte. In support of a role of these proteins for parasite growth within their host cells we provide evidence that Fam-A proteins have a role in uptake and transport of (host) phosphatidylcholine for parasite-membrane synthesis.
| Malaria parasites (Plasmodium spp.) invade both liver cells and red blood cells (RBC) in the vertebrate host. They actively remodel the infected RBC (iRBC) by exporting and trafficking various proteins to the RBC cytoplasm and the plasma membrane [1–7]. These Plasmodium proteins are involved in different processes ranging from uptake of nutrients, the formation of membranous structures, protein trafficking and mediating adherence of iRBC to host cell receptors [5]. A large proportion of these exported proteins are encoded by multi-copy gene families and for several of these families evidence exist for their involvement in antigenic variation and immune evasion [6, 8–15]. PfEMP1 proteins encoded by the P. falciparum var gene family are transported to the surface of iRBC, where they mediate interactions of iRBC with host cells. These proteins operate as ligands that bind to host cell receptors on the capillary endothelium resulting in iRBC tissue sequestration, which is believed to prevent iRBC being removed by the spleen [16–20]. Furthermore, the small variant proteins of the P. falciparum stevor and rifin multigene families are involved in iRBC interactions such as rosetting and iRBC tissue sequestration [21–24]. STEVOR-mediated rosetting provides a growth advantage by protecting merozoites from invasion-blocking antibodies [21]. Both sequestration and rosetting of iRBC have been linked to virulence of P. falciparum infections [6, 18, 19, 25–29]. Proteins encoded by the P. falciparum stevor and rifin multigene families are thought to be distantly related to the pir multigene family of rodent and other non-human primate and human malaria parasites [8, 10, 30] and it has been suggested that PIRs of these species also mediate iRBC rosetting as well as iRBC tissue sequestration [21, 31, 32]. However, the molecular determinants mediating the interactions between PIRs and host cells are unknown.
The established functions for several P. falciparum exported proteins of multigene families in interactions between iRBC and host cells (i.e sequestration) or between iRBC and other RBC (i.e rosetting) may indicate that these proteins have mainly or exclusively a role in promoting parasite survival during the erythrocytic part of the life cycle. Most studies on exported proteins, however, have been performed in P. falciparum and the functions of exported proteins encoded by multigene families of other Plasmodium species remain largely unknown. For example, rodent malaria parasites (RMP) have several multigene families that encode variant exported proteins whose functions are unclear [10, 33–35]. In addition to the pir multigene family, RMP contain two large gene families, fam-a and fam-b, that encode exported proteins [33, 35, 36]. The fam-a family is the only multigene family of exported proteins in which a structural domain has been identified that is known from other organisms. This domain has similarity to the steroidogenic acute regulatory-related lipid transfer (START) domain [33, 35], suggesting that these proteins may play a role in the transport of lipids. Different PIR members may fulfil distinct functions as is suggested by the discovery of structurally distinct phylogenetic clades of PIRs represented in multiple species, coupled with differential expression and cellular location of different PIR members during the erythrocytic cycle [8, 9, 32, 33, 35, 37, 38]. PIR proteins are also expressed in blood-stages of Plasmodium species that lack iRBC sequestration or rosetting and in non-sequestering blood-stages such as (young) trophozoites and gametocytes of P. berghei [36]. Moreover, many PIR proteins are not located at the RBC surface membrane [36], suggesting that these proteins have additional roles beyond the promotion of sequestration or rosetting through interactions between iRBC and host cells. Multiple distinct functions have also been proposed for the P. falciparum stevor multigene family [21, 39].
In this study we have performed phylogenetic analyses of the RMP fam-a and fam-b multigene families and analysed expression and cellular localization of members of the pir, fam-a and fam-b families throughout the complete life cycle of P. berghei, including mosquito and pre-erythrocytic stages. To perform the phylogenetic analyses, we first re-sequenced the P. berghei genome utilizing the Pacific Biosciences single-molecule real-time (SMRT) sequencing technology [40], resulting in improved annotation of the fam-a, fam-b and pir families. For all three families we found that multiple proteins of the same family were expressed simultaneously in a single parasite. Most proteins were localized in the iRBC cytoplasm and not transported onto the surface membrane, indicating that these members play no direct role in interactions between iRBC and host cells. Unexpectedly we found that members of all three families that were expressed in iRBC were also expressed in late liver-stages where they are located in the parasitophorous vacuole (PV). For two Fam-a proteins we provide evidence for a location at the PV membrane. Their expression in late liver-stages suggest that these proteins promote parasite development not only in the blood but also in the liver either by supporting parasite development in both the infected hepatocyte and RBC and/or through manipulation of the host immune response. In support of a role in parasite growth within the host cell, we found that several Fam-A members transfer phosphatidylcholine (PC) in vitro. Since host cell PC has been shown to be key for malaria liver stage development [41], these observations indicate that Fam-A proteins may mediate the transfer of PC from the host cell into the parasite for membrane synthesis. This is the first demonstration of a biological function of any exported variant protein family of rodent malaria parasites.
Recently we have reported an improved annotation of the genomes of the RMP P. berghei, P. yoelii and P. chabaudi, permitting the exact determination of the number and genomic location of the pir, fam-a and fam-b family members in the latter two species [33]. However, the presence of large arrays of 2.3kb repeat sequences and low complexity regions inside the subtelomeric regions of P. berghei chromosomes has hampered correct annotation of P. berghei multigene family members. To further improve the P. berghei genome annotation we first assembled the P. berghei genome sequence based on new sequence information obtained with the Pacific Biosciences single-molecule real-time (SMRT) sequencing technology that successfully can resolve long repeat sequences [40]. This resulted in a core P. berghei genome without gaps but with a remaining 11 contigs that could not be attached due to the presence of repeat sequences such as the tandem 2.3 kb repeat sequences. A total of 34, 18 and 44 novel fam-a, fam-b and pir genes, respectively, were annotated (S1 Table) and these were included in the phylogenetic and expression analyses shown below. The new sequences, gene annotation and genomic location are published on the GeneDB website (www.GeneDb.org).
A recent phylogenetic analysis of the pir family of all three RMP demonstrated the presence of robust clades, each characterized by distinct structural motifs and represented in multiple species, which may indicate a long-standing functional diversification among members of this family [33]. In order to determine whether the fam-a and fam-b families can be similarly differentiated into consistent and robust phylotypes, we performed a phylogenetic analysis of these families using approaches similar to those used to generate the RMP pir phylogeny. Most fam-a genes have a subtelomeric location but all three RMP have multiple copies that share a syntenic location in an internal region of chromosome 13 [33, 35]. The fam-a genes comprise of six exons and five introns and encode proteins of approximately 300 amino acids. RMP Fam-a proteins are characterized by the PYST-A domain and most members have a predicted signal peptide but lack a PEXEL and transmembrane domain (www.GeneDB.org; [33]). The maximum likelihood (ML) phylogeny of the RMP fam-a family is shown in Fig 1, where branches with nodes that are supported by bootstrap values >75 are shown in bold. The topology is relatively well resolved and most terminal nodes are robust. Although basal nodes are less robust, significant bootstrap values were recovered for the major clades (black squares in Fig 1). The gene number differs between the RMP with P. berghei ANKA having significantly fewer genes (45) than P. chabaudi chabaudi AS (134) or P. yoelii YM (113)(S1 Table). The distribution of genes from each species in the tree is punctate, i.e. they are not monophyletic but are found throughout the tree indicating the presence of multiple ancestral lineages. These ancestral lineages (a total of 20) are represented by the different clades that contain orthologous genes from the three RMP (orthologous loci are labelled green in Fig 1). In seven cases there has been a secondary loss in one species (small black crosses in the green bars in Fig 1). The putative orthologs from all clades have a conserved location in the RMP genomes, confirming their relatedness. However, in P. chabaudi and P. yoelii most genes are species-specific paralogs (87% in both species), that are more closely related to fam-a genes in the same genome than to fam-a genes in the genome of another RMP, emphasising that considerable gene duplication occurred after RMP speciation.
The RMP fam-a phylogeny can be robustly rooted since primate malarias contain a single copy fam-a gene [33], thereby providing an ideal outgroup. The primate fam-a genes share a syntenic, internal chromosomal location and all homologs cluster robustly together in the tree (Figs 1 and 2A). These out-group fam-a genes lack a signal peptide, whereas in the RMP fam-a genes a predicted signal peptide is linked to the conserved PYST-A domain via a low complexity region of variable length. All three RMP have a cluster of several fam-a genes on chromosome 13 that share a syntenic, internal location with the fam-a genes of primate malarias (www.GeneDB.org; Fig 2B) with which they form a robust clade (Fig 2). Unlike the human primate malaria species, the RMP loci with internal fam-a copies show diversity with respect to the location and number of fam-a genes (Fig 2B), indicating that rearrangements and duplications of fam-a genes occurred after speciation in this internal locus. Although the internal RMP fam-a genes on chromosome 13 cluster together in the phylogenetic tree, they do not cluster tightly with the outgroup (Fig 1). It is therefore unclear from the tree topology which of the (internal) RMP fam-a loci is the ancestral gene from which all others were derived.
In conclusion, our phylogenetic analysis shows that evolution of the fam-a family in RMP species occurred in two phases. Firstly, gene duplication in the common RMP ancestor has created 20 gene loci that maintain orthology across RMP species and are conserved in structure and genomic location. Secondly, subsequent independent expansions have created abundant species-specific clades, mainly in P. chabaudi and P. yoelii. The presence of structurally different and conserved Fam-a proteins may indicate that conserved, functional differences may exist between family members in the three RMP as has been suggested for members of the pir family [9, 33, 38].
In contrast to the phylogenetic trees of fam-a and pir families, the fam-b tree contains only a few branches with nodes that are supported by bootstrap values >75 (even at some terminal nodes; Fig 3). The P. berghei ANKA fam-b family consist of 48 members, whereas P. c. chabaudi AS and P. yoelii YM have 26 and 53 members, respectively (S1 Table). All fam-b genes are located in the subtelomeric regions and comprise two exons and one intron, encoding proteins of approximately 260 amino acids. They are characterized by the presence of the PYST-B domain and most members have a predicted signal peptide, PEXEL motif and transmembrane domain (www.GeneDB.org; [33]. The majority of fam-b genes (above line i in Fig 3) are separated from a small group of atypical sequences that have long branches. Most of these atypical sequences are found at four loci conserved in all three species. Orthology is very rare in the other fam-b genes. Instead, there are multiple, species-specific clades that are paraphyletic with heterospecific genes, suggesting that ancestral lineages have diversified independently after RMP speciation. In short, the phylogeny indicates that each RMP has expanded its fam-b repertoire independently through duplication of only a few (at least two) ancestral lineages. In contrast to the fam-a family, there are no fam-b homologs in non-RMP that could be used as an outgroup. Therefore, it cannot be stated unambiguously that the atypical fam-b genes represent the ancestral RMP genes. However, their positional conservation in RMP and the distinct structure may indicate that these represent ancestral lineages that have not been modified subsequently.
To obtain more insight into the transcriptional activity of members of the RMP multigene families we analysed existing genome-wide RNAseq data of blood-stage parasites [33]. For the P. berghei fam-a, fam-b and pir families we first re-analysed the RNAseq data using the improved annotation described above (see S1 Table for RNAseq FPKM values of all genes). Visualization of the transcript levels of all fam-a and fam-b genes in the phylogenetic trees shows that multiple members of nearly all phylogenetic clades are transcribed in RMP blood-stage populations (Figs 1 and 3). A similar widespread transcriptional activity of multiple members of different clades in blood stages had also been demonstrated for the pir family [33]. For the pir family it is known that significant changes in transcriptional activity of individual members can occur during blood-stage infections [8, 33, 42–44] and that mosquito transmission results in large scale changes of transcription levels of many members [45]. Less is known about variation in transcription levels of the fam-a and fam-b members. We therefore analysed existing RNAseq data of blood-stage populations of two isogenic lines of P. berghei ANKA [33]. In these populations 23 to 40% of the fam-a and fam-b members are transcribed (Fig 4A and 4B). In both lines the percentage of transcribed pir genes is lower, between 17 and 19%. Although the total number of pir genes transcribed in blood stages of the two lines is higher than the number of fam-a and fam-b, the total transcript abundance for the pir genes is lower (Fig 4C). In gametocytes the percentages of transcribed genes of both the fam-a and fam-b family are lower in comparison to asexual blood-stages. Remarkably, the percentage of transcribed pir genes is higher in the (non-sequestering) gametocytes compared to that of asexual blood-stages (Fig 4B).
Both the proportion of transcribed genes and transcript abundance of the families are very similar in in the two P. berghei ANKA lines (Fig 4A and 4B). Since these lines have a different history of blood-stage propagation in the laboratory, these observations suggests the absence of large scale changes in expression of family members during blood-stage propagation. However, when we analyse transcript abundance of individual genes in the two populations, significant differences in transcription of particular genes are observed between these populations (Fig 4D; S1 Fig). Five to nine percent of the pir and fam-a members show a more than 1.5x difference in transcript abundance whereas 21% of the fam-b members show a difference in transcript abundance higher than 1.5x (Fig 4D).
Combined, our analyses of RNAseq transcription data demonstrate transcriptional activity of a relatively high percentage of fam-a and fam-b members (from different phylogenetic clades) in populations of blood-stage parasites. In addition, transcript abundance of individual genes of the three families can differ significantly between blood-stage populations of two isogenic lines, notwithstanding the relatively stable proportions of the total number of transcribed genes and the total transcript abundance.
To further analyse expression at the protein level we generated and analysed a number of transgenic parasites expressing a fluorescently-tagged family member. An important criterion for selection of genes for fluorescent tagging was the availability of existing data for transcription and/or protein expression in P. berghei blood-stage parasites. In Table 1 the 12 selected members for tagging are shown. For all selected genes, except for one (pir1), transcriptome evidence for expression has been reported [33]. In addition, proteome evidence for expression has been reported for all selected fam-a and fam-b proteins and for three out of eight selected pirs (Table 1; [36]).
Transgenic parasites containing a single fluorescently-tagged member were generated by standard genetic modification technologies for P. berghei [46]. Parasites were transfected with linear constructs that introduce the fluorescent tag at the C-terminus of the endogenous gene by integration through single cross-over homologous recombination. The generation of eight of the twelve single-gene tagging (SGT) mutants have been reported previously by Pasini et al. (S2 Table; [36]). In this study we tagged four additional pir genes. For two genes that were tagged before with mCherry, we generated additional mutants that have GFP-tagged versions of the same gene (Table 2). In addition to tagging of multigene family members we tagged two blood-stage exported proteins encoded by single copy genes, i.e. smac [16] and ibis1 [47] (Table 2). The tagging of ibis1 was performed as described for the other SGT mutants. Details of the generation and genotyping of all mutants have been published in the RMgmDB database (www.pberghei.eu; S2 Table).
All SGT mutants showed expression of the fluorescently-tagged protein in blood-stage parasites and these proteins were exported to the RBC cytoplasm (Table 2; Fig 5). For 8 of the 12 proteins export in blood-stages had been shown before. Fluorescence microscopy analysis of live cells revealed that most of these proteins have a punctate/patchy or diffuse localisation pattern within the iRBC cytoplasm. For only one fam-a member (Fam-a1, EMAP1) we observed a fluorescence-staining pattern indicative for a location at the iRBC surface membrane ([36]; Fig 5). The additional four pir genes (PIR4, 5, 7 and 8) that were tagged in this study were also exported into the RBC cytoplasm and showed a patchy or diffuse localisation pattern with the iRBC cytoplasm (Fig 5) as had been observed for all other tagged pirs. The single-copy genes smac and ibis that were tagged were also exported into the RBC cytoplasm with a pathchy/punctate localisation pattern (Figs 5D and S2).In the uncloned blood-stage populations of most mutants more than 5% of the parasites were fluorescent-positive (Table 2). Only in two mutants (with tagged versions of pir6 and pir7) the percentage of fluorescent blood-stages was very low (less than 0.1%). Both pirs belong to the ‘large-form’ pirs [33].
To further analyse expression we first cloned all SGT mutants, except for three mutants that had a low percentage of fluorescent parasites in the parent population (pir4, pir6 and pir7). For all cloned SGT mutants, we were able to obtain one or more clones that produced fluorescent-positive blood-stages. However, for multiple SGT mutants we also selected clones that did not show any fluorescent blood-stages (Table 2), despite having a tagged gene as shown by genotyping through Southern analysis of separated chromosomes (S3 Fig). The selection of fluorescent-negative clones suggests a relatively high rate of ‘expression switching’ of these family members resulting in silencing of expression.
We next determined the percentage of fluorescent blood-stages of the fluorescent-positive clones after a period of 1–2 weeks of asexual multiplication (7–14 asexual cycles) and in blood-stages after passage through mosquitoes. In the blood-stage populations before mosquito passage the percentage of fluorescent parasites ranged between 5 and 99% (Table 2). The presence of significant numbers of non-fluorescent parasites in the clonal populations is in line with a relatively high switching rate of expression and is in agreement with significant differences in transcript levels of individual genes between the blood-stage populations of two isogenic P. berghei ANKA lines (S1 Fig).
Eight of the nine cloned SGT mutants were passaged through mosquitoes. After mosquito passage we observed fluorescent-positive parasites in blood-stage populations of all 8 mutants. The percentage of fluorescent blood-stages after mosquito passage was in most mutants comparable to that before passage (Table 2). Only for the mutant expressing tagged PIR1 we found a strong increase of the percentage of fluorescent parasites after mosquito passage (an increase from 15 to 60–70%; Table 2). These observations indicate that mosquito passage does not result in resetting of expression that would result in large scale differences in expression levels between blood-stages before and after mosquito passage.
To analyse whether significant changes in expression of individual members occur during prolonged blood-stage infections in vivo, we determined the percentage of fluorescent parasites during infections in Brown Norway rats of three cloned mutants (expressing a tagged fam-a, fam-b and pir gene). P. berghei infections in Brown Norway rats show a characteristic course of parasitemia with one or two early peaks with a maximal parasitemia of less than 5%, after which the rats are able to control the infections with occasional small waves of recrudescent parasitemias, ranging between 0.01–0.5% (S4 Fig). During these prolonged infection periods (38–51 asexual multiplication cycles) the fam-a and fam-b mutants showed relatively high and stable percentages of fluorescent parasites. Only in the pir1 mutant did this percentage dropped significantly in both rats (from 30% to lower than 10%; Two-way ANOVA; p = 0.0062;S4 Fig).
The observations of relatively stable percentages of fluorescent parasites before and after mosquito transmission and during prolonged periods of asexual multiplication demonstrate the absence of large changes in expression levels of these members at a population level. However, the observed differences in transcript levels of individual genes in blood-stage populations of two isogenic P. berghei ANKA lines (S1 Fig) and the selection of fluorescent-negative clones of SGT mutants (Table 2) demonstrate that significant changes can occur in expression levels of individual genes.
The expression of most of the tagged proteins in a relatively high percentage of blood-stage parasites suggests that a single parasite may express simultaneously multiple members of the same family. To obtain formal proof for simultaneous expression of multiple members in a single blood-stage parasite, we generated four ‘double gene tagging’ mutants (DGT) that contain two fluorescently-tagged genes from the same family (Table 2). These DGT mutants were generated by transfection of SGT mutants with linear constructs for C-terminal tagging by single cross-over homologous recombination. These constructs contain the hdhfr selectable marker that allows for selection of the DGT with the drug WR99210 [48]. The four DGT mutants contain the following pairs of genes tagged with either mCherry or GFP: fam-a1/fam-a2 (2 independent mutants; RMgm ID 1244 and 1245), fam-b1/fam-b2 (RMgm ID 1246) and pir1/pir3 (RMgm ID 1247; Table 2).
In all mutants we detected blood-stage parasites (trophozoites, schizonts) that expressed and exported the two tagged proteins simultaneously (Fig 6). For the fam-a and fam-b DGT mutants, 20–90% of the parasites expressed both proteins simultaneously. For the pir DGT mutant this percentage ranged between 1 and 3%. The two tagged fam-a members had a different cellular location in a single parasite. In both independent fam-a1/fam-a2 DGT mutants, Fam-a1 is located at the RBC surface membrane whereas Fam-a2 locates in the RBC cytoplasm. The same difference in localisation we had also observed in blood-stages of the SGT mutants that expressed either Fam-a1 or Fam-a2 (Fig 5). The simultaneous expression of multiple family members in a single parasite demonstrates that expression of members of these families is not mutually exclusive as has been shown for the P. falciparum var gene family [49–51].
We next analysed the expression of several tagged proteins throughout development in the mosquito and during development in the liver. Parasites of both SGT and DGT mutants were fed to Anopheles stephensi mosquitoes and oocysts (day 8 to 12) and salivary gland sporozoites were analysed for expression of the tagged proteins by fluorescence microscopy. In all mutants analysed, six SGT (Fam-a1, Fam-a2, Fam-b1, Fam-b2, PIR1, PIR8) and the four DGT mutants, we did not observe any fluorescence-positive oocysts or sporozoites. After mosquito passage, all SGT and DGT mutants expressed the tagged proteins in blood-stage populations (Table 2) demonstrating that the absence of expression in oocysts and sporozoites is only a temporarily switch-off of expression during development in the mosquito.
During the first 30 hours of liver-stage development in cultured Huh7 cells no fluorescence could be detected in all SGT and DGT mutants. However, between 40 and 48 hour after infection of hepatocytes, we observed expression of Fam-a, Fam-b and PIR proteins (Fig 7). In five out of six SGT mutants we observed expression in maturing liver-schizonts (Table 2, Fig 7). Also liver-stages of the DGT mutants were fluorescent-positive (Fig 8A–8C) and for one DGT mutant, fam-a1/fam-a2, we observed simultaneous expression of both members in 40–45% of the liver-schizonts by live imaging of infected hepatocytes (Fig 8A). In addition, we detected simultaneous expression of the two members in the fam-b1/fam-b2 DGT by immunofluorescence analysis of fixed liver-stages using anti-GFP and anti-mCherry antibodies (Fig 8D; Table 2). As observed for the blood stage, multiple family members are simultaneously expressed in a single liver-stage parasite, indicating that also at this stage of development there is an absence of mutually exclusive expression of members of these families. In the pir1/pir3 DGT we only observed expression of PIR1::mCherry, both by live imaging of infected hepatocytes and by immunofluorescence analysis of fixed parasites using both antibodies.
All tagged proteins were exported into the parasitophorous vacuole (PV), except for PIR8 which was located in the cytoplasm of the parasite. We did not detect fluorescence signals in the cytoplasm of the host cell, both by live imaging and by staining of fixed cells with anti-mCherry or GFP antibodies. The single-copy genes smac and ibis were also expressed in liver-stages and both proteins were located in the PV. For IBIS a PV location in infected hepatocytes had been shown before [47].
With standard fluorescence microscopy it is difficult to determine whether the proteins are located in the lumen of the PV or are located on/at the PV membrane (PVM). For two Fam-a members we analysed the location in more detail using confocal microscopy. In these analyses we stained the parasites with anti-mCherry recognizing the tagged fam-a proteins and with antibodies against two known PVM-resident proteins (EXP1 and UIS4). A clear overlap was observed between the staining of the PVM-resident proteins and the anti-mCherry staining location indicating that these members are indeed located on/at the PVM (S5 Fig). Follow up studies are required to more clearly define the location of the different family members in the PV/PVM and also to explore the possibility that members of the same family may be present either at the PV or the PVM, much like fam-a members are able to locate to different locations in the iRBC (i.e. RBC cytoplasm or RBC plasma membrane).
The expression of variant exported proteins in liver-stages indicates that these proteins promote parasite development not only in the blood but also in the liver either by supporting intracellular parasite development, or through the manipulation of the host immune response. By sequence analysis both PIR and Fam-B proteins do not contain domains with homology to known functional protein domains from other eukaryotes, including other Apicomplexan parasites, which would reveal their function in parasite growth inside RBC and hepatocytes. In contrast, Fam-A proteins are the only Plasmodium variant exported proteins that contain a domain with homology to a functional domain of proteins of other eukaryotes. This domain, the steroidogenic acute regulatory-related lipid transfer (START) domain, is involved in translocation of phospholipids, ceramide or fatty acids between membranes. The START domains represent the majority of these fam-A protein sequences. For example the START domain comprises 83% of the Fam-a protein PBANKA_1327251 and 96% of the FAM-a protein PCHAS_1331900. In Fig 9 a schematic representation of the predicted tertiary structure of the START domain of Fam-a2 is shown, threaded against the resolved holotypic structure of the STAR-D2 domain in the human phosphatidylcholine transfer protein. In the Plasmodium START domains an expanded protein loop between β-sheets 7 and 8 is present (Fig 9).
As Fam-A proteins are expressed in liver stages, where the parasite requires host-derived phosphatidylcholine (PC) [41], we tested the ability of recombinant Fam-A proteins from P. berghei and P. chabaudi (and the single P. falciparum Fam-A) to transfer PC using a standard in vitro PC assay. In this assay protein-dependent transfer of radioactive PC from a small population of donor vesicles to a larger population of acceptor vesicles is measured. To capture potential differences between different Fam-A orthologues, we choose four P. chabaudi orthologues that were distally distributed on the Fam-A phylogenetic tree (see above), two P. berghei orthologues and the single P. falciparum Fam-A protein (Fig 9D). We expressed these as recombinant proteins containing a single hexahistidine tag at their N-terminus, which were purified over Ni2+ affinity resin and further purified by size exclusion chromatography. In the in vitro PC assay one of the four recombinant Fam-A proteins from P. chabaudi (PCHAS_1331900) and one of the two recombinant P. berghei proteins (PBANKA_1327251) showed robust PC transfer activity, whereas P. falciparum Fam-A reproducibly showed a lower, but detectable, level of PC transfer activity (Fig 9D). The activity level of the P. chabaudi and P. berghei proteins was identical to that of the previously described P. falciparum START-domain containing protein PFA0210c (PF3D7_0104200) and its orthologues in Plasmodium knowlesi and P. chabaudi [52]. The five other P. chabaudi and P. berghei Fam-A proteins did not show PC transfer activity, with activity levels comparable to the no protein control (Fig 9D). The activity of these Fam-A proteins (or lack thereof) was detected in multiple assays and was independent of the affinity tag as addition of an MBP tag (at the N terminus). In addition we have expressed and purified Fam-a protein PBANKA_132751 in which we have deleted the final C-terminal alpha helix of the START domain. This mutation is known to abolish the activity of START-domain-containing proteins [53] and used this mutated recombined protein to test PC transfer activity in vitro (Fig 9E). We found that PC transfer activity is significantly reduced of the mutated protein compared to the PC transfer activity of the full length protein, confirming the functionality of the START domain. It has previously been speculated that Fam-A proteins may be cholesterol transfer proteins based on the structural resemblance of the Fam-A proteins with MLN64, a human cholesterol transfer protein [35]. We tested the ability of two Fam-A proteins, PCHAS_120120 (which does not transfer PC), and PCHAS_1331900 (which does transfer PC), to transfer cholesterol using a standard cholesterol-binding assay. No cholesterol binding activity was detected for either protein (S6 Fig). Combined our results indicate that at least one P. chabaudi and one P. berghei Fam-A protein are phospholipid transfer proteins, whereas no evidence was found for cholesterol transfer activity.
Most studies of Plasmodium exported proteins encoded by multigene families have focussed on the P. falciparum-specific var family [1–6, 10, 54]. This family encodes variant proteins that are exposed on the outside of the iRBC surface membrane and have an essential role in sequestration of iRBC, as they bind directly to certain endothelial cell receptors [6, 19, 55–58]. Also for proteins of two other P. falciparum-specific multigene families, stevor and rifin, evidence for an iRBC surface membrane location has been reported and both are believed to mediate interactions between iRBC and uninfected RBC, resulting in rosetting [21, 23, 24, 59]. With the exception of Plasmodium species of the subgenus Laverania (such has P. falciparum and P. reichnowi [60]), other mammalian Plasmodium species lack genes with clear orthology to either the var, stevor or rifin multigene families. However, many other Plasmodium species contain a large multigene family encoding PIR proteins that are exported into the iRBC. These proteins have structural similarities with domains of P. falciparum RIFIN proteins [8, 21, 30, 35] and it has been suggested that PIRs are adhesins that enable parasites to rosette and sequester in the absence of the var, stevor or rifin families [21, 32]. For P. vivax PIR proteins some evidence has been presented for a role in iRBC sequestration [9, 31] and a subset of recombinant P. chabaudi PIRs showed binding to mouse RBC suggesting a role for these proteins in rosette formation and/or invasion [32]. PIR proteins are, however, also expressed in Plasmodium species that lack sequestration of iRBC, such as P. yoelii, and PIRs are abundantly expressed in RMP blood stages that do not sequester, for example (young) trophozoites and gametocytes of P. berghei [33, 36]. These observations suggest that Plasmodium PIRs are not exclusively involved in iRBC sequestration and/or rosetting. Our observations on the cytosolic location of most tagged family members and their absence from the outer host-cell surface membrane also indicate that these members are not involved in direct interactions between iRBC and host cells that lead to sequestration or rosetting.
Several features of the RMP multigene families suggest that different members of these families may fulfil different functions during blood stage development. The presence of structurally different phylogenetic clades that are conserved between different RMP suggest functional diversification among family members ([8, 33] and this study). In addition, differences in timing of expression and cellular location between family members may indicate functional diversification ([33, 36] and this study). However, it is possible that the different members share the same function but do so at a different cellular location or at a different time point during development. Therefore, the conserved, structural differences between family members might not be related to separate functions but rather linked to differences in stage-specific expression, trafficking and location.
The expression of multigene family members in liver-stages also calls into question their exclusive role in iRBC sequestration and rosetting or in other processes related to remodelling of the iRBC. Although we have tagged only a limited number of members of the three families, we have found so far no evidence that members of different phylogenetic clades are specifically expressed either in liver- or in blood-stages. Indeed seven out of eight family members that were expressed in blood-stages were also expressed in liver-stages.
In the absence of a known function for members of the three families in iRBC we can only speculate on their role in liver-stages. For example, these proteins may facilitate an efficient formation of mature schizonts and daughter merozoites in the liver (see below) or they could help establish and/or promote a subsequent blood infection just after the infected hepatocyte ruptures and discharges the protein contents of its PV in the blood in addition to the release of merozoites.
As many of the multigene family members appear not to be located at the outer side of the iRBC surface membrane, the immune pressure on these variant proteins is likely not due to their detection on the surface of iRBC and removal of iRBC by macrophages in the spleen as has been suggested for iRBC surface-exposed proteins [18–20]. Interaction with the immune system will mainly occur when these protein complexes are released into the blood after host cell rupture. It has been shown that changes in PIR repertoire expression correlates with differences in virulence of P. chabaudi. Blood stage populations with a limited PIR expression were more virulent than blood stage populations with a larger repertoire of expressed PIRs. Mosquito transmission promoted a generalized increase in PIR expression in the parasite population that infected RBCs [45]. These observations may indicate that PIR expression influences immune responses, which may affect parasite survival and virulence. Since the liver has a unique local immune system with systemic impact [61], it is conceivable that the release of parasite proteins after rupture of the PVM and subsequent merozoite release impact on innate and adaptive immune responses that later may influence blood-stage infections.
All examined proteins were exported into the PV of the liver-stages, apart from one PIR protein. However, unlike in blood-stages, we did not find evidence for transport of these proteins across the PVM, suggesting that these proteins are not involved in either remodelling the host hepatocyte or transport parasite or hepatocyte factors through the hepatocyte cytoplasm. For two Fam-a members we provide evidence for a location on/at the PVM. In the PVM of blood-stages the translocon of exported proteins (PTEX) is implicated in the translocation of parasite proteins from the PV into the iRBC cytoplasm [3, 48, 62]. Interestingly, two recent studies provide evidence that a putative PTEX in the PVM of liver stages is different from those in the PVM of blood stages [63, 64]. Specifically, one of the main components, HSP101, is not expressed in liver-stages. HSP101 is a member of the ClpA/B chaperone family and is proposed to unfold cargo proteins [65]. Thus a different PTEX composition (or the absence of a functional translocon) might be responsible for the retention of blood-stage exported proteins in the PV of liver stages [63, 64]. However, it remains possible that the members of the multigene families are exported into the hepatocyte cytoplasm but below the detection limit or that the fluorescent-tag hampered the efficient passage of proteins through the PVM. This latter explanation is less likely, since most tagged proteins were able to cross the PVM of blood-stages ([36]; this study).
For several family members we found that the mCherry-tagged protein was expressed in liver stages, whereas no or weaker expression was detected when the same member was tagged with GFP. This may suggest that the GFP-tag hampered expression or transport of the protein to the correct location, resulting in protein degradation or may be explained by differences in specific properties of GFP and mCherry that influence fluorescence intensity, i.e. the higher pH sensitivity of GFP compared to mCherry [66–68]. However, the lack of expression of the GFP-tagged version might also be due to chance, i.e. that the encoding gene had not been switched on in the liver on this occasion as the expression switching rate of family members is relatively frequent. This is supported by the observation of the presence of both the GFP- and the mCherry-tagged version of the IBIS protein in the PV of liver stages, a protein encoded by a single copy gene that is stably expressed in all parasites.
In blood-stages the IBIS protein is found in discrete membranous structures in the iRBC cytoplasm that exhibits characteristics of P. falciparum Maurer’s clefts and its location in the PV of liver-stages had previously been demonstrated [47]. The SMAC protein, the other tagged protein that is encoded by a single copy gene, is also located in the iRBC cytoplasm and has been implicated in the transport of proteins that facilitate CD36-mediated sequestration of P. berghei schizonts [16]. It is therefore possible that IBIS and SMAC are both components of larger protein complexes in the iRBC cytoplasm, protein complexes which may also interact with members of the three multigene families. Such protein complexes may fulfil similar functions in the iRBC cytoplasm and in the PV of liver-stages. For example, these complexes may be involved in transport of factors essential for schizont maturation and merozoite formation.
Both PIR and Fam-B proteins do not contain domains that have homology to functional domains of proteins from other eukaryotes including proteins of other apicomplexan parasites that would point to a function for parasite growth within RBCs or hepatocytes. In contrast, Fam-A proteins contain a START (StAR-related lipid-transfer) domain with homology to a functional domain of StAR (Steroidogenic Acute Regulatory) proteins in other eukaryotes, so named based on their capacity to transport a large variety of lipids and sterols between membranes [35, 69–71]. The START domain is a module of about 210 residues, which binds lipids such as glycerolipids, sphingolipids and sterols. We show that several Fam-A proteins have robust phosphatidylcholine (PC) transfer activity in vitro. This activity was in the same range as previously described for the exported P. falciparum phospholipid transfer protein PF3D7_0104200 that also contains a START domain [52]. Such PC transfer activity is in support of a role of these proteins in parasite development within the host cell. Uptake of host cell PC has recently been shown to be key for malaria liver stage development [41] and the presence of Fam-A proteins with PC transfer activity in the PV may indicate that they are involved in uptake and transfer of PC from the hepatocyte. PC is also a main component of membranes of Plasmodium blood stages and the expression of Fam-A members in the cytoplasm of iRBC may further designate their role in uptake of host PC. However, these proteins may also be involved in transfer of parasite-derived phosphatidylcholine into the host cell for the assembly of membrane-bound compartments in iRBC, as has been suggested for the P. falciparum START-domain containing protein PF3D7_0104200 [52]. Plasmodium parasites have different enzymatic pathways for the de novo-synthesis of PC [72]. Interestingly, rodent and non-rodent Plasmodium species differ in their phospholipid metabolic pathways [72, 73]. Non-rodent Plasmodium species have an additional PC synthesis pathway, a plant-like pathway that relies on serine decarboxylase and phosphoethanolamine N-methyltransferase activities, which diverts host serine to provide additional PC and phosphatidylethanolamine to the parasite. The absence of this pathway in rodent malaria parasites may explain why the rodent parasites have an expanded family of START-domain containing proteins compared to non-rodent Plasmodium species. In the absence of this de novo synthesis pathway rodent Plasmodium species may be (more) reliant on uptake of host PC for synthesis of their membranes during growth and multiplication. It is also conceivable that rodent parasites owe their very short developmental cycles (24 and 48 hour cycle of blood- and liver-stages, respectively) to a higher and more efficient uptake of host lipids for membrane synthesis during intracellular growth. It is interesting to note that Fam-A members were not only localized in the iRBC cytoplasm but also at the surface membrane which may suggest that these proteins may be involved in uptake of host lipids from the serum and subsequent transfer into the parasite.
It is known that also hepatocyte-derived cholesterol is transported to the parasitophorous vacuole of liver stages [74, 75] and that blood stages scavenge host cholesterol since Plasmodium parasites cannot synthesize cholesterol [76, 77]. It has been speculated that the Fam-A proteins may be cholesterol transfer proteins based on the structural resemblance of the Fam-A proteins with MLN64, a cholesterol transfer protein [35]. We found no evidence of cholesterol transfer activity of the few Fam-A proteins we have tested. It is however possible that other Fam-A members play a role in transport in lipids other than PC, including cholesterol. In other eukaryotes START domain proteins have been divided into distinct subfamilies, each subfamily being more specialized in the transport and/or sensing of specific lipid ligand species and START domain containing proteins act in a variety of distinct physiological processes, such as lipid transfer between intracellular compartments, lipid metabolism and modulation of signaling events [70, 71].
Combined, our observations indicate that (a subset of) Fam-A proteins are involved in the transfer of PC. Transfer of PC may either be from the host cell into the parasite for membrane synthesis during intracellular growth or may be of parasite-derived PC onto the PV/PVM and into the host cell cytoplasm to form trafficking networks contiguous with the PV/PVM. This PC transfer activity is the first demonstration of a biological function of any exported variant protein family of rodent malaria parasites and further exploration of capacity of different members of Fam-A family to transfer specific/different lipid moieties transfer is required to better understand the role of host nutrient uptake and intracellular parasite development. Our observations on liver stage expression expand the role for these proteins beyond their role in iRBC interactions with host cells, such as sequestration or rosetting. Our observations of proteins of RMP multigene families may also be of relevance for understanding the role of multigene families, other than the var family, in human malaria. PIR proteins are also expressed by P. vivax and P. knowlesi and several features of PIRs are shared with P. falciparum RIFIN and STEVOR proteins [10, 21].
Using P. berghei ANKA (cl15cy1) genomic DNA we generated a C2P4 library of fragments with a mean length of 8kb. By Pacific Biosciences sequencing 20 SMRT cells were generated (Accession number ERS531640), which were assembled with the HGAP assembler using the default parameter [40]. The resulting 61 contigs were ordered using ABACAS (Algorithm-Based Automatic Contiguation of Assembled Sequences; [78]) against the current P. berghei ANKA (cl15cy1) reference genome (version2; [33]). Contigs that contained mouse DNA contamination or contained incorrect assemblies (for example chimeras) were excluded. This resulted in the assembly of 14 chromosomes, two plastid genomes and a remaining 11 unassigned contigs in the bin). Contigs were corrected for frameshift with ICORN (Iterative Correction Of Reference Nucleotides; [79]). The improved sequence was annotated through RATT (Rapid Annotation Transfer Tool; [80]) and ab initio gene finding [81] and combined as described in [33]. The annotation was further manually improved.
Fam-a: All homologous sequences were extracted from PlasmoDB (version 12.0) after BLAST analysis and from GeneDB for the re-annotated sequences. Initial alignment of coding sequences established that structural variation was relatively low and therefore, to maximise resolution, the phylogenetic analysis would be carried out on nucleotide sequences containing both exons and introns. Multiple sequence alignment was carried out with ClustalW [82] within the BioEedit suite [83], and then manually adjusted. The final alignment contains 1328 characters; due to substantial length variation at the 5’ end of the genes, the first exon, first intron and a portion of the second exon were omitted because they do not align. The data set includes 313 taxa; nine P. berghei sequences and 22 P. yoelli 17X sequences were omitted because they failed to align satisfactorily. Phylogenetic analysis was carried using two methods; maximum likelihood (ML) using RAXML v8.0 [84] and Bayesian Inference (BI) using MrBayes v3.2.3 [85]. In both cases, a general time-reversible model with a correction for rate heterogeneity (GTR+Γ) was applied. Node robustness was assessed in the ML analysis through 100 non-parametric bootstraps. The Bayesian analysis was carried out with four independent Monte Carlo Markov Chain (MCMC) chains, each running for 5 million generations, with a sampling frequency of 1000 generations and a burn-in of 10%. The potential scale reduction factor (PSRF) for tree length approached one (1.000074), indicating that the MCMC analysis reached convergence. Analysis with Tracer [86] confirmed that the effective sample sizes for both log-Likelihood (ESS = 1128) and tree length (ESS = 18630) were sufficient for convergence. To assess the effect of dynamic differences in substitution rate across the genome, a second, partitioned Bayesian analysis was carried out on the same alignment in which each exon and intron was independently modelled.
Fam-b: All homologous sequences were extracted from PlasmoDB (version 12.0) and from GeneDB for the re-annotated sequences and aligned as described above. Initial alignment again established that, to maximise resolution, the phylogenetic analysis should be carried out on nucleotide sequences containing both exons and the single intron. The final alignment contains 1082 characters. The data set includes 120 taxa; nine P. berghei sequences, two P. chabaudi sequences and 13 P. yoelli 17X sequences were omitted because they failed to align satisfactorily. ML and BI phylogenetic analyses were carried out as described above. In the Bayesian analysis without partitioning, the MCMC analysis failed to reach convergence, (PSRF for tree length = 1.028711); effective sample sizes for both log-Likelihood (ESS = 23) and tree length (ESS = 89) were substantially less than 100. A second, partitioned Bayesian analysis also failed to reach convergence (PSRF = 1.080465; log-Likelihood ESS = 4; tree length ESS = 18). We attempted to produce convergence in the MCMC runs by constraining each with a starting tree topology estimated using maximum parsimony in MEGA v6 [87]. However, this analysis also failed to achieve convergence in log-Likelihood after 5 million generations (PSRF = 1.005; log-Likelihood ESS = 20; tree length ESS = 1865). Hence, it seems ambiguity in these data make the estimate inherently unstable. In light of this, our BI estimate is based on a single MCMC run from the MP-primed analysis that did converge (log-Likelihood ESS = 731; tree length ESS = 3154).
RNA-seq data: RNA-seq data of the three multigene families from RMP blood-stages was obtained from Otto et al. [33]. For P. berghei fam-a, fam-b and pir families the RNAseq reads were remapped to the improved P. berghei genome as described [33]. Transcript abundance is expressed in FPKM (fragments per kilo base of exon per million fragments mapped). FPKM cut-off values were calculated for each RMP as described [33]. Expression evidence was defined as FPKM values >21 for P. berghei, >11 for P. chabaudi and >11 for P. yoelii. We defined the following classes for the level of expression for the fam-a and fam-b genes: class 1: less than twice the cut-off values; class 2: between 2 and 4 times the cut-off levels; class 3: between 4 and 8 times the cut-off levels; class 4: >8 times the cut-off levels. Due to the improved P. berghei genome sequence (see above), several collapsed repeats were now separated in the assembly, resulting in base perfect duplicated copies of the fam-a, fam-b and pir genes. As default, the expression level (FPKM value) of these duplicated genes is zero. We estimated the amount of unique (non-repetitive) base pairs of all genes of one family by blasting all genes against themselves and subtracting from the top blast hit, the length (bp) of the overlap of the total gene length (bp) of the gene (only for those genes that show an identity >98.5%).
Female Swiss OF1 mice (6–8 weeks; Charles River, F) and female Wistar and Brown Norway rats (7 weeks, Charles River, F) were used. Three reference ‘wild type’ P. berghei ANKA parasite lines were used: i) line cl15cy1 (ANKAwt) [46] ii) reporter line 1037cl1 (ANKA-GFP-Lucschiz; mutant RMgm-32; www.pberghei.eu) which contains the fusion gene gfp-luc gene under control of the schizont-specific ama1 promoter integrated into the silent 230p gene locus (PBANKA_030600) and does not contain a drug-selectable marker [79] and iii) reporter line 676m1cl1 (ANKA-GFP-Luccon; mutant RMgm-29; www.pberghei.eu) which contains the fusion gene gfp-luc gene under control of the constitutive eef1α promoter integrated into the silent 230p gene locus (PBANKA_030600) and does not contain a drug-selectable marker [88].
Ethics statement: All animal experiments of this study were approved by the Animal Experiments Committee of the Leiden University Medical Center (DEC 07171; DEC 10099; DEC12042). The Dutch Experiments on Animal Act is established under European guidelines (EU directive no. 86/609/EEC regarding the Protection of Animals used for Experimental and Other Scientific Purposes).
To generate transgenic parasites expressing C-terminally tagged mCherry proteins, construct pL1419 was used [16]. The smac targeting region was then replaced by a targeting region of the candidate genes listed in supplemental S2 Table. For pL2067 and pl2069 (double cross-over recombination plasmids), a second targeting region was introduced (HindIII/ApaI). Construct pL1817 was used to generate transgenic parasites expressing C-terminally tagged GFP proteins [36]. To create C-terminally tagged fluorescent plasmid that contain the human dhfr/ts selectable marker, the T. gondii dihydrofolate reductase/thymidylate synthase (dhfr/ts) selectable cassette was exchanged for the h dhfr/ts selectable cassette of pL0006 (HindIII/EcoRV; MR4, http://www.mr4.org). Details of the primers, DNA constructs and the genotype analysis of all mutants have been submitted to the database of genetically modified rodent malaria parasites (RMgmDB, www.pberghei.eu). For the generation of parasite expressing two C-terminally tagged fluorescent proteins (mCherry and GFP), parasites were first transfected with a DNA construct that contains the dhfr/ts and targets the first gene. These transfected parasites were selected with pyrimethamine. Subsequently these transgenic parasites were transfected with a DNA construct that contains the human dhfr/ts selectable marker and targets the second gene. These transfected parasites were selected with pyrimethamine WR99210 [48]. Transfection of P. berghei parasites with linearized plasmids, selection and cloning of transgenic and mutant parasite lines was performed as described [46]. Correct integration of the DNA constructs was determined by diagnostic PCR and Southern analysis of chromosomes separated by pulse-field gel (PFG) electrophoresis. Southern blots were hybridized with the following probes: 3’UTR dhfr/ts of P. berghei ANKA and a mixed probe human dhfr [89]; ~800bp fragment of 5’UTR of PBANKA_0508000 located on chromosome 5 (primer set: 4100 5’- GGGGTACCGCACATCTACAAATTGCATGTC and 4101 5’- CCCAAGCTTTTGAACCAGTTACAGGCTTG).
For analysis of transgene expression in blood stages, most parasites were collected from asynchronous blood stage infections in Swiss mice. These mice were either infected intravenously with single infected red blood cell (during the cloning procedure) or intraperitoneally (i.p.) with 105 infected red blood cells. To monitor parasite development in Brown Norway, rats were injected i.p. with 105 parasites. Parasitemia in rats was monitored by analysis of Giemsa-stained thin smears of tail blood collected during the course of infection to a maximum of 51 days post infection. To collect parasites for FACS analysis of fluorescence (see below) 50μ of tail blood of rats was used to infect 2 mice. At a parasitemia of 1–3% tail blood was collected from these mice for FACS analysis.
For analysis of mCherry and/or GFP expression of the transgenic lines, tail blood of infected mice or infected erythrocytes from in vitro cultures [36] were collected in PBS or complete 1640-RPMI culture medium and were examined by FACS (see below) or fluorescence microscopy using a Leica DMR fluorescent microscope with standard GFP and Texas Red filters. Parasites nuclei were labelled by staining with Hoechst-33258 (Sigma, NL) and red blood cell surface membranes were stained with the anti-mouse TER-119-FITC labelled antibody (eBioscience, NL). Briefly, erythrocytes were stained with TER-119-FITC antibody (1:200) and Hoechst-33258 (2μmol/L) at room temperature (RT) for 30 min and washed with 500μL of RPMI-1640 medium (400g; 2min). For DNA visualization, Hoechst-33258 (2μmol/L) was added during the incubation of the secondary antibody. Pelleted cells (400g, 2min) were resuspended in RPMI-1640 medium.
The percentage of blood stage parasites that express mCherry and/or GFP was determined by FACS analysis of cultured blood stages. In brief, infected tail blood (10 μL) with a parasitemia between 1 and 3% was cultured overnight in 1ml complete RPMI1640 culture medium at 37°C under standard conditions for the culture of P. berghei blood stage [46]. Cultured blood samples were then collected and stained with Hoechst-33258 (2 μmol/L, Sigma, NL) for 1 hr at 37°C in the dark and analysed using a FACScan (BD LSR II, Becton Dickinson, CA, USA) with filter 440/40 for Hoechst signals and filter 610/20 for mCherry fluorescence. For FACS analysis the population of mature schizonts was selected on their Hoechst-fluorescence intensity and the percentage of mCherry-expressing parasites was calculated by dividing the number of mCherry-positive schizonts by the total number of mature schizonts [16].
Statistical analyses were performed using Student’s t-test and Two-way ANOVA with the GraphPad Prism software package 5 (GraphPad Software, Inc).
Feeding of A. stephensi mosquitoes and determination of oocyst production was performed as described [90]. Oocyst infection and expression of fluorescently-tagged proteins were monitored between day 8 and 12 post mosquito infection. P. berghei sporozoites were isolated from salivary glands of infected Anopheles stephensi mosquitoes 18–24 days after an infectious blood meal. Fluorescence in oocysts and sporozoites was analsysed using a Leica MZ16 FA microscope.
The human hepatocyte carcinoma cell line Huh7 (JCRB0403, JCRB Cell Bank, JP) is used for in vitro cultures of the liver stages. Isolated sporozoites (5×104) were added to monolayers of Huh7 cells on coverslips in 24 well plates (with a confluency of 80–90%) in ‘complete’ DMEM [90]. At different time points after infection (30, 44 and 48hpi), nuclei were stained with 1 μg/ml Hoechst 33342 and live imaging of the different liver stages, GFP and or mCherry-expressing Huh7 was performed using a DM RA Leica fluorescence microscope.
For immunofluorescence analysis of liver stages, 5×104 sporozoites were added to a monolayer of Huh7 cells on coverslips in 24 well plates in ‘complete’ RPMI 1640 medium supplemented with 10% (vol/vol) fetal bovine serum (FBS), 2% (vol/vol) penicillin-streptomycin, 1% (vol/vol) GlutaMAX (Invitrogen), and maintained at 37°C with 5% CO2. At 30, 44 and 48 hr after infection, cells were fixed with 4% paraformaldehyde, permeabilized with 0.5% Triton-X 100 in PBS, blocked with 10% FBS in PBS, and subsequently stained with primary and secondary antibodies overnight at 4°Cand for 1h, respectively. Primary antibodies used were anti-GFP (594 Alexa Fluor # A21209; Life-Technologies) anti-mCherry (DsRed polyclonal 632496; Clontech and monoclonal # M11217; Life-Technologies). Secondary antibodies used were anti-mouse conjugated to Alexa Fluor 488 (# A11029 Life-Technologies) for GFP and respectively anti-rabbit conjugated to Alexa Fluor 594 (A-21207; Invitrogen) and anti-GFP mouse IgG1k (# 11814460001; Roche) for mCherry. Additionally parasites were stained with primary antibodies: anti-PbEXP1 (PBANKA_092670) raised in chicken [91] or with anti-rabbit PbUIS4 (PBANKA_092670; [92]) and anti-chicken or anti-rabbit secondary antibodies, conjugated to Alexa Fluor® 488 were used for immune-detection (Invitrogen).
Nuclei were stained with Hoechst-33342. Cells were mounted in Image-iT FX Signal Enhancer (Molecular Probes) and examined using a DM RA or a TCS SP8 Leica fluorescence microscope. Images analysis was done with the Leica LAS X software.
For analysis of blood infections after sporozoite infection, Swiss mice were inoculated with 1x104sporozoites by intravenous injection. Blood stage infections were monitored by analysis of Giemsa-stained thin smears of tail blood collected on day 5–8 after inoculation of sporozoites. Infected tail blood was collected at a parasitemia of 1–3%.
A structural model of P. berghei Fam-a gene (PBANKA_1327251) was generated by threading against the resolved structure of the STAR-D2 domain in the human Phosphatidylcholine Transfer Protein (1LN1; [93]) using Phyre2 [94].
Sequences encoding MLN64 (human isoform 2) and the Plasmodium Fam-A family members tested were synthesized (GeneArt and IDT) without introns and codon-optimized for expression in Escherichia coli. The region encoding the START domain was amplified using the primers listed in S3 Table and the resulting DNA fragment was cloned into pMALc2x. Additionally, for the Fam-a protein PBANKA_1327251, we generated a recombinant protein that lacks the final C-terminal alpha helix of the START domain. In the case of MLN64, the restriction sites were included in the synthesized DNA and this was cloned directly into pMALc2x. The sequences of the resulting plasmids were verified by sequencing (Source Biosciences). The plasmids were subsequently transformed in the E. coli strain BL21(DE3) and expression of hexahistidine-tagged Fam A protein was induced in 200 ml bacterial cultures by the addition of isopropyl β-D-1-thiogalactopyranoside (IPTG) to 0.5 mM when the culture was at an OD600 of ~0.5. The bacteria were harvested after an overnight incubation at 18°C and resuspended in column buffer (20 mM Tris, pH 7.4, 500 mM NaCl, 20 mM imidazole) containing protease inhibitors (Complete EDTA-free Cocktail, Roche). The bacteria were lysed with a cell disruptor (Constant Cell Disruption Systems) and the lysate sonicated with a microtip for three 30-second pulses (50% duty cycle, setting 4; Vibracell, Sonics and Materials). The lysate was clarified by centrifugation in a JA25.5 rotor at 9000 x g for thirty minutes. The clarified lysate was mixed with Ni2+ resin (Qiagen) and incubated at 4°C for one hour while rotating. The mixture was poured into a 1.5 x 12 cm chromatography column (BioRad) and the resin was washed with 50 column volumes of column buffer. The protein was eluted with five column volumes of column buffer supplemented with 250 mM imidazole. The eluate was concentrated to 0.5–1.5 ml using a Vivaspin 15 concentrator with a molecular weight cut-off of 10,000 MW (Sartorium Stedim Biotech) and loaded onto a HiLoad 26/60 Superdex 200 prep-grade column equilibrated in standard assay buffer (10 mM HEPES-Na+, pH 7.4, 1 mM EDTA, 50 mM NaCl, pH7.4). Elution of protein was detected through monitoring the UV absorption of the eluate, followed by SDS-PAGE. The fractions containing monomeric protein were concentrated as described above, aliquoted and snap frozen in liquid nitrogen. The MBP-PFA0210c-His6 fusion used as positive control was produced as described previously [52].
The phospholipid transfer activity of the Fam A proteins was measured as previously described [95]. Briefly, acceptor vesicles were produced by mixing phosphatidylcholine and phosphatidic acid in a molar ratio of 98:2, whereas donor vesicles were produced by mixing phosphatidylcholine, phosphatidic acid, N-lactosyl-phosphatidylethanolamine (all non-radioactive lipids were obtained from Avanti Polar Lipids, Inc. and were dissolved in chloroform) in a molar ratio of 88:2:10, with a trace of 14C-labeled phosphatidylcholine (L-α-DiPalmitoyl-Phosphatidylcholine; Perkin Elmer). The phospholipid mixtures were supplemented with 200 μl chloroform, dried under a stream of N2 gas until completely dry and then resuspended in standard assay buffer (10 mM HEPES, pH 7.4, 1 mM EDTA and 50 mM NaCl) such that the total concentration of phospholipid was 2.3 mM. This mixture was sonicated in a sonicating water bath (Ultrawave U300H) until the solution became completely translucent. Transfer assays were set up by mixing 30 μl acceptor vesicles (69 nM total phospholipid), 10 μl donor vesicles (23 nM total phospholipid) and 5 μl 20 mg/ml essentially fatty-acid free bovine serum albumin (1 mg/ml final concentration; Sigma), followed by the addition of Fam A or control protein to a final concentration of 25 μg/ml and standard assay buffer to bring the reaction volume to 100 μl. This mixture was incubated at 37°C for thirty minutes. A 5 μl aliquot was removed to measure total radioactivity in the sample using scintillation counting. To the remainder, 37.5 μl of a 400 μg/ml solution of agglutinin RCA120 (lectin from Ricinus communis; Sigma) in standard assay buffer was added (to a final concentration of 110 μg/ml) to agglutinate the donor vesicles and the samples were incubated on ice for thirty minutes and then at room temperature for ten minutes. The agglutinated donor vesicles were removed by centrifugation for six minutes at 13,000 rpm in a microcentrifuge. The radioactivity in the supernatant, which represents the amount of radioactive phospholipid that was transferred to the acceptor vesicles, was then measured using scintillation counting and the amount of transfer of the radioactivity was calculated.
The fluorescent cholesterol reporter NBD-cholesterol (22-(N-(7-Nitrobenz-2-Oxa-1,3-Diazol-4-yl)Amino)-23,24-Bisnor-5-Cholen-3β-Ol) (Life Technologies) was used to verify cholesterol binding. The sterol was added to a final concentration of 600 nM in a Suprasil quartz fluorescence cuvette (path length 3x3mm) (Hellma) pre-incubated at 37°C, containing 200 μl of 25 mM potassium phosphate buffer (pH 7.4) with 5% DMSO, and mixed. Protein was titrated into the cuvette and the fluorescence was recorded every 0.5 seconds for a 5 min period. The measurements were performed using an FP-6300 spectrofluorometer (JASCO). NBD-cholesterol was excited at 473 nm and fluorescence emission was monitored at 530 nm, at high intensity.
|
10.1371/journal.ppat.1003413 | Cryotomography of Budding Influenza A Virus Reveals Filaments with Diverse Morphologies that Mostly Do Not Bear a Genome at Their Distal End | Influenza viruses exhibit striking variations in particle morphology between strains. Clinical isolates of influenza A virus have been shown to produce long filamentous particles while laboratory-adapted strains are predominantly spherical. However, the role of the filamentous phenotype in the influenza virus infectious cycle remains undetermined. We used cryo-electron tomography to conduct the first three-dimensional study of filamentous virus ultrastructure in particles budding from infected cells. Filaments were often longer than 10 microns and sometimes had bulbous heads at their leading ends, some of which contained tubules we attribute to M1 while none had recognisable ribonucleoprotein (RNP) and hence genome segments. Long filaments that did not have bulbs were infrequently seen to bear an ordered complement of RNPs at their distal ends. Imaging of purified virus also revealed diverse filament morphologies; short rods (bacilliform virions) and longer filaments. Bacilliform virions contained an ordered complement of RNPs while longer filamentous particles were narrower and mostly appeared to lack this feature, but often contained fibrillar material along their entire length. The important ultrastructural differences between these diverse classes of particles raise the possibility of distinct morphogenetic pathways and functions during the infectious process.
| Influenza viruses that have been cultivated in the laboratory usually produce particles that are spherical. However, viruses isolated from patients frequently produce long filamentous particles, as well as smaller elliptical particles that we term “bacilliform virions”. Long filaments may be important for cell-to-cell transmission or facilitate release of the smaller particles by disrupting the mucous layer of the respiratory tract. We have used three-dimensional electron microscopy to investigate the structure of influenza virus filaments ‘budding’ from cells. We found that many of the long filaments had a large bulbous head at the end furthest from the cell. Many of these bulbs were empty while some contained tubules that we believe are made of a scaffold-protein M1 that usually lines the inner surface of the viral membrane. Bacilliform virions contain genomes comprised of eight segments of RNA; these are each wrapped up in protein and packaged in an ordered manner. None of the bulb-headed filaments and very few narrower ones had this feature. We hypothesise that the diverse viral structures we have seen suggest distinct assembly pathways and moreover functions. Long filamentous structures that do not appear to contain genomes may combat the immune response or help the smaller virus particles spread.
| Each year influenza A viruses cause seasonal epidemics, in which many millions of people worldwide become infected. Pandemic strains emerge periodically as a consequence of the segmented nature of the influenza virus genome that predisposes these viruses to reassortment. Complementary subsets of genome segments from two parental strains come together to form a novel virus with a new antigenic character and possibly altered virulence or species specificity. Influenza A viruses are enveloped, single-stranded negative-sense RNA viruses within the family Orthomyxoviridae. The viral envelope is derived from the host cell plasma membrane and bears the glycoproteins haemagglutinin (HA) and neuraminidase (NA) as well as the ion channel protein M2, all of which are critical for virus entry and egress. Beneath the viral envelope is a layer of matrix protein (M1), which is important for virion morphogenesis [1]. The virus interior contains the viral genome, which consists of eight separate RNA molecules [2], [3]. These genome segments are encapsidated by the nucleoprotein (NP) forming eight ribonucleoprotein complexes (RNPs, also termed nucleocapsids), each of which is associated with a viral RNA dependent RNA polymerase (RdRp). Segments one, two and three code for the RdRp proteins (PB2, PB1 and PA respectively), segment four for HA, segment five for NP, segment six for NA, segment seven for M1 and M2, and segment eight for the non-structural protein NS1 and the nuclear export protein (NEP) [4]–[6]. Efficient packing of each segment into a budding virion is directed by specific cis-acting RNA sequences. [7], [8].
Influenza virus particles are pleomorphic, showing significant variations in virion morphology among strains characterized as being either spherical or filamentous. Clinical isolates in particular frequently form filaments that can be many microns long when grown in eggs and cell culture [9]–[11]. Metal shadowing electron microscopy experiments showed that filamentous virions sometimes had large varicosities at one end, proposed to bear spherical particles (known as Archetti bodies, [12]). Spherical virion formation is a trait of laboratory-adapted strains, these particles range in diameter between 80 and 170 nm [13]. The process of virus budding and the resulting virion morphology depend on several viral gene products as well as cellular factors. Each envelope-associated gene product (HA, NA, M1 and M2) plays an important role in virion morphogenesis. Virus-like particles can be generated in the absence of M1, suggesting that HA and NA drive budding [14]. HA also plays a critical role in directing M1 to lipid rafts, the site of assembly and release [15]. However while HA and NA are sufficient for budding, M1 appears to be the principal determinant of virion morphology [11], [16], for example a single point mutation (K102A) induces the spherical A/WSN/33 strain to produce filamentous virions [17]. Virion morphology also depends on cell type, a greater proportion of filaments being produced in polarized cells. Disruption of the actin microfilament network is deleterious to filament production [18].
Electron tomography has recently emerged as a powerful tool to investigate pleomorphic virus ultrastructure [19], [20]. Tomograms of resin embedded and frozen-hydrated short rod shaped influenza virions revealed ordered packing of RNPs, with a single central RNP surrounded by a further seven [21]–[24]. It has been suggested that the same ordered arrangement of RNPs is also present at one end of long filaments [21], [22]. This proposition implies that morphogenesis of short and long filaments initiates through the same mechanism of packaging eight genome segments. It also raises the prospect that long filaments may serve a specific function, perhaps in cell-to-cell transmission or possibly in propulsion of progeny virions away from the infected cell as has been reported for vaccinia virus [25].
Here we present a detailed ultrastructural analysis of particles produced by a filamentous strain of influenza A virus (A/Udorn/72 [H3N2]). Immunofluorescent confocal microscopy and cryo-electron microscopy (CEM) were used to image virus infected cells revealing a profusion of long filaments, many of which terminated in a bulbous structure at their leading ends. Cryo-electron tomography (CET) of these Archetti bodies showed that they frequently contained very little material in the terminal varicosities. In those Archetti bodies that did appear to bear contents, the density resembled tubular assemblies of M1 rather than RNPs. CET of purified filamentous virions yielded improved reconstructions of viral filaments showing two distinct forms: short, rod-shaped particles and long narrower ones. Fewer virions with varicosities at the termini were observed suggesting that the majority of Archetti bodies are lost during the purification process.
Overall we found that long filamentous virions exhibited several morphologies and mostly did not appear to contain RNPs at their leading end. Ultrastructural differences observed between classes of filaments raise the possibility of distinct functions and morphogenetic pathways.
To achieve an overview of the formation of viral filaments we performed confocal immunofluorescence imaging of MDCK cells infected with influenza virus (A/Udorn/72) at low multiplicity of infection (MOI). This revealed an abundance of viral filaments, many of which had bulbous heads (Archetti bodies). Our images demonstrated the position of these heads to be at the leading (distal) end of the filaments (Fig. 1). Some filaments appeared to have varicosities along their lengths, however such features were not seen in subsequent CET analysis, strongly suggesting that these features were a consequence of several filaments of different lengths lying in close proximity. Viral filaments were extremely long, some measuring greater than 10 µm. A time course was performed revealing the presence of virions as early as 6 hours post-infection (p.i.) (Fig. 2) while filaments and Archetti bodies were seen from 8 hours p.i. (Fig. 2–3, movie S1). Fluorescence imaging of unpermeabilised infected-cells confirmed that the filamentous and Archetti structures seen were not artefacts of preparation (Fig S1). Further control experiments were performed to compare these data with patterns of fluorescence seen in A549 cells infected with Udorn (Fig S2). This revealed that fewer and shorter filaments were produced although the filamentous trait was still in evidence. MDCK cells infected with the spherical Influenza (A/WSN/33) strain on the other hand did not show any filamentous forms (Fig S3).
To examine Archetti bodies and other cell associated filamentous structures in more detail and in three-dimensions, we performed CET of infected frozen hydrated cells (Fig. 4, Movie S2). Filamentous structures and Archetti bodies were densely covered in surface spikes confirming their viral origin and had a contiguous matrix layer. The filamentous regions of these particles had a diameter of 74.7±0.78 nm (mean +/− SEM, measured to the tips of the glycoproteins) and extended beyond 10 µm in length (Fig. 4C). Diameters of the bulbous heads ranged from approximately 200 nm to over 550 nm. Of 41 Archetti bodies imaged 25 (61%) were found to be empty while the remaining 16 (39%) had contents within the termini. Segmentation of such particles (Fig. 5A–B, Movie S3) and close inspection of the reconstructed density (Fig. 5C–E) strongly suggested that the contents were tubules formed from M1. These features were seen to be single or paired curved sheets of density (resembling a bracket when viewed in cross-section, Figs. 5D and E). Single sheets lay parallel and closely apposed to the particle envelope (spaced between 25 and 35 nm from the envelope) while paired sheets were 30–35 nm apart. The most likely interpretation of these structures is that they are tubes in which the top and bottom are not well resolved owing to the missing-wedge artefact (features in the z-axis are poorly resolved in electron tomographic reconstructions owing to incomplete sampling caused by the geometry of the transmission electron microscope that prevents tilting the specimen to +/−90°). The missing-wedge also complicated efforts to segment the ‘top’ and ‘bottom’ portions of the viral envelope, however the extent of the bulb was rendered visible by the presence of the gold fiducial markers, revealing that the particle is substantially flattened in the vitreous ice-layer. Furthermore the position of the fiducial markers (which will not have entered the particle) indicates that those putative M1 tubes that are seen as pairs of sheets are also closely apposed to the viral envelope. In all cases the measured internal diameter of the tubes is very similar to the interior diameter of the filamentous particles, supporting our view that these structures are most likely composed of the matrix protein M1.
We have previously reported the presence of tubular matrix derived structures in the paramyxovirus Sendai virus [19]. These structures have been shown to enclose the nucleocapsid in the related measles virus [26]. Our reconstructions of Archetti bodies do not however show evidence of RNPs within these putative M1 tubes.
The filamentous regions of the Archetti bodies sometimes contained sparsely distributed density that could not be attributed to one specific viral protein on the basis of morphology. In addition to being extremely long these filaments were often seen to be flexible (Fig S4A) while others were straight and appeared to be rigid (Figs. 4B, S4A). Such particles sometimes showed fractured apparently open ends (Fig S4B) suggesting that their significant lengths may predispose them to breakage, releasing them from the cell surface. Archetti bodies were however also seen to have pinched off, forming intact particles (Fig S4C–D).
It has proven difficult to visualise the site of budding, as regions of the cell thicker than 500 nm may not be imaged in the cryomicroscope at high-tilt angles owing to the commensurate increase in ice-thickness as tilt angle increases. We have however recorded both low magnification images and tomograms of several filamentous particles that appear to emanate from the cell surface (Fig. 4), some of which are surrounded by cellular processes and vesicles (Fig S5).
In addition to Archetti bodies, we saw many filaments that did not terminate with a bulbous varicosity. Some of these filamentous particles had density at their termini reminiscent of the ordered arrangement of genome segments previously described in smaller filamentous particles [21]. The density was not however clearly enough resolved to image the classical ‘seven around one’ arrangement of genome segments in transverse sections (Fig. 6A inset 1 and 3, Movie S4). In these experiments however, we more commonly saw filamentous particles with no distinct density at their termini, rather they had indistinct density along their entire lengths or they were empty (Figs. 6B–C). To evaluate the relative numbers of the various classes of filaments observed we classified 175 long filaments imaged by CET according to the structures seen at their ends. We found that 21.7% of filaments appeared to contain RNPs, 20% of filaments terminated in bulbous Archetti varicosities while 58.3% had no distinct structures at their ends and simply terminated in a hemispherical cap. Thus 78.3% of filaments had no obvious RNP-like structures.
Interestingly in some filaments we saw extended helical density that we hypothesise to be M1 (Figs. 6A inset 1,2, and 6D, Movie S4). The M1 layer is more usually seen as a tightly packed helical array that is closely associated with the envelope but is harder to resolve without prior bromelain digestion of the surface glycoproteins [21]. Negative stain EM experiments of disrupted virions have also unambiguously demonstrated the helical nature of the influenza virus M1 matrix layer [27].
Experiments to image cell associated filamentous structures did not yield data on particles released into the media and in particular smaller virions were only rarely seen (Fig S6A, B). To provide a structural view of all classes of particle produced by infected cells and to compare the morphology of cell-associated filaments with virions and filaments released into the media we performed CET of purified virus particles (Fig. 7A, movie S5). These preparations were predominantly filamentous and very few Archetti bodies were seen (Fig S6C). Some filamentous structures were found to have varicosities along their lengths however these were more irregular in shape and did not resemble the Archetti bodies seen in our study of infected cells (Fig S6D). In our analysis of virus infected cells Archetti varicosities were predominantly seen to be at the termini of budding filaments, confocal imaging on the other hand appears to show filaments with varicosities along their length. Given the greater clarity of the CET data, we conclude that in most cases such features seen in confocal imaging are most likely the result of several Archetti filaments clustering together to give the appearance of a single entity. Very long filaments and Archetti bodies were seen to be fragile and liable to shear, it is likely then that the majority of longer filaments and Archetti bodies were lost during the purification process and were therefore not frequently seen in our study of purified filamentous particles.
Three forms of virion were observed in our preparations: two distinct classes of filamentous particle: short capsule-shaped particles (52% of the population, Fig. 7B) and long filamentous particles (31%, Fig. 7C) as well as small numbers of spherical particles (17%, Fig. 7D). Longer filaments frequently extended to over 2 µm i.e. beyond the field of view. Our classification of filamentous particles into distinct groups was based on an analysis of their dimensions in four tomograms of one virus preparation.
96 particles were selected and classified as either filamentous or spherical (having an axial ratio <1.2). Diameter and length measurements were made on the dataset and are plotted in Fig. 7E. For filamentous particles that extended out of the field of view the length of the proportion of the filament that was imaged was plotted. Several filaments that appeared concatenated, perhaps having failed to pinch-off, were measured as a single virion. Plotted particle dimensions clearly showed that short filaments had a larger diameter than the longer structures. To determine the statistical validity of this observation, particles were grouped according to whether they were longer or shorter than 250 nm. Student's t-test confirmed that the two filamentous classes had significantly different mean diameters (t = 13.745, d.f = 77, p<0.0001). The shorter class was found to have a mean (+/− SD) diameter of 94.9 (+/−5.6) nm, while longer filamentous particles had a mean diameter of 78.8 (+/−3.6) nm (measured to the tips of the glycoproteins). We introduce the term “bacilliform” to refer to the shorter class of filamentous particles and distinguish them from the long filaments. A strict definition on the basis of a single measurement is not helpful however owing to an overlap in dimensions between filaments of intermediate length and diameter. Consideration of morphological differences between these two kinds of particles is also important therefore.
Close inspection of the filamentous structures revealed major morphological differences between the long narrow filaments and bacilliform particles. Transverse sections through tomograms of bacilliform particles showed a pattern of RNP packaging that adheres to that previously described [13], [22]. Eight segments were arranged in an orderly fashion: with a single RNP at the centre of the virion and a further seven arranged around it (Fig. 8A, Movie S6). In bacilliform particles viewed in longitudinal section three segments were sometimes seen lying side by side (Figs. 8B–C). Such views were readily observed in our data, as the majority of particles were oriented with their long axis parallel to the ice layer. Transverse sections through these reconstructions showed that these particles also had the characteristic RNP arrangement (Fig. 8D). RNPs measured between 10 and 14 nm in diameter with a clear channel running along their centre.
Small numbers of longer bacilliform particles (190–300 nm with a diameter of approximately 85–90 nm) appeared empty at one end and contained RNP density that may correspond to a single complement of genome segments (Figs. 8E and 8F). Similar images have led to the suggestion that very long filamentous virions have a single genome copy at one end [21], [22]. However, internal structures in long filaments were distinct and somewhat harder to interpret in our data. Density frequently extended along significant proportions of these particles and was fibrillar in appearance, measuring approximately 5 nm in diameter with no central channel. Other filaments were only sparsely packed or contained clumps of density (Fig. 8G). In particles that were densely packed, fibrillar density sometimes appeared as long straight rods (Fig. 8H), or appeared wound around itself (Fig. 8J). Transverse sections through these filaments reveal that the fibrillar contents clearly do not correspond with the ordered arrangement of RNPs seen in bacilliform virions (Compare Figs. 8I and K with 8A and D). Indeed those long filaments having an ordered complement of genome segments that were rarely observed in studies of infected cells were even more infrequently seen in tomograms of purified filaments (One seen - Fig. 8L). While many long-filament termini were imaged, the great majority of which were devoid of RNPs, fewer tomograms of filaments were available in which both ends were seen. Those that were imaged however did not appear to contain RNPs and were either empty or had fibrillar contents along their entire length (Fig S7).
We have examined influenza A/Udorn/72 filament formation by CET and immunofluorescent confocal microscopy. Confocal microscopy revealed the presence of a variety of filamentous forms: long straight filaments, flexible filaments and Archetti bodies. CET demonstrated that filaments produced in MDCK cells frequently do not contain ordered RNPs at their distal ends and more often terminate with empty ends or Archetti varicosities. These data represent the first structural analysis of influenza virus filaments budding from the host cell under near native conditions. Early metal shadowing TEM studies of filamentous virions produced in eggs described Archetti bodies and proposed that they might bear spherical virions [12], [28]. Our data refute these suggestions and reveal the absence of RNPs or spherical virions in the terminal varicosities of Archetti bodies. It is unclear from our analysis whether these particles bud from the cell surface as bulbous structures or whether the terminal varicosity forms after budding. The presence of tubular assemblies that we attribute to M1 in many Archetti termini suggests that the latter may be the case. However, the envelope measures approximately 12 nm thick in this region, similar to measurements made of the envelope/matrix component of purified particles, budding filaments and empty Archetti bodies. This suggests that the presence of M1 tubes might not be the result of them having detached from the inner surface of the envelope, an occurrence that could result in loss of filamentous form. We frequently saw density within filaments that appears to be a second layer of M1, such as the helical assemblies shown in figure 6 and inclusions within varicosities (Fig S6D). It is possible then, that concentric layers of M1 may be a common feature of influenza A filaments. Owing to the missing wedge we are unable to unambiguously image the entire envelope of the Archetti varicosity however. Thus the origin of the putative M1 tubes cannot be conclusively proven and further experimentation is required to establish the origin of these structures and the morphogenetic pathway of Archetti particles.
As in our study of budding virus, imaging of purified virus also showed the presence of long filamentous particles. Previous structural studies of purified influenza A/Udorn/72 virus imaged filaments of intermediate length (∼500 nm) and did not highlight the fundamental differences in internal features between filaments and bacilliform particles that we have observed [21]. Our data clearly showed the presence of fibrillar material running along the filament interior and longer filaments containing RNPs were very rare. We saw very few spherical particles and the majority of smaller particles would be better described as obloid, prolate or bacilliform. These particles contained a well-ordered arrangement of RNPs when viewed in transverse and longitudinal sections that appeared similar to those previously described and have been shown to be supercoiled circularized nucleocapsids [21]–[24], [29]–[31]. Harris et al. observed dense ‘solenoid shaped’ material proposed to be nucleocapsids in their CET study of spherical influenza A virions [13]. Similar density observed in filamentous virus by Calder et al. was attributed to M1 however [21]. This illustrates the difficulties associated with attributing density in tomograms to specific components on the basis of morphology and in the absence of firm biochemical evidence. Likewise we have encountered difficulties in our interpretation of density within very long filaments and have yet to determine the origin of material seen in purified and cell associated filaments. It should be noted that in our experiments we have visualised different influenza filament populations in purified preparations than in virus-infected cells. The fibrillar density seen in purified particles is more clearly resolved than interior density in cell-associated filaments. Thicker ice in the latter experiments may reduce the contrast of these features; indeed surface glycoproteins are also less well resolved. Alternatively this may reflect genuine differences in filament composition.
Long filamentous particles had a narrower diameter than the bacilliform particles and there was frequently no evidence of a well-ordered arrangement of RNPs inside. Our findings indicate that the longer filamentous forms do not in the main simply represent elongated versions of the shorter bacilliform particles. Various classes of filaments are present and these might assemble according to distinct morphogenetic pathways. Both filaments and bacilliform virions have a regular cylindrical shape, capped at each end by a hemisphere, or in the case of Archetti bodies, with a bulbous head at one end. There is considerable evidence that the filamentous phenotype is controlled by M1 [11], [16], [17], moreover preservation of the filament morphology correlates with an intact M1-layer. It seems plausible therefore, that formation of one or other class of virion may be controlled by the oligomerisation of M1 at the site of virion assembly, where the resulting curvature of the leading end might be influenced by a number of factors including M2 [32] and the presence or absence of eight RNPs. This latter possibility provides perhaps the simplest explanation of the various forms seen. If viral proteins were capable of initiating assembly at the plasma membrane in the absence of RNPs this may result in a smaller radius of curvature and consequently a narrower filament. The proximal end of RNPs in the budding virion may also be required to initiate pinching off, with the absence of this signal therefore leading to long filament formation. The smaller radius of curvature or absence of RNPs at the distal end may also result in a less robust structure, prone to loss of integrity and formation of the Archetti varicosity (although we see no evidence of loss of the envelope M1 layer). Small numbers of long filaments that contain RNPs might also be formed as a consequence of a failure in the pinching off process, leading to a switch to the alternative helical packing of M1 necessary for the formation of the narrower diameter long filament.
Despite clinical isolates frequently exhibiting the filamentous phenotype, the role played by these long filaments in the infectious process is yet to be established. They have been suggested to be involved in cell-to-cell transmission [18]. Clearly virions that can be greater than ten microns in length could not initiate infection via receptor-mediated endocytosis. It has recently been shown that filaments can however enter cells by macropinocytosis [33]. Recent work has also highlighted possible advantages of the filamentous shape over spherical in the context of effective transport and trafficking to the respiratory epithelium [34]. The various functions of the diverse filaments we observe may be resolved upon identification of the density at the particle interior. The small number of filaments that comprise an ordered complement of RNPs at their leading end may indeed play a role in cell-to-cell transmission. There is also evidence to suggest that filaments may contain multiple sets of RNPs; early studies showed higher infectivity in filament preparations in comparison to spherical virions while UV inactivation experiments suggested that filaments are polyploid [16], [35], [36]. Our data do not refute this however it is very difficult to create pure preparations of long filaments to verify these data. If the material in these particles were found not to be RNPs this would point to an alternative role, perhaps in virus pathogenesis. Immunoglobulin A (IgA) is highly expressed in the respiratory tract and plays a key role in resistance to influenza infection [37]. It is possible that those long filaments and/or Archetti bodies with no genome segments, being studded with significant quantities of HA might act as a “decoy” for the host immune response, binding antibody and helping the infectious virions to evade neutralisation. It is also conceivable that due to their extreme lengths they could physically disrupt the mucociliary layer in the respiratory epithelium, thereby facilitating rapid spread of infectious bacilliform particles.
We have shown that the filamentous Udorn/72 strain of influenza A virus produces a variety of filamentous particles and Archetti bodies, the majority of which do not contain ordered RNPs at their leading end and may have a unique role to play during viral infection. The functional significance of these diverse structures remains to be confirmed. Future studies employing human airway model systems may provide further insights and a more biologically relevant view of the role these particles play in the infectious process and viral pathogenesis.
The H3N2 strain influenza A/Udorn/72 was cultivated in MDCK cells. Cells were grown to confluence at 37°C in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% foetal calf serum (FCS). 3×108 cells were infected at a multiplicity of infection (MOI) of 0.001. Following an incubation period of 1 h, the media was replaced with serum-free DMEM containing 2.5 µg/ml N-acetyl trypsin (NAT, Sigma) followed by incubation at 37°C. 36 hours post-infection (p.i.), the supernatant was harvested and clarified by centrifugation at 2500 rpm for 5 min. Following a second clarification step (10,000 rpm for 30 minutes), virus was pelleted by centrifugation onto a 30% sucrose cushion in NTE buffer (1 mM EDTA 10 mM, 150 mM NaCl, Tris-HCl, pH 7.5) at 25,000 rpm for 2.5 h. The pellet was then resuspended in 250 µl NTE and run through a continuous sucrose gradient (30%–60%) at 25,000 rpm for 2.5 h. Finally, banded virus was collected and centrifuged at 31,000 rpm for 2 h. The pelleted virus was resuspended in 100 µl of NTE buffer.
Confluent monolayers of MDCK cells grown on cover slips were infected at an MOI of 0.6. Cells were then fixed 24 h p.i. with 4% formaldehyde/2.5% Triton X-100 in PBSA for 30 min. They were then washed three times with PBS, followed by blocking in sheep serum or rabbit serum for 1 h. Labelling was then performed by incubating cells for 1 h at room temperature with primary antibodies. Unbound antibody was removed by washing three times with PBSA prior to incubation for 30 min at room temperature with fluorescent-tagged secondary antibodies. Finally cells were washed three times with PBSA and mounted using ProLong Antifade plus DAPI reagent (Invitrogen, UK). For the time course experiment, infected MDCK cells (moi 3) were fixed at various time points; 6 h, 8 h, 10 h, 12 h, 14 h, 16 h, 18 h and 27 h. Infected MDCK cells fixed at 27 h with only 4% formaldehyde served as a non permeabilised control. Controls were also carried out by infecting A549 cells with H3N2 Udorn (moi 3) and MDCK cells with H1N1 WSN (moi 5) followed by fixation with 4% formaldehyde/2.5% Triton X-100 13 h p.i. Labelling for the time course and control samples were performed as detailed above. All samples infected with H3N2 Udorn were immunolabelled for the H3 haemagglutinin with a mouse monoclonal antibody raised against A/X-31/1968 H3N2 virus, kindly provided by Prof. John Skehel (NIMR, London), while the H1N1 WSN samples were labelled for H1 with a mouse monoclonal antibody raised against A/PR/8/34 H1N1, kindly provided by Prof. Paul Digard (Edinburgh). H3 and H1 were detected with a rabbit anti-mouse Alexa Fluor 633 (Invitrogen, UK). Nucleoprotein (NP) labelled using a mouse monoclonal antibody (Abcam, UK) was detected using a sheep anti-mouse-FITC conjugate (Sigma, UK). Phalloidin Alexa Fluor 568 (Invitrogen, UK) was used to stain actin. Immunofluorescent imaging was carried out with Zeiss LSM510 Meta and LSM710 laser confocal microscopes.
Gold 200 mesh TEM grids with holey carbon support film (Quantifoil Micro Tools GmbH, Jena, Germany) were sterilized with ethanol and then coated with laminin overnight in glass-bottomed dishes (MATTEK Corporation Inc, USA). Grids were then washed in water and seeded with 100,000 MDCK cells per dish in DMEM media supplemented with 10% FCS. They were then incubated overnight at 37°C. Cells were infected at an MOI of 0.6 for 1 h at 37°C. The media was then replaced by serum free DMEM supplemented with 2.5 µg/ml NAT and incubated for 19 h.
For cryo imaging of purified virus, preparations were mixed with 10 nm colloidal gold (British Biocell International, Cardiff, UK) in a ratio of 1∶3 v/v. A 5 µl aliquot was applied to freshly glow-discharged Quantifoil holey carbon support films (R2/2 200 mesh copper grids - Quantifoil Micro Tools GmbH, Jena, Germany), blotted and frozen by plunging into liquid ethane as previously described [38].
For cryo imaging of virus infected cells grown on Quantifoil EM grids 15 nm colloidal gold (British Biocell International, Cardiff, UK) was added in a ratio of 1∶3 v/v. Grids were then blotted and frozen by plunging into liquid ethane.
Tilt-series imaging was performed on a JEOL 2200FS energy-filtering transmission electron microscope equipped with a Gatan Ultrascan 4 k×4 k CCD camera and a Gatan 914 high-tilt cryo-stage. The microscope was operated at 200 kV and zero-loss energy filtered imaging with a slit-width of 30 eV was used to enhance image contrast. Tilt series were recorded using the SerialEM software package [39]. Images were acquired at two-degree increments from −70° to +70° between 10,000× and 20,000× magnification for cells on grids and at 20,000× or 40,000× magnification for purified virus. Images were recorded with two-times binning, corresponding to a pixel sizes ranging from 21.2 to 5.4 Å/pixel in the specimen. The target defocus was set to between 4 and 6 µm under-focus and the electron dose ranged from 83 e/Å2 to 100 e/Å2 per tilt-series. Tomograms were calculated and visualized using the IMOD software package [40]. Reconstruction was performed using weighted back projection followed by denoising using non-linear anisotropic diffusion. Figures were prepared by averaging 10 tomogram sections using IMOD's 3dmod slicer routine. Segmentation was performed manually using Amira (Visual Sciences Group).
|
10.1371/journal.pcbi.1002227 | Patient-Specific Data Fusion Defines Prognostic Cancer Subtypes | Different data types can offer complementary perspectives on the same biological phenomenon. In cancer studies, for example, data on copy number alterations indicate losses and amplifications of genomic regions in tumours, while transcriptomic data point to the impact of genomic and environmental events on the internal wiring of the cell. Fusing different data provides a more comprehensive model of the cancer cell than that offered by any single type. However, biological signals in different patients exhibit diverse degrees of concordance due to cancer heterogeneity and inherent noise in the measurements. This is a particularly important issue in cancer subtype discovery, where personalised strategies to guide therapy are of vital importance. We present a nonparametric Bayesian model for discovering prognostic cancer subtypes by integrating gene expression and copy number variation data. Our model is constructed from a hierarchy of Dirichlet Processes and addresses three key challenges in data fusion: (i) To separate concordant from discordant signals, (ii) to select informative features, (iii) to estimate the number of disease subtypes. Concordance of signals is assessed individually for each patient, giving us an additional level of insight into the underlying disease structure. We exemplify the power of our model in prostate cancer and breast cancer and show that it outperforms competing methods. In the prostate cancer data, we identify an entirely new subtype with extremely poor survival outcome and show how other analyses fail to detect it. In the breast cancer data, we find subtypes with superior prognostic value by using the concordant results. These discoveries were crucially dependent on our model's ability to distinguish concordant and discordant signals within each patient sample, and would otherwise have been missed. We therefore demonstrate the importance of taking a patient-specific approach, using highly-flexible nonparametric Bayesian methods.
| The goal of personalised medicine is to develop accurate diagnostic tests that identify patients who can benefit from targeted therapies. To achieve this goal it is necessary to stratify cancer patients into homogeneous subtypes according to which molecular aberrations their tumours exhibit. Prominent approaches for subtype definition combine information from different molecular levels, for example data on DNA copy number changes with data on mRNA expression changes. This is called data fusion. We contribute to this field by proposing a unified model that fuses different data types, finds informative features and estimates the number of subtypes in the data. The main strength of our model comes from the fact that we assess for each patient whether the different data agree on a subtype or not. Competing methods combine the data without checking for concordance of signals. On a breast cancer and a prostate cancer data set we show that concordance of signals has strong influence on subtype definition and that our model allows to define prognostic subtypes that would have been missed otherwise.
| Molecular data show great promise to stratify patients into distinct subgroups that are indicative of disease development, response to medication and overall survival prospects [1]. Such subgroups are highly useful in informing treatment decisions [2], [3]. Most current computational diagnostic approaches are based on gene expression data and cluster patients by co-expression of genes. For example, multivariate gene expression signatures have been shown to discriminate between disease subtypes, such as recurrent and non-recurrent cancer types or tumour progression stages [3]–[6].
In addition to expression data there are also many other data types that can be informative about a patient's disease status. For example, somatic copy number alterations provide good biomarkers for cancer subtype classification [7]. For this reason, the focus of research has recently shifted towards integrative clustering of complementary data types, e.g. [8]. The goal of integrative analysis is to identify clusters of samples that share not only expression profiles, but also other molecular characteristics such as copy number alterations. The subtypes of tumours identified in this way are more likely to share the same regulatory programs and underlying genomic alterations.
Data integration for subtype discovery poses several challenges that we address in this paper.
Challenge 1: Separating concordant from contradictory signals. While different molecular data are expected to share complementary information on common cellular processes, they can also contain contradictory signals because of the complexity of living cells and noise in the data. For example, genomic gains and losses may or may not be accompanied by concordant expression changes of the genes in the altered regions. The level of concordance may differ dramatically from patient to patient due to cancer heterogeneity. However, most existing integrative methods force different data types to be fused in all samples without reference to whether the data are concordant or contradictory in each patient.
Challenge 2: Selecting informative features. Identifying which measurements are informative about the underlying subtypes is particularly important when using genomic data because the number of measurements can be very large, e.g. in the tens of thousands or more in the case of microarrays. Because a priori we expect only a fraction of measurements to contain useful clustering information, extracting these features accurately will improve the quality and stability of clustering outcome. Additionally, identifying the relevant biological features can inform us about the underlying processes driving the disease.
Challenge 3: Estimating the number of subtypes. In many clustering algorithms this number is a parameter that needs to be set by the user [8]. Afterwards, the quality of the clusterings need to be compared, e.g. using stability indices [9]. However, jointly estimating the clusters together with their optimal number in a unified framework can improve results, because the most likely number of clusters can be inferred directly from the data.
These three challenges are not independent of each other: Whether or not the data show concordant signals for a subgroup of patients has a direct effect on which features should be selected as informative, which in turn has a direct influence on the estimate of the number of clusters. Thus, all three challenges need to be treated in an unified model.
Our approach is Patient-specific Data Fusion (PSDF) by Bayesian nonparametric modeling. In this paper, we propose a statistical model based on a two-level hierarchy of Dirichlet Process (infinite mixture) models (DPMs) [10], [11] that integrates copy number and expression data to jointly classify patients into cancer sub-groups. This model is an extension of the model presented in [12], modified to include a method of feature selection and adjusted to address a different problem with a number of advantages:
Thus, the model not only identifies copy number alterations driving gene expression changes but simultaneously finds differences in regulation that distinguish one cancer subtype from the other. In doing so it explores the basic scientific question to which extend copy number data can be fused with expression data in integrative cancer studies.
everal integrative clustering approaches have been proposed in the literature [8], [13], [14]. A recent method is iCluster [8]. iCluster is based on a k-means approach that is extended to include more than one data type and performs feature selection in each data type independently. iCluster is fast and easily applied to more than two data types. However, compared to iCluster we have a more flexible mixture model underlying our own approach that in particular does not need the number of clusters (the ‘k’ in ‘k-means’) to be specified beforehand. In contrast to our model, iCluster assumes that both data are informative for all patients without checking for patient-specific consistency. In two case studies with cancer data sets [7], [15], we will show what impact these differences have and that our model compares favourably with iCluster in clinically important analysis results.
We introduce PSDF as an unified model to address the above three key challenges in patient subtype discovery. To demonstrate the power of this patient-specific integrative method, we analyse a breast cancer data set and a prostate cancer data set. High degree of concomitant changes has been observed in copy number and expression changes in breast cancer [15], [16]. In contrast, prostate cancer data display entirely different characteristics with relatively few co-ordinated genomic-transcriptomic changes [7], [17]. Therefore, these two cancer types represent two very different cases in terms of fusion ability, making them ideal for validating PSDF. Both the Matlab code for PSDF and pseudo-code for our work flow of data preprocessing and downstream analysis are available at https://sites.google.com/site/patientspecificdatafusion/.
Bayesian nonparametric modeling provides a principled way to learn unknown structure in the data. Dirichlet Process (infinite mixture) models (DPMs) [10], [11] are Bayesian nonparametric models that have been widely used for clustering [18]–[25]. DPMs give us a sound interpretation of common cluster membership, that the data for those samples are drawn from the same underlying distribution. They also allow us to infer the most likely number of clusters given the data as part of the unified model.
PSDF groups patient samples on the basis of both gene expression and copy number alteration data. It also simultaneously distinguishes, on a sample-by-sample basis, between samples that can share concordant signal across the data types (fused) and those for which there is contradiction (unfused). We note that throughout this paper we will use the following terminology, relating to the concordance (or otherwise) of the two data sets for a given patient.
The breast cancer data from [15] contains both copy number and expression data for 106 tumour samples, with 26,755 copy number probes and 37,411 expression probes. Even for a clustering method with feature selection capability, it is convenient to remove the mostly obviously uninformative “noise” features. To preselect features with functional implications in a principled, controlled manner, we take the following steps.
First, copy number data are filtered based on whether there is a concomitant change between a locus's copy number and its own expression. This is to exclude passenger events without explicit downstream effects. Each expression probe is matched to its nearest copy number probe allowing for multiple matches, i.e. a copy number probe can be matched to multiple expression probe. This resultes in 37,411 matched pairs of copy number and expression data annotated by expression probes. We then calculate the adjusted -values of the correlations of each pairs of copy number and expression probes, and a copy number probe is selected if the corresponding -value is smaller than 0.1. Still there are highly similar copy number profiles among the selected copy number probes. To remove redundancy, copy number data of the selected probes are then merged based on their similarity using CGHregions [26], which results in 379 regions. Finally, both of the copy number signatures from the merged regions and all expression profiles passing the above -value threshold are ranked by the Wald test in predicting breast-cancer-specific survivals. The best 200 of each type of data are used for clustering.
For the prostate cancer data set, there are 150 tumour samples with both copy number and expression data [7]. The expression data were profiled with Affymatrix Human Exon 1.0 ST array which contains 229,581 probes after quality filtering. For the copy number data, there are 43,416 probes on Agilent 244K array comparative genomic hybridization array.
To extract features, we use a slightly different approach since the scale of this data set is much larger than that of the breast cancer data. Substantially larger number of probes compared to the breast cancer study means that the probe-centric method is not suitable, hence we take a gene-centric method by aggregating copy number and expression data to 12,718 genes based on array annotation. For copy number data, the aggregation is done by taking the median for probes within a gene. For the expression, the probe most highly correlated with the copy number profile of a gene is chosen to represent this gene. Even if so, only modest correlations are observed between the two data types. Finally, 286 genes with highly correlated copy number and expression (adjusted ) from the two data sets are used as clustering input.
This paper explores the potential of patient-specific data fusion to enhance prediction power in cancer subtype discovery. Cancer subtype discovery combining both genomics and transcriptomics leads to a more comprehensive understanding of the heterogenous cellular contexts. By using a flexible, nonparametric model such as the model presented in this paper, we can learn both the concordant and contradictory structures underlying those multiple data types. This structure leads to an improved understanding of the functional components and pathway regulations for each cancer subtype, something that is essential for the future development of targeted therapeutics. Our contributions are therefore as follows.
With both breast cancer and prostate cancer data, PSDF is able to discover poor outcome subtypes with early-stage, highly frequent recurrences/deaths. These subtypes are not identified by other methods which either force to fuse data on all samples, or cluster patients based on single data type. We show that there exist both concordant and contradictory signals in these data, which, when forced to cluster together, can result in inferior subtype identification. Moreover, data fusion is necessary in predicting both events and timing of cancer survivals/recurrrences. Hence, taking this approach is vital in the discovery of new disease subtype consisting of early-stage events.
A promising aspect of studying cancer subtypes is the identification of key pathways altered unique to this subtype. Our network analyses show functionally interacting genes in the subtype-specific network modules whose deregulations may contribute to the poor outcome of a cancer subtype. The pathway enrichment analysis facilitates functional interpretation of the new clusters/subtypes in a coherent manner with the network modules. Under-lying driver events for poor outcome may be revealed during this process, such as the over-expression of the Cell Cycle pathway in breast cancer, and the under-expression of Endocytosis and Chemokine signaling pathway in prostate cancer. Further exploration of these results may lead to the discovery of new genes participating in the cancer-related pathways, as well as the identification of treatment target and the development of pathway inhibitors.
Our analysis results also highlight the difference between different cancer types. Previously, relatively low concordance between prostate cancer copy number and expression has been reported [17], in contrast to the high-level correlations generally observed in breast cancer. In addition, unlike breast cancer where RNA expression are predictive of recurrence, copy number changes in prostate cancer have been found to outperform expression in prediction [7]. Different degrees of concordance in the data lead to significantly different clustering results – while fused clusters in highly concordant breast cancer data are prognostic, an unfused subtype in prostate cancer turns out to be extremely aggressive. The results from the breast and prostate cancer data sets are in fact strong statements that different cancer types should be treated differently by statistical methods. Hence, a versatile tool such as PSDF is particularly suitable for this field.
PSDF extends the model of [12] to include feature selection. The model is motivated by the need to address three main challenges in data-fusion-based clustering, namely (i) to separate concordant from contradictory signals, (ii) to identify which features are informative and (iii) to estimate the number of disease subtypes.
PSDF is constructed from a two-level hierarchy of Dirichlet Processes, as shown in Fig. 7. Each patient has a binary state () that defines whether their data are concordant across the data sets, either fused () or unfused ().
Within any given mixture component from the Dirichlet Processes, we model the (discretised) data as being drawn from a multinomial distribution with a weakly informative multinomial prior. The features are assumed to be independent, giving rise to a naive Bayes data model for each data set. We use this data model for both gene expression and copy number data sets. Since our method use discretised data as input, copy number calls are made with R package CGHcall [43]. Without match normal expression data, we use quantile discretisation to deem the top 10% log2 ratio data as over-expressions and bottom 10% data as under-expressions, similar to [44], [45]. In cases when match normals are available, appropriate methods such as the one in [46] can be used for discretising the expression data. As a result, the copy number data are discretised into three levels of loss, neutral, and gain, and the expression data are discretised into three levels corresponding to under-, normally- and over-expressed.
We note that in principle, this model could be extended to 3+ data sources. In practice however, this will become unwieldy, and so we restrict ourselves in this paper to considering fusion between two data sources. We are currently developing a related model that will scale much better with increasing numbers of data sources.
The naive Bayes data model used in [12] models data for a given feature as being drawn from a multinomial distribution with unknown class probabilities. Choosing a conjugate (Dirichlet) prior, these unknown class probabilities can be marginalised out to give a marginal likelihood for each feature in each cluster.(1)Where and , is the index over features and is the index over discrete data values. The are the Dirchlet prior hyperparameters, which in this case are set to match the known proportions of each data value in the data set (which is prior knowledge here, as we define the data discretisation). These proportions are scaled to sum to 1.5, which is the sum of the Jeffreys' value (0.5) over the three possible data values, hence representing only a weakly-informative constraint.
To perform feature selection, we will consider two different likelihoods for a given feature, corresponding to the feature being off/on, as denoted by an indicator variable . For , we simply use the multinomial-Dirichlet marginal likelihood, as before. For , we fix the class probabilities to the expected prior values, given the spread of discrete input values for the given feature.(2)Where again is the index over features and is the index over discrete data values. The are simply taken as the proportion of each data value in a given feature across the whole data set, with a minimum count of one assigned to each data value.(3)Where and are required to have minimum of one count per class.
This has the effect of defining an ‘indifference’ likelihood, where it makes no difference to the overall posterior (for the given feature) to which cluster any given sample is assigned. It is straightforward to write down the conditional distribution for a single indicator variable , so we Gibbs sample each in turn when producing a new MCMC sample.
The switching on/off of a given feature can be regarded as a kind of model selection. Considering the limit of many samples (and hence negligible uncertainty in the value of the class probabilities for ), the ‘indifference’ likelihood is simply the expected case if the samples are randomly assigned to clusters. For finite numbers of samples, the ‘indifference’ likelihood is inherently simpler (in the sense that the class probabilities are known), so the feature selection becomes a competition between this simplicity and the greater ability of the case to explain non-random cluster assignments.
To give improved mixing, we run 50 MCMC chains for each analysis. The chains are samples long, with the first removed as a burn-in. The remainder are sparse-sampled by a factor of 10 for computational convenience and then used to produce the outputs.
All chains are examined using the R package CODA. In particular, the time-series and histograms for each parameter/chain pair are examined by eye for any obvious anomolies that would indicate incomplete mixing.
The multiple MCMC chains are used to compute uncertainties in statistics of interest (for example, the probability that a given feature is selected). This gives us a direct measure of chain mixing quality.
Each chain runs to completion in less than 48 hours on nodes of the University of Warwick's high performance computer cluster.
In order to validate our model, we performed a simulation study. We constructed a pair of synthetic data sets. For each synthetic data set, we started with the 106 signal items and 200 signal features in the copy number variation data from [15] (which is also analysed in Section. These items will therefore (by construction) be fused as they share identical clustering structure across the two synthetic data sets. We note that this is a reasonable test of the method because in the real analyses both copy number and gene expression data sets are discretised into three levels. These synthetic data represent a good way of constructing items that share concordant signals across the two data sets.
To each synthetic data set, we then added 50 noise items. These items are drawn by replacement from the signal items and are drawn separately for each synthetic data set. For example, a given noise item may be a copy of signal item 15 in the first synthetic data set, and signal item 59 in the second synthetic data set. These noise items are therefore drawn from the existing clustering structure of each synthetic data set, but in general they will not be fused (excepting the case where by coincidence they are both drawn from the same underlying cluster). This then gives us 156 items in total.
Finally, we added to each synthetic data set 200 noise features. The data for these features are drawn with replacement from the original data. Therefore, while they reflect the distribution of data values in the signal features, they are entirely random and without clustering structure. As such, we expect them o be rejected by feature selection.
Table 1 shows the results of an analysis of these synthetic data. The method successfully rejects all 400 noise features across the two data sets. 8 signal features are also rejected at this level, but we note that some level of feature rejection is expected of signal features, as some of them will be uninformative.
The method successfully finds 105 of the 106 fused items. It also identifies 17 of the noise items as being fused. We note that we expect some level of coincidental fusion for the noise items, where they happen to have been drawn from the same cluster. For example, if we assume there are 5 (equally-sized) underlying clusters in the copy number data, we expect coincidentally fused noise items. We note that here, 25 MCMC chains of length samples are sufficient to achieve reasonable convergence. We conclude that our method performs well in identifying both fused/unfused items and selecting appropriate features in each data set.
|
10.1371/journal.ppat.1005693 | Hepatitis B Virus-Induced Parkin-Dependent Recruitment of Linear Ubiquitin Assembly Complex (LUBAC) to Mitochondria and Attenuation of Innate Immunity | Hepatitis B virus (HBV) suppresses innate immune signaling to establish persistent infection. Although HBV is a DNA virus, its pre-genomic RNA (pgRNA) can be sensed by RIG-I and activates MAVS to mediate interferon (IFN) λ synthesis. Despite of the activation of RIG-I-MAVS axis by pgRNA, the underlying mechanism explaining how HBV infection fails to induce interferon-αβ (IFN) synthesis remained uncharacterized. We demonstrate that HBV induced parkin is able to recruit the linear ubiquitin assembly complex (LUBAC) to mitochondria and abrogates IFN β synthesis. Parkin interacts with MAVS, accumulates unanchored linear polyubiquitin chains on MAVS via LUBAC, to disrupt MAVS signalosome and attenuate IRF3 activation. This study highlights the novel role of parkin in antiviral signaling which involves LUBAC being recruited to the mitochondria. These results provide avenues of investigations on the role of mitochondrial dynamics in innate immunity.
| Hepatitis B virus (HBV) chronic infection is one of the major causes of hepatocellular carcinoma. HBV infection is associated with mitochondrial dysfunction. We previously showed that persistent infection of HBV requires rapid clearance of impaired mitochondria by mitophagy, a cellular quality control process that insures survival of HBV infected cells. During the process, Parkin, an RBR E3 ligase, is recruited to mitochondria to induce mitophagy. In this study, we show that the Parkin, plays a critical role in the modulation of innate immune signaling. Using HBV expressing cells, we show that the Parkin recruits linear ubiquitin assembly complex (LUBAC) to the mitochondria and subsequently inhibits downstream signaling of mitochondrial antiviral signaling protein (MAVS). Mitochondrial LUBAC then catalyzes linear ubiquitin chains on MAVS, which abrogates its downstream events such as MAVS-TRAFs interaction and abolishes IRF3 phosphorylation. The results of this study highlight the molecular details explaining how HBV can suppress interferon synthesis implicating a mitophagy-independent role of Parkin. HBV-induced mitochondrial damage serves as the platform for recruitment of Parkin and LUBAC, which together modify MAVS by ubiquitination and cripples its downstream signaling.
| Infection by the human hepatitis B virus (HBV) is a major public health burden associated with about 600,000 deaths annually and 350 million chronic carriers worldwide [1]. Chronic hepatitis is associated with the progression of disease to liver failure and hepatocellular carcinoma [2]. HBV belongs to the Hepadnavirus family. The small HBV genome contains multiple translational reading frames to produce different HBV proteins [2]. These open reading frames (ORFs) include; S, C, P and X. The S ORF codes for the hepatitis B surface antigen (HBsAg). The C ORF codes for the core (HBcAg) and the e antigen (HBeAg) proteins. HBV core protein contains a cluster of highly basic amino acids and intrinsically has a property of self-assembly and RNA binding. The P ORF codes for the polymerase protein, which contains a reverse transcriptase activity that catalyzes the conversion of pregenomic RNA into viral DNA [2]. The X ORF codes for a multifunctional X protein (HBx) affecting a wide variety of cellular functions [3]. HBx is required for productive HBV replication [3].
Mitochondrial injury is a prominent feature underlying the pathogenesis of chronic hepatitis B virus-associated liver disease [4–6]. We previously reported that HBx primarily localizes to the mitochondria and directly interacts with the outer mitochondrial voltage-dependent anion channel, VDAC3 [7, 8]. HBx expression results in the loss of mitochondrial transmembrane potential (ΔΨm), increase in the level of reactive oxygen species (ROS), and mitochondrial calcium levels suggestive of its profound effect on mitochondrial homeostasis and function [7, 9].
Mitochondria serve as a platform for innate immune signaling and play indispensable role in cellular antiviral defense [10]. MAVS, a mitochondrial membrane protein, is the central adaptor molecule on which signals from many pattern recognition receptors (PRRs) that specifically recognize viral nucleic acids converge [10, 11]. RIG-I like receptors (RLRs) are the well-characterized cytoplasmic sensors that sense viral RNA [12, 13]. RIG-I oligomerizes around the bound RNA in ATP-dependent manner and interact with MAVS through CARD-CARD domain association [12]. Activated MAVS recruits multiple effector components to initiate a complex cascade of signaling events that lead to the recruitment and activation of TANK-binding kinase 1 (TBK1) [14]. Activated TBK1 phosphorylates and activates interferon-regulatory factor-3 (IRF-3) and IRF-7 leading to IFN-β synthesis [10]. Although HBV is a DNA virus, it replicates by the reverse transcription of a pre-genomic RNA (pgRNA) intermediate [2]. A recent report demonstrates that RIG-I senses the 5’-ε region of the HBV pgRNA and induces type-III IFN synthesis with no significant induction of IFN β [15]. In support of this, HBV polymerase has been previously shown to dampen RIG-I signaling and inhibit the IFNβ synthesis by inhibiting the interaction between TBKI and DDX3 [16]. Moreover, the HBx protein is also shown to suppress IRF3 activation by disrupting the MAVS-complex as well as by downregulating MAVS expression [17–19]. It has been shown that HBV expression can activate RIG-I–MAVS axis to invoke countermeasures to target downstream steps to abrogate IRF3 activation and thereby IFN β synthesis [20, 21].
We previously reported that HBV induces mitochondrial translocation of Parkin and subsequent Parkin-dependent mitophagy to promote viral persistence [4]. Recent studies implicate mitochondrial dynamics and mitophagy in the modulation of antiviral signaling [10, 22]. Parkin, a cytosolic RBR ubiquitin ligase protein linked with Parkinson’s disease, is a hallmark of mitophagy [23]. It is recruited to mitochondria where it ubiquitinates several target proteins on the outer mitochondrial membrane (OMM) [24, 25]. The mitochondria are among the key organelles that mediate antiviral signaling. Therefore it is very likely that the mitochondrial surrounding environment, polarization status and ubiquitin abundance at OMM can significantly affect the signal transduction induced by the PAMP-PRR interaction. Hence, we reasoned that Parkin via its E3-ligase activity may affect mitochondria-associated antiviral signaling. In this study, we explored the role of Parkin in mitochondria-mediated antiviral signaling in HBV expressing cells. HBV-induced mitochondrially-localized Parkin interacts with MAVS and causes its ubiquitination. We further show that Parkin recruits LUBAC to the mitochondria, which leads to the enrichment of M-1 linked polyubiquitin chains on MAVS which disrupts its interaction with downstream TRAFs and abrogates IRF-3 activation. Parkin has been previously shown to modulate the LUBAC activity [26] and LUBAC is also reported to abrogate MAVS signaling via disruption of MAVS-TRAF3 [27] or TRIM25 [28]. This study also revealed an additional pathway demonstrating how Parkin-dependent accumulation of M-1 linked polyubiquitin chains on MAVS affects IRF3 activation in IFN signaling. Altogether, our results highlight the novel role of Parkin as a negative modulator of MAVS-mediated innate immune signaling and unravels how HBV usurps Parkin to cripple the cellular antiviral response.
Mitochondria associated protein, MAVS is a central molecule on which, signals from the various RLRs, which sense viral RNA and DNA converge [29]. HBV suppresses IFN β synthesis both in vivo and in vitro cultured cells infection [30]. Moreover, the HBx, a regulatory protein encoded by HBV is shown to target MAVS-IRF3 signaling and inhibit IFN β production [19]. In agreement with previous reports, we found that HBV expression rendered the cells less responsive to polyI:C (pI:C) as evidenced by reduced interferon-stimulated responsive element (ISRE) activity (Fig 1A). However, Parkin silencing in these cells, restored MAVS/IRF3 signaling (Fig 1B), suggesting that HBV usurps Parkin’s function to abrogate antiviral signaling. In our previous report, we showed that HBV is able to promote Parkin translocation to mitochondria while the HBV genome defective for HBx expression (HBV-ΔX) did not affect Parkin expression or mitochondrial translocation [4]. Huh7 cells transfected with the wild type HBV genome showed reduced induction in ISRE activity with pI:C stimulation compared to the untransfected control. Comparable induction of ISRE activity was observed in control Huh7 and transfected with the HBV-ΔX (HBx defective) genome suggesting that the HBx expression is required for the inhibition of ISRE activity (Fig 1C). Parkin silencing in Huh7 cells transfected with wild type HBV genome restored ISRE activity upon pI:C stimulation (Fig 1C). Similarly, HBx expressing Huh7 cells showed reduced ISRE activity, however silencing Parkin expression was sufficient to restore ISRE activity in these cells (Fig 1C). Altogether these findings confirmed that the ability of HBV/HBx to regulate antiviral signaling is mediated by Parkin via enhanced mitochondrial recruitment, as demonstrated in our previous study [4]. To further substantiate our observations, we evaluated the effect of HBx on IRF3 phosphorylation upon pI:C stimulation. Stimulation with pI:C led to a robust increase in IRF3 phosphorylation in controls cells. In the control cells, a modest (basal) level of Parkin is associated with the mitochondria which may modestly enhances IRF3 phosphorylation during Parkin silencing in control cells. In contrast, pI:C stimulation did not lead to IRF3 phosphorylation in HBx expressing cells (Fig 1D). However, Parkin silencing significantly restored IRF3 phosphorylation upon pI:C stimulation in HBx expressing cells (Fig 1D). Overall, the results presented so far, clearly establish that HBV/HBx expression utilizes Parkin to abrogate IRF3 activation.
Our previous report demonstrated that HBx interacts with Parkin [4] while others have shown that it can interact with mitochondrial MAVS [19]. In order to explore how Parkin mediates its effect on MAVS downstream signaling, we characterized Parkin-MAVS interaction. Upon activation, MAVS recruits multiple E3 ligases to form a functional signalosome. Here, we investigated if Parkin (an E3 ligase) is a part of the MAVS signalosome complex. Cell lysates obtained from HBV replicating cells (HepAD38) were immunoprecipitated using anti-Parkin antibody followed by immunoblotting using anti-MAVS antibody. Parkin was able to co-precipitate MAVS suggesting that Parkin physically associates with MAVS (Fig 2A). The MAVS-Parkin interaction was also confirmed by reciprocal immunoprecipitation (S1 Fig). Similarly, in HBx transfected Huh7 cells using a similar co-immunoprecipitation (co-IP) strategy, MAVS and Parkin interaction was confirmed (Fig 2B). Interestingly, the HBV or HBx expression appeared to further enhance the MAVS-Parkin interaction. Since HBx, Parkin and MAVS, interact with each other, we next reasoned that HBx and Parkin may be a part of MAVS signalasome. We performed co-IP analysis using cells co-transfected with expression vectors encoding HA-Parkin, Flag-MAVS and Flag-HBx. Immunoprecipitation of cell lysates with anti-HA antibody (to immunoprecipitate Parkin) followed by western blot analysis with anti-Flag antibody (which will detect both MAVS and HBx) revealed that Parkin co-precipitates with both MAVS and HBx (Fig 2C). These data suggested that all three proteins (MAVS, Parkin and HBx) interact with each other and therefore Parkin and HBx may likely be a part of the MAVS signalosome. To further validate this interaction, we performed immunofluorescence microscopy to determine the co-localization between HBx, Parkin and MAVS. We observed enhanced co-localization between Parkin and MAVS in cells expressing HBV (Fig 2D). Confocal microscopic analysis of cells transfected with HBx-Flag also displayed prominent co-localization between HBx-Parkin-MAVS (Fig 2E, see white spots). Parkin-MAVS localization index in control and HBx expressing cells was quantified by overlaying red (MAVS) and green (Parkin) channels using the Image J software (Fig 2F). The cells expressing HBx displayed enhanced Parkin expression (Fig 2E and 2G), relatively similar levels of MAVS (Fig 2E and 2H), and enhanced co-localization between Parkin and MAVS (Fig 2E and 2I). Based on these observations, we conclude that HBV expression potentiates Parkin association with MAVS that could be instrumental in suppressing MAVS downstream signaling.
It has also been reported previously that few viruses promote mitophagy to downregulate innate immune signaling, by facilitating the delivery of mitochondria associated antiviral signaling proteome to the lysosomes for degradation [10, 31]. Since Parkin recruitment to mitochondria can also initiate mitophagy, we asked the question if mitophagy affects innate immune signaling. Inhibition of mitophagy did not affect the ISRE activity in HBV expressing cells (S2 Fig) suggesting that the HBV expression abrogates MAVS-IRF3 signaling by mechanism independent of mitophagy. These results suggest that during HBV infection, the effect of Parkin on MAVS downstream signaling is largely independent of its role in mitophagy.
These results directed us to explore in more detail the possible mechanism(s) by which Parkin may mediate its negative influence on MAVS signaling. Parkin ubiquitinates several mitochondrial proteins [25]. Parkin is able to interlink ubiquitin monomers with various lysine residues and transfer on the target protein in many different ways [25]. Since Parkin interacts with MAVS, we first determined the effect of Parkin silencing on the overall ubiquitination status of MAVS and its turnover. HBV expression induced MAVS associated ubiquitin chains to a significant level, which was in agreement to previous report [17]. However Parkin silencing in HBV expressing cells resulted in reduction of overall level of MAVS-associated ubiquitin chains (Fig 3A). Intriguingly in our subsequent experiments using HBV or HBx expression system, we did not observe any Parkin-dependent change in the MAVS expression level or turnover. This observation demonstrated that the Parkin-mediated ubiquitination of MAVS does not target to proteasome for subsequent degradation (Fig 3A, lower panel) contradicting the previous report [17]. We further evaluated the expression levels of endogenous MAVS in HBV-replicating or HBx-expressing cells with and without Parkin silencing (Fig 3B and 3C). The western blot analysis revealed that there was no significant change in MAVS expression levels in HBV- or HBx-expressing cells and Parkin silencing did not affect MAVS turnover (Fig 3B and 3C). Taken together, these data demonstrate that the Parkin can enhance ubiquitin chains on MAVS but does not target MAVS for proteasomal degradation. After confirming the non-degradative nature of MAVS associated ubiquitin chains mediated by Parkin, we next characterized the linkage specificity of MAVS associated ubiquitin chains. There are several types of linkages that polymerize the ubiquitin monomers on the target proteins [32]. These distinct ubiquitin chains can be associated with any target protein or signaling complex covalently or non-covalently [13, 33]. Sometime these chains get anchored to the target protein or modulate the signaling via unanchored associations [34]. Among all the distinct linkages, the proteasomal machinery predominantly recognizes the target protein tagged with K-48 linked ubiquitin chains [35]. Other linkages are destined for other signaling events [36]. In order to further characterize the kind of ubiquitin linkages involved in Parkin-dependent ubiquitination of MAVS, we probed the MAVS immunoprecipitates with the ubiquitin antibodies specific to different types of linkages. The HBV expression enhanced the levels of linear (or M-1) and K-63 linked chains attached to MAVS and this enhancement was significantly affected when the Parkin was silenced (Fig 3D, first 4 lanes at left). On the other hand, the association of K-48 linked chains with the MAVS did not show any difference in Parkin-silenced and control cells which explained our initial observation of no change in the MAVS turnover (Fig 3B and 3C).
To further characterize if the polyubiquitin chains are anchored to MAVS, we incubated the lysates for 30 minutes at 50°C prior to MAVS immunoprecipitation (as shown in the flow diagram at right). The samples were brought to the 4°C and MAVS immunoprecipitation was performed followed by immunoblot analysis with respective ubiquitin linkage-specific antibodies (Fig 3D last 4 lanes at right). It should be noted that heating the lysate at 50°C for 30 minutes followed by MAVS immunoprecipitation at 4°C did not affect the MAVS immunoprecipitation efficiency. In contrast to untreated lysates, in the lysates preheated at 50°C, all types of ubiquitination associated with MAVS were eliminated. It is known that covalent ubiquitination remains unaffected even at 90°C and in this case, treating the cell lysate merely at 50°C dissociated all MAVS associated ubiquitin chains. These data strongly suggest that in HBV expressing cells, the linear (M-1) and K-63 linked polyubiquitin chains associated with MAVS were predominantly unanchored in nature (Fig 3D) and that the accumulation of M-1 and K-63 linked ubiquitin chains on MAVS is Parkin-dependent. The quantification of the different ubiquitination in various conditions are depicted in the Fig 3E, 3F and 3G.
The inhibition of Parkin reduced the MAVS associated M-1 ubiquitin chains. It should be noted that the M-1 (or linear) ubiquitination is only catalyzed by linear ubiquitin assembly complex (LUBAC) [37, 38]. Therefore it became interesting to explore how Parkin, being unable to catalyze the linear ubiquitination can modulate linear ubiquitin chains on MVAS. This also suggests that Parkin may exert its inhibitory effect on MAVS antiviral signaling via LUBAC. We first established a direct link between the LUBAC and HBV mediated suppression of antiviral response by silencing the LUBAC subunits by RNA interference. Similar to Parkin silencing, HBV expressing cells restored the response against pI:C when LUBAC subunits were inhibited. This results confirmed the involvement of LUBAC in HBV mediated suppression of antiviral signaling (Fig 4A). Similar effect was seen on IRF3 activation that further demonstrated that the LUBAC remains a critical factor for inhibition of antiviral response (S3A Fig). Our experiments confirmed that in HBV expressing cells, the Parkin mediated modulation of MAVS signaling is actually mediated via LUBAC activity. We next wondered Parkin and LUBAC are interlinked in HBV expressing cells that contribute to the suppression of antiviral response. We further found that the Parkin co-eluted with the larger subunit of LUBAC, HOIP (which is the main catalytic site of the complex) and concluded that Parkin is an interacting partner of LUBAC. The interaction between LUBAC and Parkin was observed to be enhanced by HBV expression as in the HBV expressing cell, we observed approximately 3 fold higher level of Parkin co-eluted with the LUBAC-IP compared to the control cells (Fig 4B). We presumed that the enhanced interaction between Parkin and LUBAC in HBV replicating cells could modulate the stability of LUBAC subunits. However, the inhibition of the Parkin in HBV expressing cells had no significant effect on the expression level LUBAC subunits that ruled out the possible involvement of Parkin in altering the overall turnover or stability of any of the LUBAC subunits (Fig 4C). Interestingly, by confocal microscopy we found the striking Parkin dependent difference in the distribution pattern of the LUBAC (Fig 4D). In the HBV off condition, most of the HOIP signals were cytosolic as very less overlap between LUBAC and MAVS was seen (Fig 4D, top panels and the top 2 insets). On the other hand, in the hepatocytes expressing HBV, there was an enhanced overlap of HOIP and MAVS signal that raised the possibility that upon HBV replication, LUBAC subunits are recruited to the mitochondria (Fig 4D, middle panels and middle 2 insets). Finally the cells expressing HBV with silenced level of Parkin did not show any LUBAC accumulation on the mitochondria (Fig 4D, bottom panels and 2 insets at the bottom). Pixel depiction of the LUBAC-MAVS co-localization is shown in the adjacent panel. The in-silico analysis (Fig 4E) of the LUBAC subunits by different algorithms (as described previously) [39] however revealed that none of the LUBAC subunit revealed promising probability for inherent localization to the mitochondria (that indicates the requirement of additional proteins(s) for LUBAC to be targeted to the mitochondria). Conclusively, the confocal microscopy revealed that the HBV expression affect the cellular distribution of LUBAC and directs LUBAC subunits to be recruited to the mitochondria where the Parkin remained a key player facilitated this recruitment of LUBAC to the mitochondria as in Parkin silenced HBV expressing cells, the mitochondrial recruitment of LUBAC was drastically reduced. In support, the cells fractionation experiment also strengthened our observation obtained in microscopy and showed that the HBV replication enriched the LUBAC in mitochondrial fractions while the inhibition of the Parkin abolished it (Fig 4F). This observation was further substantiated when we analyzed the level of mitochondria-associated linear ubiquitin chains (mito-M1 Ub). We observed that the mito-M1 Ub chains were significantly enhanced in the mitochondrial fraction in the HBV expressing cells while the inhibition of Parkin significantly eliminated it (S3B Fig). Lastly we confirmed that the Parkin expression supports the HOIP-MAVS interaction in HBV expressing cells that further validates the Parkin involvement in recruiting LUBAC to MAVS (Fig 4G). Altogether, the above experiments substantiated our notion the Parkin facilitates the LUBAC redistribution in HBV expressing cells. We confirmed the Parkin mediates the redistribution of all three subunits of LUBAC and silencing of HOIP, HOIL-1L and Sharpin restored the IRF3 signaling in HBV expressing cells. It should be noted that the silencing of all subunits in control cell (non HBV), had no or modest effect on ISRE signaling which is in agreement to the previous report [14]. However the same silencing had drastically restore the ISRE activity in HBV expressing cells that strongly reconcile the discrepancy over the LUBAC’s role in modulating IRF3 activation.
MAVS signaling includes its interaction or recruitment of many downstream partner molecules like TRAFs (TNF receptor associated factors) [14]. Notably the TRAF3 has been shown to mediate the IRF3 activation by direct association with MAVS [40, 41]. We observed that MAVS-TRAF-3 interaction was inhibited in HBV expressing cells (Fig 5A), whereas in Parkin and HOIP silenced cells, this interaction was restored. The recent advancement has led to expand our understanding in MAVS-TRAFs interaction and further revealed that not only TRAF3 but other TRAFs like TRAF2,5 and 6 also play a role in IRF3 activation. Therefore we evaluated the effect of HBx on the various TRAFs and how their interaction is modulated by Parkin or LUBAC at endogenous level. We observed that the expression of HBx inhibited the interaction of MAVS with TRAF 2,3 5 & 6 and this interaction was restored when Parkin of HOIP was inhibited (Fig 5B). This experiment convincingly demonstrates that the HBV expression is able to disrupt the MAVS signalasome and utilizes Parkin or LUBAC for this disruption. To further substantiate our hypothesis that in HBV expressing cells, the MAVS signalosome is disrupted in Parkin/LUBAC dependent fashion, we used an in vitro reconstitution assay described previously [14, 42]. We observed that the Parkin/LUBAC affected the MAVS signaling in VSV (vesicular stomatitis virus)-infected cells (S4 & S5 Figs). We, next performed in vitro reconstitution assay using purified mitochondrial preparation from VSV infected control and HBx-expressing cells respectively transfected with non-targeting, Parkin, and LUBAC specific siRNAs. VSV infection was used to prime the mitochondrial antiviral signaling pathway. Cytosolic fractions were prepared from control cells as described in the schematics (Fig 6A–6F). The mitochondria from VSV-infected control cells were able to stimulate IRF3 phosphorylation in the in vitro reaction when mixed with the cytosol obtained from the control cells. However mitochondria from VSV-infected cells expressing HBx did not promote significant level of IRF3 phosphorylation. Interestingly, the mitochondria prepared from Parkin or LUBAC silenced-HBx expressing cells responded better (Fig 6G). We further analyzed TRAF3 recruitment in a similar in-vitro reconstituted assay using mitochondrial and cytosolic fractions, as described above. The mitochondrial fractions were mixed and incubated with the purified cytosolic extract of the control cells expressing HA-TRAF3. After the in vitro reaction, the mixture was centrifuged to pellet the mitochondria, which were subjected to western blot analysis to analyze TRAF3 binding. It was observed that the mitochondrial fraction from HBV replicating cells showed reduced recruitment of cytosolic TRAF3 (Fig 6H). However the recruitment of TRAF3 was restored when either Parkin or LUBAC subunits were silenced (Fig 6H). This in vitro analysis further confirms the fact that mitochondria from the HBV/HBx expressing cells exhibit reduced recruitment of effector molecules (such as TRAF3). These results suggest that HBV/HBx disrupts MAVS signalasome that is modulated in Parkin/LUBAC-dependent manner. The proposed mechanism for these combined events is summarized in S6 Fig.
Parkin’s role in cellular events other than mitophagy remains largely unexplored. The role of Parkin in the activation of classical NF-kB pathway [26] and our observation that Parkin serves as negative regulator of MAVS signaling are among the few examples of mitophagy-independent functions of Parkin. Previous report pointed out that in Drosophila, Parkin impairment is associated with the induced level of interferon stimulated genes (ISGs) [43]. On the other hand, the linear ubiquitination by LUBAC could negatively regulate MAVS signaling [27]. However, how LUBAC gets activated and recruited to antiviral signaling complex was previously unknown [44]. Our observations explained Parkin-dependent recruitment of LUBAC to the mitochondria can disrupts MAVS signaling via enrichment of M-1-linked ubiquitin chains associated with MAVS. The LUBAC was initially discovered as a dimeric enzyme that consists two different RBR E3 ligases i.e. HOIP and HOIL-1L [37]. Subsequently SHARPIN was identified as a third component of LUBAC [45]. It should be noted that the HOIP is the main catalytic subunit of LUBAC that remains in auto-inhibited state. The interaction of HOIP with HOIL-1L or Sharpin releases this auto-inhibition and makes this complex active. During the formation of M-1 ubiquitin chains, the ubiquitin transfer proceeds via thioester intermediate [38]. It should be noted that the role of LUBAC in anti-viral signaling has been controversial with inconsistencies in the proposed mechanism [14, 27, 28]. By using HBV replication model, we explain that as such, LUBAC has very modest effects on anti-viral signaling as reported previously [14], however when recruited to the mitochondria, LUBAC turns into a strong modulator of MAVS-mediated innate immune signaling. This conclusion is based on the fact that the presence of LUBAC was ineffective when Parkin was silenced. The reported role of Parkin in the activation of classical NF-kB pathway establishes its involvement in other cellular events beyond its role in selective-autophagy of mitochondria [26]. Under cellular stress, Parkin activates LUBAC and increases the linear ubiquitination of NF-κB essential modulator (NEMO), which is required for NF-kB activation [26]. On the other hand accumulating evidence clearly indicate that the MAVS, upon activation mediates the signaling of IRF3 and NF-kB. However the involvement of various factors separately modulating the IRF3 and NF-kB signaling pathways through MAVS, still remain mysterious. Recent investigation has revealed that the domain III (aa401-450) is specifically essential for IRF3 activation while domain I and II are required for NF-kB signaling [46]. Intriguingly, our observations suggest that the enrichment of linear-linked ubiquitin chains on the mitochondria-associated MAVS disrupt its downstream signaling. This implicates Parkin’s dual involvement in the regulation of cellular antiviral and inflammatory responses. It further signify our finding and explains that although LUBAC is critical for NF-kB signaling, it can potentially inhibit IRF3 activation if translocated to mitochondria. How the cells fine- tune the balance between these two contrasting roles of Parkin and its consequences on antiviral defense and stress response (via NF-kB) remains to be determined.
The role of unanchored polyubiquitin chains has been investigated in multiple ways. For instance, in the presence of K-63 unanchored polyubiquitin chains, RIG-I is activated and mediates the conversion of MAVS into prion like structure [47]. It should also be noted that the activation of RIG-I, not only requires the RNA but also needs specific binding with the K-63 polyubiquitin chains [48]. In addition, the unanchored K-48 polyubiquitin chains can activate IKKε via TRIM6 and subsequently activate STAT1 [49]. In contrast, our study reveals that the accumulation of unanchored M-1 polyubiquitin chains can negatively affect MAVS signaling. Our results do not rule out the possibility that these unanchored polyubiquitin chains associated with MAVS may be due to other MAVS interacting proteins with covalently attached poly ubiquitin chains. The role of K-48, K-63, and linear ubiquitin linkages in regulating innate immune signaling is well documented [32, 37, 38, 50]. A recent study demonstrates that Parkin can induce the accumulation of various lysine linked polyubiquitin chains including K-48, K-63, and M-1 on the mitochondria [51]. Parkin, as such is unable to catalyze the M-1 linkage and so far it has been puzzling how Parkin can modulate the M-1 ubiquitin chains on the mitochondria. Our study further explains that Parkin is able to accumulate M-1 ubiquitin chains on mitochondria through LUBAC recruitment. Interestingly the LUBAC inhibition reduced the enrichment of M-1 linked polyubiquitin chains associated with MAVS or mitochondria. We therefore concluded that the Parkin-dependent enrichment of M-1 linked ubiquitin chains on MAVS could be a LUBAC mediated consequence. Interestingly, inhibition of Parkin or LUBAC components restored MAVS signaling in HBV expressing cells. These observations also suggest that Parkin-dependent enrichment of M-1 polyubiquitin chains on mitochondria negatively modulates MAVS-IRF3 signaling. Our analysis with in vitro reconstituted assays using Vesicular Stomatitis Virus (VSV)-primed mitochondria revealed that the accumulation of M-1 linked polyubiquitin chains on MAVS attenuates MAVS signaling by perturbing MAVS interaction with the downstream effector proteins like TRAF3 thereby inhibiting IRF3 phosphorylation and IFN β production.
We envisaged that Parkin localized to the mitochondria may influence the orchestration of mitochondria-based antiviral signaling during HBV infection. HBV is considered a stealth virus due to its ability to evade host immunity and cause/establish chronic infection [30]. Previous reports establish that HBV cripples RIG-I-MAVS signaling [17–19] and a recent study shows that the RIG-I senses HBV pgRNA to stimulate IFN λ production but not IFN β [15]. Notably only peroxisomal MAVS stimulates IFNλ production via IRF-1 [52]. From this, it can be concluded that despite the activation of RIG-I and MAVS, the mitochondrial MAVS signaling that stimulates IFN β production is suppressed in HBV infection [15, 52]. This conclusion further advocates the likely role of Parkin in suppressing mitochondrial MAVS signaling via LUBAC recruitment to mitochondria. We observed that Parkin interacts with MAVS and HBV further potentiates this interaction. Moreover, Parkin silencing restores IFN β synthesis in HBV expressing cells upon stimulation with RIG-I agonist pI:C. These findings establish the novel role of Parkin in influencing MAVS signaling during HBV infection. Our investigations provide molecular mechanism(s) of HBV-induced suppression of innate immunity. This study opens a new paradigm involving Parkin-dependent spatio-temporal modulation of LUBAC activity and elucidates how viruses manipulate host factors to regulate antiviral signaling.
The HepG2, Huh7, HEK293 cells were obtained from ATCC (American Type Culture Collection) and HepAd38 cells were a kind gift from Dr. C. Seeger, Philadelphia [53]. The cells were maintained as described previously [4]. The pHBV1.3mer and pHBV-ΔX plasmid DNAs encoding wild-type HBV genome and HBx-deficient HBV genome, respectively, were a kind gift from Dr. Jing-hsiung James Ou (University of Southern California). The plasmids pHBx-flag (Addgene# 42596) [54], FLAG tagged LUBAC subunits HOIL-1L (Addgene#50016) HOIP (Addgene# 50015) [55], HA-TRAF3 (Addgene#44032) [56] and MAVS-FLAG were used for in vitro transfections. The Lyovec pI:C (invivogen) and the reporter assay for ISRE (Interferon stimulated regulatory element) luciferase was used. The GFP tagged Vesicular stomatitis virus (VSV) was kindly provided by Dr. Juan de La Torre (The Scripps Research Institute La Jolla, CA). For the preparation of cells expressing HBx-FLAG under the control of tetracycline promoter, the coding region of HBx-FLAG (from Addgene# 42596) was inserted into pTRE2-Hyg (Clonetech) and transfected into the cells stably expressing rTA. The selected clone expressing HBx-FLAG were maintained in 0.5mg/ml G418 and hygromycin. For western blot and immunoprecipitation assays, rabbit anti-Parkin (abcam), mouse anti-parkin (abcam), mouse anti-MAVS (Santa Cruz), anti-HOIP (abcam), and anti-HOIL-1L (abcam) antibodies were used as per the manufacturer’s instructions. The anti-K48 and K-63 antibody (Cell Signaling); Mouse anti-linear Ubiquitin LUB9 (Lifesensors) were used for characterization of MAVS associated ubiquitin chains.
To conduct laser scanning confocal microscopy, the cells grown on coverslips were transfected with the indicated plasmid DNAs followed by immunofluorescence assay, as described previously [4]. Images were visualized under a 60x or 100x oil objectives using an Olympus FluoView 1000 confocal microscope. Quantification of images was conducted with ImageJ, Adobe and MBF ImageJ softwares.
Small interfering RNA (siRNA) pools used in this study were siGENOME SMARTpool for Parkin, nontargeting #1 control (NT), HOIL-1L, SHARPIN and HOIP from Dharmacon. The cells were transfected with siRNA (50 nM) for the indicated times using Dharma- FECT 4 transfection reagent according to the manufacturer’s instructions (Dharmacon).
Immunoprecipitation and subcellular fractions for analyzing LUBAC enrichment and abundance of linear ubiquitin chains on the mitochondria were prepared as per the previous reports [14, 27, 42]. All procedures were carried out at 4°C unless otherwise specified. The cells were homogenized in hypotonic buffer containing 10 mM Tris-Cl [pH 7.5], 10 mM KCl, 0.5 mM EGTA, 1.5 mM MgCl2, and EDTA-free protease inhibitor cocktail. The homogenates were centrifuged at 1000x g for 5 min to pellet nuclei and unbroken cells. The supernatant was subjected to centrifugation at 5000 x g for 10 min to separate crude mitochondrial pellet from cytosolic supernatant. Mitochondrial pellet was washed once with Mitochondria Resuspension Buffer (MRB) (20 mM HEPES-KOH [pH 7.4], 10% glycerol, 0.5 mM EGTA, and EDTA-free protease inhibitor cocktail) and resuspended in MRB buffer. After centrifugation at 10,000 x g for 15 min, the supernatant was used in all assays. For VSV infection, the cells were infected with VSV for 15 hours. Most of the assays were carried out using mitochondrial and cytosolic extracts. Various preparations of mitochondrial fractions from different sets, were mixed with the cytosolic extracts of the control cell or the cells expressing HA-TRAF3 and incubated in the presence or absence of ATP for 60 minutes at 30°C. For in vitro IRF3 phosphorylation assay, the reaction mix was resolved on native or SDS PAGE and probed for total IRF3 (Cell Signaling) or p396-IRF3 (abcam). For TRAF3 recruitment assay, the reaction mix was centrifuged at 5000xg for 10 minutes and pelleted crud mitochondria were washed 3 times with cold assay buffer. After washing, the mitochondrial fractions were loaded on SDS-PAGE and the level of TRAF3 (HA) recruited to the mitochondria were analyzed by using anti-HA antibody in western blot.
For immunoprecipitation, the cells were transiently transfected with the indicated expression plasmids. Cells were harvested and immediately lysed in a 1% Triton X-100 lysis buffer (20 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1% Triton X-100, 10% glycerol, 0.1% protease inhibitor cocktail, 1 mM PMSF, 1 mM Na3VO4, 5 mM NaF, 1 mM DTT, 10 mM NEM). Immunoprecipitation was carried out by using 2 milligram of WCL and 5 μg of antibody. The immune complex was captured by affinity purification using protein A/G coupled sepharose beads (GE healthcare) at 4°C with constant rotation. Following five washes with supplemented lysis buffer, samples were denatured in 1x loading dye, separated by SDS-PAGE and transferred to a nitrocellulose membrane (Bio-Rad) and the co-immunoprecipitate were analyzed by using various antibodies as described in the figure legends.
All the data are representative of three independent sets of experiment. For each result, error bars represent the mean ± s.e.m. from at least three independent experiments. Statistical significance was performed with two-sided unpaired Student’s t-test.
|
10.1371/journal.pntd.0001825 | Proof-of-Principle of Onchocerciasis Elimination with Ivermectin Treatment in Endemic Foci in Africa: Final Results of a Study in Mali and Senegal | Mass treatment with ivermectin controls onchocerciasis as a public health problem, but it was not known if it could also interrupt transmission and eliminate the parasite in endemic foci in Africa where vectors are highly efficient. A longitudinal study was undertaken in three hyperendemic foci in Mali and Senegal with 15 to 17 years of annual or six-monthly ivermectin treatment in order to assess residual levels of infection and transmission, and test whether treatment could be safely stopped. This article reports the results of the final evaluations up to 5 years after the last treatment.
Skin snip surveys were undertaken in 131 villages where 29,753 people were examined and 492,600 blackflies were analyzed for the presence of Onchocerca volvulus larva using a specific DNA probe. There was a declining trend in infection and transmission levels after the last treatment. In two sites the prevalence of microfilaria and vector infectivity rate were zero 3 to 4 years after the last treatment. In the third site, where infection levels were comparatively high before stopping treatment, there was also a consistent decline in infection and transmission to very low levels 3 to 5 years after stopping treatment. All infection and transmission indicators were below postulated thresholds for elimination.
The study has established the proof of principle that onchocerciasis elimination with ivermectin treatment is feasible in at least some endemic foci in Africa. The study results have been instrumental for the current evolution from onchocerciasis control to elimination in Africa.
| The control of onchocerciasis, or river blindness, is based on annual or six-monthly treatment with ivermectin of populations at risk. This has been effective in controlling
the disease as a public health problem but it was not known whether it could also eliminate infection and
transmission to the extent that treatment could be safely stopped. Many doubted that this was feasible in Africa. A
study was undertaken in three hyperendemic onchocerciasis foci with seasonal transmission in Mali and Senegal
where treatment has been given for 15 to 17 years. As a result of this treatment, infection and transmission levels
had fallen everywhere below postulated thresholds for elimination. Treatment was therefore stopped in each
focus. Follow-up evaluations up to 5 years after the last treatment showed no evidence of recrudescence after
stopping treatment but instead a consistent decline in infection and transmission levels, reaching zero in two
sites. The study has established the proof-of-principle that onchocerciasis elimination with ivermectin treatment is
feasible in at least some endemic foci in Africa. The results of the study have greatly contributed to the current
evolution from onchocerciasis control to elimination in Africa.
| Onchocerciasis control is currently nearly exclusively based on large-scale treatment with ivermectin [1]. Annual or six monthly treatment of all eligible members of high risk communities effectively controls the disease as a public health problem [2]–[4], and following the donation of the drug free of charge for as long as needed by the manufacturer in 1987, large-scale ivermectin treatment programmes have been established in nearly all endemic areas in Africa where over 99% of all cases in the world are found [5]. The African Programme for Onchocerciasis Control (APOC) was established in 1995 to support the establishment of community directed treatment with ivermectin (CDTi) in all areas where onchocerciasis was a public health problem in 19 African countries [6]. The CDTi strategy has been very successful in ensuring sustained high treatment coverage and by the year 2010 some 75 million people at risk were treated annually with ivermectin in the APOC countries [7]. The remaining 11 endemic African countries had been supported since 1975 by the Onchocerciasis Control Programme in West Africa (OCP) [8]. OCP was based on vector control to which ivermectin treatment was added in 1988. Following the successful completion of the vector control activities and the closure of the OCP programme in 2002, large-scale ivermectin treatment using the CDTi strategy has been maintained by these countries themselves.
As a result of these sustained ivermectin treatment activities, nearly all endemic areas in Africa are under annual ivermectin treatment and the control of onchocerciasis as a public health problem has already been achieved in the majority of these areas [9].
Following this success, the principal question became how long these treatments needed to continue and whether in the long term it would ever be possible to eliminate onchocerciasis infection and transmission with ivermectin treatment so that treatment could be safely stopped. Epidemiological models predicted that elimination would be feasible in the long term [10], and in the Americas where most onchocerciasis foci are small and most vectors relatively inefficient, elimination has been set as the objective by the Onchocerciasis Elimination Program for the Americas (OEPA) [11], [12]. However, in the absence of empirical evidence on the feasibility of elimination in Africa, most experts doubted that elimination would be possible in the African continent where onchocerciasis is highly endemic over vast areas, and where the vectors are highly efficient and some vector species can migrate over long distances [13]–[15].
In order to study this question, a longitudinal study was undertaken in three onchocerciasis endemic foci in Mali and Senegal. The three foci were among the first areas where large scale ivermectin treatment was started in Africa and by 2006 they had received 15 to 17 years of ivermectin treatment. Interim epidemiological evaluations had indicated that the prevalence of infection had fallen to very low levels [16]. Because of the duration of treatment and the promising interim evaluation results, it was considered that if elimination with ivermectin treatment would be feasible in endemic foci in Africa, these would be the foci where this could be first demonstrated. The aim of the study was to determine if after 15 to 17 years of ivermectin treatment onchocerciasis infection and transmission levels had fallen so low that transmission would be unlikely to sustain itself, and then to test this hypothesis by actually stopping treatment and undertaking follow-up surveys for another 3 years to confirm that was no recrudescence in infection and transmission after cessation of treatment.
The study began in 2006 and was completed in 2011. The first results of the study covering the period 2006 to mid 2008 have been reported by Diawara et al [17]. The final results of the study, including the full results for the 3-year follow-up evaluations after cessation of treatment, are reported here.
The study was undertaken in three onchocerciasis foci along the River Bakoye in Mali, the River Gambia in Senegal, and the River Faleme on the border of the two countries (figure 1). These three areas were part of the Western Extension area of the OCP where onchocerciasis control has been exclusively based on ivermectin treatment which started in 1988–1989. According to skin snip surveys undertaken by the OCP before the start of control, in each of these three foci there were hyperendemic villages, i.e. villages with a prevalence of microfilaridermia ≥60% or a Community Microfilarial Load (Cmfl, the geometric mean number of microfilariae per skin snip among adults aged 20 years and above) >10 microfilariae per skin snip (mf/s) [18]–[20]. In the River Gambia focus, 8 out of 22 surveyed villages had a Cmfl >10 mf/s (range 12.0 to 48.1 mf/s) [21]. In the River Bakoye focus 5 out of 11 surveyed villages had a Cmfl >10 mf/s (range 10.2 to 21.6 mf/s) and in the River Faleme focus this was the case for 3 out of 27 surveyed villages (range 13.3 to 21.0 mf/s) [22]. The rural population of the three foci has about the same size with 20,000 to 30,000 people living in 75 to 94 villages per site. In the R. Gambia focus there is also one town with a population of about 18,000.
The onchocerciasis vectors in the study areas are the savanna vectors Simulium sirbanum and Simulium damnosum s.s. and transmission is limited to the rainy season when the rivers flow from about July to December. All three areas are isolated with respect to long-distance migration of the Simulium vectors except for the first few weeks of the rainy season. During the dry season, the rivers do not flow and there are no blackflies. At the beginning of the rainy season, when the Inter-tropical-conversion-zone (ITCZ) moves to the north, the breeding sites are reinvaded by simuliids from the south (mainly S. sirbanum) that migrate with the prevailing winds and start the repopulation of the breeding sites [15], [23], [24]. After a few weeks, when the winds change, this long distance migration stops and the vector population becomes purely local with virtually no migration from outside or from neighboring river basins. All river basins involved in this migration pattern are either free from onchocerciasis or under large-scale ivermectin treatment since 1990. For the R. Bakoye, S. dieguerense has also been reported but this is a non-migratory Simulium species that only plays a local role in onchocerciasis transmission [25].
Along all three rivers there are onchocerciasis endemic villages downstream of the study areas but their endemicity levels are generally lower and they are all covered by the same national ivermectin treatment programs of Mali and Senegal. The neighboring river basins are also endemic for onchocerciasis and undergoing ivermectin treatment. Hence, the three study areas cannot be considered completely isolated areas, but rather as the most endemic sections of onchocerciasis zones along three rivers that are fully covered by the national ivermectin treatment programs.
Ivermectin treatment was given annually in the R. Bakoye and R. Faleme, and at six monthly intervals in the R. Gambia making this the only onchocerciasis endemic area in Africa where six monthly treatment with ivermectin has been given for more than 10 years. The months of treatment were April or May, just before the rainy season, in order to optimize the impact of treatment on transmission. In the R. Gambia there was a second round of treatment in October or November of each year. In the R. Gambia the first round of treatment was given in 1988 and in the other two foci in 1989. The treatment programs were introduced stepwise, covering only the most endemic villages during the first year and gradually expanding the coverage to all villages over the next few years. Hence the number of years that each village has received treatment by the time of the study ranged from 14 to 19 years. We have defined the number of years of ivermectin treatment for each study area as the number of years that all first line villages had been under ivermectin treatment [17]. For the R. Gambia this was 17 years, for the R. Faleme 16 years and for the R. Bakoye 15 years.
During the first few years (1988 to 1991) treatment coverage was not yet satisfactory at 59% to 69% of the total population, but from 1992 onwards the reported treatment coverage was high at 75% to 89% of the total population (corresponding to some 89% to 100% of eligibles). The only exception was the year 1997 when there was a temporary drop in coverage following an abrupt change in drug delivery policy.
More detailed information on the study sites and ivermectin treatment history is provided in the first article on the study [17]
Onchocerciasis elimination is defined as the reduction of local onchocerciasis infection and transmission to such low levels that transmission can no longer sustain itself and treatment can be safely stopped without risk of recrudescence of infection and transmission [26].
To assess whether elimination has been achieved in the three study areas, the study was designed in three phases (figure 2). The aim of the first phase was to undertake a detailed assessment of onchocerciasis infection and transmission levels after 14 to 17 years of treatment. Skin snip surveys were to be undertaken in a stratified random sample of some 40 villages in each study site, and transmission monitored for a full transmission season through entomological evaluations in 4 to 6 fly-catching points per study site. If the observed infection and transmission levels in a study site were below predefined thresholds (see section on indicators below), phase 2 would start in which treatment would be stopped in a test area of 5–8 villages located around one of the catching points in the study site. The effect of stopping treatment on infection and transmission would be evaluated by epidemiological surveys 20 to 22 months after the last treatment in the test villages, and by entomological evaluation in all catching points during another full transmission season. If there was no recrudescence of infection and transmission in the test area, phase 3 would start in which treatment would be stopped throughout the study site and infection and transmission monitored for another two years in all sample villages and catching points. Detailed information on the dates when the different treatment and evaluation activities were undertaken in each of the study foci is provided in the results section.
In the R. Bakoye and the R. Gambia foci, the study has followed the original design. In the R. Faleme focus one step was added to phase 2 to collect information from two additional test areas in the southern part of the focus, where the phase 1 results were less clear, before making the decision to proceed to phase 3 and stop treatment throughout the focus. Hence, in the R. Faleme site the study has lasted one year longer than in the other two sites (figure 3).
Skin snip surveys were done in all selected evaluation villages. In each village, all persons above the age of 1 year who agreed to participate (or whose parent agreed for them to participate in the case of children) were examined for onchocerciasis infection. The surveys used established skin snip examination methods in which the national onchocerciasis teams have been trained by the OCP. Two skin snips were taken from the iliac crests with a 2 mm Holth corneoscleral punch and microscopically examined after incubation for 30 minutes in distilled water (and a further 24 hours in saline for negative skin snips) for the presence and number of O. volvulus microfilariae [27]. The numbers of microfilariae were counted and the results recorded for each person examined. Basic information on the migration history for each person during the last 10 years before the survey was also collected. The prevalence of mf was estimated as the percentage of examined persons who had microfilariae in at least one skin snip. Confidence intervals for the prevalence were calculated using the Clopper-Pearson method.
During each year, a detailed entomological evaluation was done throughout the transmission season in order to determine the levels of O. volvulus transmission. Four vector catching points were selected for the R. Bakoye and R. Gambia and six for the R. Faleme which covers a larger area in two countries. The location of the catching points is shown in figure 4. Every week, 3 days of capture were carried out at each catching point during the transmission period which generally covers 5 to 6 months per year (July/August to November/December). Flies were collected using the method of bulk catches with a team of 3 to 4 fly catchers working from 7 AM to 6 PM. Each daily catch was preserved in 80% alcohol and sent to the DNA laboratory of the Multi-Disease Surveillance Centre (MDSC) in Ouagadougou, Burkina Faso [28]. In the laboratory, the flies were rinsed with distilled water, the heads separated from the bodies and sorted in lots for DNA extraction. The purified DNA was used as a substrate in an O-150 (an Onchocerca-specific DNA sequence) PCR, and the resulting product classified by hybridization to the O. volvulus-specific oligonucleotide probe OVS-2 [29]. A computer program (Poolscreen) was used to translate the molecular biology data obtained from screening pools of 300 flies into an estimate of the infectivity rate in the vector population [30].
The two main indicators of onchocerciasis infection and transmission used in the present study are the prevalence of microfilariae in the skin in the human population and the vector infectivity rate as measured by the number of flies with O. volvulus L3 (infective) larvae in the head per 1,000 flies (FLH/1,000). Based on model predictions as well as large-scale experience in the OCP, provisional thresholds have been defined for these indicators below which the remaining infection and transmission levels are so low that they would die out naturally, even in the absence of any intervention, and when treatment can therefore be safely stopped without risk of recrudescence [17]. The thresholds for the prevalence of infection were defined as a microfilarial prevalence <1% in 90% of sample villages, and a prevalence <5% in 100% of sample villages. The threshold for vector infectivity was defined as 0.5 FLH per 1,000 flies. To ensure that a sample with 0 FLH would imply that the infectivity rate was with 95% confidence below the threshold of 0.5 FLH per 1000 flies, a minimum of 3900 flies was to be analyzed per catching point [30].
The above thresholds were provisional thresholds to guide decision making and analysis in the current study. One of the objectives of the study was to review these thresholds, and revise them as required, based on the final study results.
Ethical review and clearance of the research protocol, research instruments and informed consent procedures were obtained from the national ethical review boards of the ministries of health in Mali and Senegal, and continuing ethical review was ensured by the ethical review committee of the World Health Organization. Community meetings were held in all villages to explain the research objectives and procedures, and the right of each individual to decide whether to participate in the examinations or not. Before each examination, each individual who had voluntarily come to the examination point and agreed to participate signed, or put a thumb print if not literate, on the examination form to indicate consent. For children, one of the parents or the responsible guardian would sign the examination form. The use of community meetings to discuss the research project and the right of individuals to refuse participation in the examination was considered the most culturally appropriate and effective method for providing the necessary information to community members, and this approach was approved by both the national ethical review boards and the WHO ethical review committee. Community leaders approved the use of the selected locations on the river banks as vector catching points.
The results for phase 1 and phase 2 of the study have been reported previously by Diawara et al [17]. In this article we report the results of the final evaluations during phase 3 after cessation of ivermectin treatment throughout the study areas.
In the R. Bakoye focus, the last round of ivermectin treatment was given in May 2006 in the villages in the test area, and in May 2007 in all other villages in the focus.
During phase 3, skin snips surveys were done in the same 40 villages that had been surveyed during phase 1 before the cessation of treatment. Of these 40 villages, 20 were surveyed in February 2009 during phase 3A, 21 months after the last treatment. Another 20 villages were surveyed in May 2010 during phase 3B, 36 months after the last treatment in 15 villages from the main area and 48 months after the last treatment in 5 villages that were located in the test area where treatment was stopped one year earlier.
The results of the epidemiological surveys are shown in figure 5 and table 1. As reported previously, the overall prevalence of mf had fallen from a precontrol prevalence of 43.4% to 0.26% after 14 years of treatment. The phase 3 results show that after cessation of treatment, the prevalence of mf continued to decline. In phase 3A, only 2 mf positives (0.05%) were detected out of 3739 persons examined. It concerned one male of 31 years who had a relatively high mf density with 34 and 81 mf in the skin snips from the left and right iliac crests respectively, and who reported to have been treated only twice, once in 1990 and once in 2000. The other was a male of 55 years with 0 and 10 mf in the two skin snips who reported to have been treated only three times in 1994, 1997 and 2007. These two individuals had not been examined during the phase 1 surveys. In phase 3B no mf positives were detected among the 3520 persons examined 3 to 4 years after the last treatment.
The entomological evaluations during phase 3A were done from September to December 2008, 16 to 19 months (about 1.5 years on average) after the last treatment in the main area and about 2.5 years on average after the last treatment in the test area. During phase 3B the entomological evaluations were done from August to December 2009, i.e. on average about 2.5 and 3.5 years after the last treatments in the main and test areas respectively. The results are given in table 2. More than 100,000 flies were collected and examined during phase 3A and phase 3B, but not a single infective fly was detected. The upper limit of the confidence interval for the infectivity rate remained for each catching point and for each evaluation year below the threshold of 0.5 F3H/1000 flies.
In the R. Gambia focus, where treatment has been given at six monthly intervals, the last round of ivermectin treatment was given in May 2006 in the villages in the test area, and in all other villages in May 2007.
For phase 3, skin snip surveys had been planned for 40 villages. However, as reported previously, the population in the study areas was becoming increasingly reluctant to submit to the skin snip examination. This reluctance was particularly strong in the R. Gambia focus where the total population of 6 villages refused to participate in the final survey. Hence only 34 villages were surveyed, 18 in February 2009 during phase 3A, 21 months after the last treatment, and another 16 villages in May 2010 during phase 3B, 32 months after the last treatment in 10 villages and 48 months after the last treatment in 6 villages from the test area.
Figure 6 and table 3 show the results of the epidemiological surveys. During the treatment period the prevalence of mf had fallen from a precontrol prevalence of 49.5% to a prevalence of 0.06%, and the phase 3 results showed that the prevalence has remained at this very low level after the cessation of treatment. Only two mf positives were detected out of 1561 examined during phase 3A. Both were adult males (32 and 50 years of age) who had very low mf counts of 1 and 3 mf per skin snip. The 32 year old male reported to have been treated irregularly. The 50 year old male disappeared after the examination and could not be interviewed about his treatment history. During phase 3B there were no mf positives among the 1540 persons examined 32 to 48 months after the last treatment.
The entomological evaluations during phase 3A were undertaken from August to September 2008, about 1.5 years on average after the last treatment in the main area, and about 2.5 years on average after the last treatment in the test area. During phase 3B the entomological evaluations were undertaken exactly one year later, i.e. about 2.5 to 3.5 years after the last treatment in the main and test areas respectively. The results were again very clear: more than 150,000 flies were collected and examined during phase 3 and no infective fly was detected (table 4). The 95% confidence interval of the infectivity rates were for all catching points below the threshold of 0.05 F3H/1000 flies.
As mentioned in the methodology section above, the study design for the R. Faleme focus was modified because the phase 1 results for this focus did not fully meet the provisional criteria for stopping treatment. Phase 2 was therefore extended by one year during which two additional test areas were introduced in the south of the focus where treatment was stopped during phase 2B. When the follow-up results for these additional test areas showed no increase in infection and transmission levels during the first year after cessation of treatment, the focus was also moved into phase 3 and treatment was stopped in all villages. Hence, the cessation of treatment in the R. Faleme went in three steps: treatment was first stopped in the initial test area where the last treatment was given in May 2006, then in the additional two test areas where the last treatment was given in May 2007, and finally in all the remaining villages which received their last treatment in May 2008 in Mali and in October 2008 in the Senegal villages where the treatment was a few months delayed because of late arrival of ivermectin in the country during that year.
During phase 3 skin snip surveys were done in a total of 57 villages in the R. Faleme focus: 20 villages during phase 3A and 37 other villages during phase 3B. Because of the special epidemiologically situation in this focus, the number of survey villages for phase 3B was increased and all villages previously surveyed in phase 2 in the three test areas were included. During phase 3A, surveys were done in May 2010 in 20 villages that were all located outside the three test areas and which had their last treatment 19 to 24 months earlier. The 37 villages surveyed during phase 3B consisted of villages from all three groups: villages from the main area where the survey was done 31 to 36 months after the last treatment, villages from the additional test areas which were surveyed 48 months after the last treatment, and villages from the first test area which had not been treated for 60 months.
A summary of the epidemiological evaluation results is given in figure 7 and table 5. The prevalence of mf had fallen from a precontrol overall prevalence of 34% to a prevalence of 0.84% after 15 years of annual treatment. However, of the 44 villages surveyed during phase 1, 80% had a prevalence of mf <1% and 91% a prevalence of mf<5%. This did not meet the provisional thresholds for stopping treatment of at least 90% and 100% of villages in these two categories, and this was the reason for proceeding more prudently with stopping treatment in this focus and for the introduction of two additional test areas. When treatment was finally stopped, this was followed by a significant decline in the overall prevalence of mf to 0.13% in phase 3A, 1.5 to 2 years after the last treatment, and 0.07% in phase 3B, 2.5 to 5 years after the last treatment. The six persons who were mf positive (3 males and 3 females between the age of 19 and 49 years) had all low mf densities between 1.5 and 12 mf per skin snip. Four of them reported to have been treated irregularly with one having received only one treatment in 2006. The other two mf positives reported to have participated regularly in treatment. Five of the mf positives had been examined previously in phase 1 or phase 2, and four of those were already mf positive before the cessation of treatment. The fifth person, a 19 years old male, had been skin snip negative during two previous surveys but was now skin snip positive although with a very low mf density of 2 mf in the snip from the right iliac crest and 1 mf in the snip from the left iliac crest. This person was from a village in the south-west of the focus (figure 7) and had been treated irregularly. The distribution of the prevalence of mf by village showed that 95% of the villages had a prevalence <1% and 100% of villages a prevalence <5%, bringing the epidemiological evaluation results for the Faleme therefore also within the provisional threshold for elimination.
The entomological evaluations of phase 3A were undertaken from July to December 2009, on average about 1 year after the last treatment in the main area, 2.5 years after the last treatment in the additional test areas and 3.5 years after the last treatment in the first test area. During phase 3B the entomological evaluations were done one year later, and on average 2, 3.5 and 4.5 years after the last treatment in the different groups of villages.
The results of the entomological evaluations are summarised in table 6. This table also includes the entomological results for phase 2B after cessation of treatment in the two additional test areas.
In each phase more than 100,000 blackflies were collected and examined, and no infective flies were found in phase 2B and phase 3A. In phase 3B there were two infective flies out of 107,100 flies examined, giving a very low infectivity rate of 0.02 F3H/1000 flies overall or 0.06 for each of the two positive catching points. The upper limit of the 95% confidence interval remains for all catching points below the threshold of 0.5 F3H/1000.
Table 7 gives a summary of the results for the main epidemiological and entomological indicators in the three study sites. In the R. Bakoye focus, the results are very clear. After the cessation of treatment at the end of phase 1, the infection and transmission levels continued to decline and 3 to 4 years after the last treatment both indicators were zero. For the R. Gambia focus, where treatment was given at six monthly intervals instead of annually, the results were equally clear and 3 to 4 years after the last treatment again no mf or infective larvae were detected. For the R. Faleme, where the prevalence of mf was significantly higher at the end of phase 1, there was also no sign of recrudescence after the cessation of treatment but instead there was a clear downward trend in the indicators and 3 to 5 years after the last treatment the prevalence of mf had fallen to extremely low levels, below 10% of the prevalence found in phase 1 before the cessation of treatment, while the vector infectivity rate had fallen below 5% of the provisional entomological threshold for elimination.
Annual or six monthly treatment with ivermectin of populations in onchocerciasis endemic areas has proven to be an effective strategy for controlling the disease as a public health problem. However, whether in the long term this strategy could also achieve elimination of onchocerciasis infection and transmission has not been clear. Until recently, most experts doubted that it would be feasible to achieve onchocerciasis elimination with ivermectin treatment in Africa where the vectors are very efficient and the disease is hyperendemic in many areas [13], [14], [16]. Although models predicted that elimination might be achieved in the long term, there was no empirical evidence to support this prediction and it was generally believed that elimination might not be possible in Africa [13], [14], [16].
The current study has fundamentally changed this perception. The first results of the study, as reported by Diawara et al [17], provided the first evidence that onchocerciasis elimination with ivermectin treatment is feasible in some foci in Africa. The current article reports the final results of the study and provides the definite evidence on the feasibility of elimination based on extensive data on onchocerciasis infection and transmission levels 2 to 3 years after stopping treatment in all villages in the three onchocerciasis foci, and 4 to 5 years after stopping treatment in parts of these foci.
The evaluation data show that after cessation of treatment there was no recrudescence of infection or transmission, but instead a continuously declining trend in infection and transmission levels up to 5 years after the last round of ivermectin treatment. In two sites, the prevalence of mf and the vector infectivity rate was zero during the final round of evaluation 3 to 4 years after the last treatment in these two areas. These results imply that the residual infection levels were so low that the vectors were no longer able to ingest and transmit the parasite, and that there was no renewed mf production by surviving worms up to 5 years after the last treatment. These results convincingly show that local elimination of onchocerciasis has been achieved in these two foci and that, as long as the parasite is not reintroduced, the population of this area will be forever free from the curse of onchocerciasis. Hence, the study has established the proof of principle that onchocerciasis elimination with ivermectin treatment is feasible in some endemic foci in Africa.
The results for the R. Faleme focus were even more remarkable. In this focus, the prevalence of mf was still relatively high at the end of the treatment period and did not completely meet the criteria for stopping treatment. However, these criteria were based on model predictions and experiences with stopping vector control, and they were still provisional criteria for ivermectin treatment to be tested in the current study. When the researchers discussed the phase 1 results for the R. Faleme focus with the Technical Consultative Committee of APOC, the committee recommended that the study should also proceed with stopping treatment and evaluating the subsequent trend in infection and transmission levels in the R. Faleme focus, given the provisional nature of the criteria and the operational importance of improved understanding of the feasibility of elimination and the risk of recrudescence in such borderline situations. It was therefore agreed to proceed with the study in the R. Faleme focus, but prudently through the introduction of two additional test areas that would be evaluated thoroughly for another year before stopping treatment throughout the focus. In view of this initial uncertainty, the final results for the R. Faleme focus were especially remarkable. After cessation of all treatment, the prevalence of mf continued to decline for a period of 3 to 5 years and the vector infectivity levels remained close to 0 throughout the follow-up period. The final results show that it was safe to stop treatment and that elimination has also been achieved in the R. Faleme focus.
It is important to note that the prevalence of mf was not equal to zero in any of the three foci when treatment was stopped, but that nevertheless there was no recrudescence of infection and transmission. This finding provides further evidence of the existence of breakpoints, as predicted by models, below which transmission cannot maintain itself and the infection will die out over time [31]. This is not the first time that empirical evidence of breakpoints has been produced. In virtually all areas where the OCP stopped vector control, the prevalence of mf was not yet zero and there were still several villages where the prevalence, although greatly reduced by vector control, was still in the range of 1 to 5% [32]. However, during the years after the cessation of vector control these residual infection levels declined to zero [33], [34]. Operationally these are important findings. It indicates that a few isolated infections, especially if it concerns only individuals with low mf counts, do not pose a significant threat and that treatment can be safely stopped as long as all indicators are below the elimination thresholds.
The study was undertaken by teams from the ministries of health of the two countries who used the regular epidemiological and entomological evaluation methods in which they had been trained by the OCP and that are the routine evaluation methods used in onchocerciasis control in Africa. In that sense, the study procedures were representative of those likely to be used by onchocerciasis control programs in other countries for future decision making on stopping ivermectin treatment. Nevertheless, our study was a carefully executed experiment with a stringent ethics protocol and one ethical requirement was to treat all individuals who were found to be mf positive during the skin snip surveys, even during surveys done after the cessation of treatment. It is unlikely that this requirement has had a significant impact on the study outcome as there were only very few mf positives detected after the cessation of treatment. In phase 3a, 7 mf positives were detected and treated (2 in R.Bakoye, 2 in R.Gambia and 3 in R.Faleme) but their treatment did not affect the results of any prevalence surveys because villages that were surveyed in phase 3A were not surveyed again in phase 3B. Furthermore, any possible impact on transmission would have been limited as phase 3A surveys were done in a sample of only 18% of villages in the study areas, and any mf positives in the other 82% of villages would have remained untreated during the study period. During phase 3B, 3 mf positives were detected but these were treated after the completion of the study and this did therefore not affect any study results.
It should be noted that one of the mf positives was an adult male from the R. Bakoye who had a relatively high microfilarial density of 58 mf per skin snip and who had been treated only twice. During phase 1, two persons with high microfilarial loads of 87 mf and 96 mf per snip were also detected in the R. Bakoye. It concerned two farmers who lived most of the year in hamlets on their farms on the river banks, near the Simulium breeding sites but very far from their village and they hardly ever received ivermectin. These findings underscores importance for CDTi programmes to ensure that ivermectin treatment reaches isolated high-risk groups such as fishermen and farmers living in hamlets near the river. One 19-year-old man from the R.Faleme, who had been skin snip negative during the surveys in phase 1 and phase 2, was diagnosed with a light infection of 1 and 2 mf in the skin snips from the left and right iliac crest respectively. He was reported to have been treated irregularly. This case might have been an isolated new infection resulting from low level transmission. Alternatively, it might represent a person with a very low intensity of infection at the border of detectability who was false negative during the surveys in 2006 and 2009. The latter explanation seems more plausible given that no other mf positives were detected in this village, nor in any of the surrounding 10 villages, and that the vector infectivity rate in the nearest catching point was zero throughout the study period. An important problem during the epidemiological surveys was the increasing reluctance of the population in the study villages to participate in the skin snip examination. This could have introduced a bias in the skin snip results if those who did not participate in the examination would also be more at risk of infection. Hence the importance of having also an extensive entomological evaluation as an independent measure of transmission levels in the study areas.
There is an ongoing debate about the potential value of a six monthly treatment strategy, as used by OEPA, instead of annual treatment strategy, as used in Africa, in order to reduce the total number of years treatment required to achieve elimination [35], [36]. The present study provides the only comparative data available to date on the long term impact of six monthly versus annual treatment. Our results show that both strategies achieved elimination after 15 to 17 years of treatment. Although the prevalence of mf in phase 1 was slightly lower in the R.Gambia focus, where treatment was given at six monthly intervals, than in the R. Bakoye focus and the R. Faleme focus where treatment has been annual, the final results after cessation of treatment were similar for all three sites, irrespective of treatment frequency. In each site there were still some mf positives after cessation of treatment, even in the R. Gambia focus after 34 treatment rounds, and transmission levels were zero or close to zero in all three sites. However, it is quite possible that elimination was achieved earlier in the R. Gambia focus but our study design does not allow us to determine exactly when the elimination threshold was reached.
One objective of the study was to develop and test a methodology and indicators for decision making on stopping ivermectin treatment. Our experiences indicate the importance of an approach that combines epidemiological and entomological evaluations as a basis for decision making to stop treatment and for subsequent evaluation to ensure that the decision to stop was correct. With respect to the epidemiological threshold for stopping treatment, the R. Faleme results seem to suggest that the current threshold is too conservative and that a higher threshold might be valid. However, onchocerciasis models predict that the risk of recrudescence depends on the precontrol endemicity level as an indicator of the local potential of transmission [10], [20], [37]. The precontrol endemicity levels in the R. Faleme were not very high, and for future evaluations we consider it prudent to maintain the current thresholds until sufficient empirical evidence has accumulated from multiple sites to justify their modification.
The results of the study have had significant impact on the strategy of APOC. Interim study results have been reported annually to the Technical Consultative Committee and Joint Action Forum of APOC. Based on the preliminary results, the Forum accepted that elimination may be feasible in at least some endemic areas, and requested APOC in December 2008 to generate the evidence to determine where and when treatment can be safely stopped [38]. APOC subsequently started an accelerated programme of systematic epidemiological evaluations of the long-term impact of ivermectin treatment and progress towards elimination in APOC projects that had at least 8 years of ivermectin treatment [39].
The current study was undertaken in hyperendemic onchocerciasis foci with seasonal transmission in the dry savanna of Mali and Senegal, and an important question is to what extent the results can be extrapolated to other endemic areas in Africa with different precontrol endemicity levels, transmission patterns and vector species. Computer simulations using the ONCHOSIM model indicate that the speed of decline in prevalence during the ivermectin treatment period, and thus the required duration of treatment to achieve elimination, depends greatly on the precontrol endemicity level. The results of the recent APOC evaluations of progress towards elimination have confirmed this [10], [40]. Hence it should not be concluded from the current study that 15–17 years are required everywhere to achieve elimination. In less endemic areas it should be possible achieve elimination in much shorter periods, maybe even less than 10 years, while it is predicted that in the areas with the highest endemicity levels up to 20 to 25 years of annual treatment may be required [10].
The main vector in the study areas is S. sirbanum which is the predominant vector of onchocerciasis in the dry savanna belt in West and Central Africa where transmission is limited to the rainy season [25], [41]. Hence, with respect to vector species and transmission patterns, our results appear representative for a vast area from Senegal to Sudan where millions of people were infected with O.volvulus. In the rest of Africa, the vectors include S.damnosum s.s. in the wet savanna, several other species of S.damnosum s.l. in forest areas where transmission is mostly perennial and S.neavei in parts of East and Central Africa [42]. The recent epidemiological evaluations by APOC have shown satisfactory progress towards elimination in the vast majority of ivermectin treatment projects in all these areas, and several projects with a population of over 7 million appear to have already reached the elimination breakpoint when treatment can be stopped [43], [44]. The results of these epidemiological evaluations are consistent with the results of phase 1 in our study. Nevertheless, due to differences in vector competence between vector species, the risk of recrudescence after cessation of treatment might still differ between endemic zones [45], [46]. We recommend, therefore, that the first ivermectin treatment projects that reach the elimination threshold in areas with different vector species or very high precontrol endemicity levels, proceed particularly carefully with stopping treatment using a methodology similar to that of our study, including detailed epidemiological and entomological evaluations of onchocerciasis infection and transmission levels for a period of 3 years after the cessation of treatment.
Recent years have seen a paradigm shift from onchocerciasis control to onchocerciasis elimination in Africa. The strategy of APOC has changed from control of onchocerciasis as a public health problem to a strategy of onchocerciasis elimination ‘where feasible’. A recent analysis has suggested that national elimination of onchocerciasis may be feasible in 23 African countries by the year 2020 [40], and a strategic plan to achieve this is under development [43]. The results of the current study have been instrumental for this evolution from onchocerciasis control to onchocerciasis elimination in Africa.
|
10.1371/journal.pcbi.1002861 | Evolutionary Optimization of Protein Folding | Nature has shaped the make up of proteins since their appearance, 3.8 billion years ago. However, the fundamental drivers of structural change responsible for the extraordinary diversity of proteins have yet to be elucidated. Here we explore if protein evolution affects folding speed. We estimated folding times for the present-day catalog of protein domains directly from their size-modified contact order. These values were mapped onto an evolutionary timeline of domain appearance derived from a phylogenomic analysis of protein domains in 989 fully-sequenced genomes. Our results show a clear overall increase of folding speed during evolution, with known ultra-fast downhill folders appearing rather late in the timeline. Remarkably, folding optimization depends on secondary structure. While alpha-folds showed a tendency to fold faster throughout evolution, beta-folds exhibited a trend of folding time increase during the last 1.5 billion years that began during the “big bang” of domain combinations. As a consequence, these domain structures are on average slow folders today. Our results suggest that fast and efficient folding of domains shaped the universe of protein structure. This finding supports the hypothesis that optimization of the kinetic and thermodynamic accessibility of the native fold reduces protein aggregation propensities that hamper cellular functions.
| Nature has come up with an enormous variety of protein three-dimensional structures, each of which is thought to be optimized for its specific function. A fundamental biological endeavor is to uncover the driving evolutionary forces for discovering and optimizing new folds. A long-standing hypothesis is that fold evolution obeys constraints to properly fold into native structure. We here test this hypothesis by analyzing trends of proteins to fold fast during evolution. Using phylogenomic and structural analyses, we observe an overall decrease in folding times between 3.8 and 1.5 billion years ago, which can be interpreted as an evolutionary optimization for rapid folding. This trend towards fast folding probably resulted in manifold advantages, including high protein accessibility for the cell and a reduction of protein aggregation during misfolding.
| The catalog of naturally occurring protein structures [1] exhibits a large disparity of folding times (from microseconds [2], to hours [3]). This disparity is the result of roughly 3.8 billion years of evolution during which new protein structures were created and optimized. The evolutionary processes driving the discovery and optimization of protein topologies is complex and remains to be fully understood. Nature probably uncovers new topologies in order to fulfill new functions, and optimizes existing topologies to increase their performance. Various physical and chemical requirements, from foldability to structural stability, are likely to be additional players shaping protein structure evolution. One indicator for foldability, i.e. the ease of taking up the native protein fold, is a short folding time.
Here we propose that foldability is a constraint that crucially contributes to evolutionary history. Optimization of foldability during evolution could explain the existence of a folding funnel [4], [5], into which a defined set of folding pathways lead to the native state, as postulated early on by Levinthal [6]. While the biological relevance of efficient folding still needs to be explored, an obvious advantage is the increase of protein availability to the cell. For instance, folding could decrease the time between an external stimulus and the organismal response. However, this increase of accessibility is probably limited by other factors such as protein synthesis, proline isomerization and disulfide formation. A probably more important point to support folding speed as an evolutionary constraint is that fast folding avoids proteins aggregation in the cell [7]. Aggregation avoidance could lead to a selection of topologically simple structures that fold rapidly or exclusion of a large number of geometrically feasible structures that compromise accessiblity. This could have reduced the catalog of naturally occurring folds [8]–[10].
The balance between the need for new structural designs and functions in evolution and the physical requirements imposing pressure on folding has remained elusive. The increasing number of organisms with completely sequenced genomes and experimentally acquired models of protein structures, combined with new techniques to study the folding behavior of proteins now open new avenues of inquiry. A common approach for such studies has been the use of molecular simulations such as lattice or coarse-grained techniques, which are efficient enough to scan sequence space. Simulations generally involve an algorithm that mimics the evolutionary accumulation of mutations. This allows to monitor how proteins are selected and evolve towards specific features that are optimized, including those linked to folding, structure and function [11]–[13]. In contrast, we have uncovered phylogenetic signal in the genomic abundance of protein sequences that match known protein structures. Specifically, phylogenomic trees that describe the history of the protein world are built from a genomic census of known protein domains defined by the Structural Classification of Proteins (SCOP) [14] and used to build timelines of domain appearance [15], [16] that obey a molecular clock [17]. This information revealed for example the early history of proteins [18], planet oxygenation [17], and the dynamics of domain organization in proteins [19]. All-atom simulations of denatured proteins folding into their native state [20], [21] are computationally too demanding to systematically evaluate the folding times of the available structural models of protein domains, currently 100,000 in total. A decade ago, Baker and co-workers [22] introduced the concept of contact order, a measure of the non-locality of intermolecular contacts in proteins. Contact order was found to be in good correlation with folding times of two state folders but not multistate proteins. Subsequent studies with extended comparison to experiments led to the definition of the Size-Modified Contact Order (SMCO),(1)where is the number of contacts, is the total number of aminoacids, and is the number of aminoacids along the chain between residues and forming a native contact. By correcting for protein size , the SMCO showed an improved correlation with experimental folding times, with a correlation coefficient of 0.74 [23].
Here, we reveal evolutionary patterns of foldability by mapping the SMCO and thus the folding time onto timelines derived from phylogenomic trees of domain structures (Figure 1). Remarkably, we find there is selection pressure to improve overall foldability, i.e reduce folding times, during protein history. Interestingly, different topologies such as all- and all- folds show distinct patterns, suggesting folding impacts the evolution of some classes of protein structures more than others.
To trace protein folding in evolution, we determined the SMCO of protein domain structures at the Family (F) level of structural organization. Figure 2a shows the folding rate of each F, as measured by its average SMCO, as a function of evolutionary time. Using polynomial regression, we observed a significant decrease (p-value = 9.5e-15) in SMCO in proteins appearing between 3.8 and 1.5 billion years ago (Gya). Trends were maintained when excluding domains from the analysis solved in multi-domain proteins (Figure S11), and also when studying domain evolution at more or less conserved levels of structural abstraction of the SCOP hierarchy. Namely, we find a significant decrease of SMCO at the level of Superfamily (SF), p-value = 2.6e-15), and at the level of domains with less than 95% sequence identity (p-value< = 2.0e-16, Figure S1a,b). Similarly, consistent results were obtained at the F level using linear regression (p-value = 1.0e-06, Figure S1c). Remarkably, even within a smaller data set of only 87 proteins for which folding times have been measured [24], we find that the experimental folding times exhibit a tendency to decrease early in protein evolution (Figure S2). As an additional way of validation, we repeated the analysis for 3 million single domain sequences with predicted SMCO [25], and obtained a decrease again of SMCO up to 1.5 Gya (p-value< = 2.0e-16, Figures S3, S4). Thus, in this initial evolutionary period, proteins tended to fold faster on average.
As suggested by the decrease in SMCO, during evolution, domains diminish long-range and favor short-range interactions, thereby becoming more strongly connected locally. This picture was further corroborated by an analogous analysis of evolutionary trend in tightness, measured by shortest paths in the network of protein contacts [26]. Tightness, and thus the lengths of paths in the interaction network, decreased in evolution until 1.5 Gya, followed by an increase, just like the SMCO (Figure S5). Our results support the hypothesis that folding speed acts as an evolutionary constraint in protein structural evolution.
In contrast, we observed an increase in SMCO between 1.5 Gya and the present (Figure 2a). Thus, the appearance of many new stuctures by domain rearrangement 1.5 Gya, also refered to as the “big bang” [19] of the protein world, affected the evolutionary optimization of protein folding. While a linear regression supports the SMCO increase (p-value = 2e-16), it was not as observed at the SF level or at the level of domains (Figure S1a,b), and for the analysis of experimentally determined rates (Figure S2).
Given the observed overall evolutionary speed-up of protein folding, we would expect a late evolutionary appearance of so-called downhill proteins, which feature ultra-short folding times on the microsecond scale. We annotated 11 downhill folders [27] by their Fs, namely a.35.1.2, a.4.1.1, a.8.1.2, b.72.1.1, and d.100.1.1, and show their average SMCO per family as black triangles in the timeline of Figure 2a. All of them, unsurprisingly, have an SMCO 2, and thus fold significantly faster on average than other structures. We find 7% of families to have a lower SMCO (SMCO 1.5) than the experimentally identified downhill folders. We predict these Fs will fold even faster than the known downhill folders, rendering them interesting candidates for folding assays. The five Fs containing the fast folders have all appeared no earlier than 2.5 Gya, suggesting that they are a result of lengthy evolutionary optimization. According to our predictions, the first fast-folding proteins appeared already 3.4 Gya. However, their frequency and optimization of folding speed continue to increase until 1.5 Gya.
The length of the amino acid chain has been reported to influence the folding kinetics of a protein, with longer chains folding more slowly [23], [27]–[29]. We therefore ask if the decrease in SMCO we observed from 3.8 to 1.5 Gya can be explained by a decrease in the chain length of proteins. Figure 2b shows how domain size has varied in evolution. Folding time measured by SMCO and domain size follow a very similar bimodal trend, with a clear decrease occuring prior to 1.5 Gya and a slight increase after the “big bang”. As expected, we find domain size, which equals in Equation 1, and SMCO to be correlated with folding rate in agreement with other studies [8], [23] (Figure S6). In line with this correlation, the downhill folders discussed above and shown in Figure 2a as triangles, have a small domain size of less than 100 residues in common.
We next eliminated the effect of domain size on the evolutionary trends observed in folding rate to analyze factors other than domain size. To this end, we dissected our dataset according to the amino acid chain length. This analysis was done with all 92,000 domains to ensure enough data points for each length. The distributions of chain length are shown in Figure 3a, b for the two time periods before and after the “big bang” (1.5 Gya). The length distribution for proteins appearing before the “big bang” exhibited a peak at around 150 amino acids, and shifted later (1.5 Gya to the present) to shorter chains with a peak at around 100 aminoacids, underlining the tendency for a decrease of domain size. We note that the resulting average chain length of three-dimensional structures in SCOP, which have been obtained from X-ray or NMR measurements, is smaller than the average length of sequences in genomes [30], apparently due to the increasing experimental difficulties when working with large proteins. We then analyzed evolutionary tendencies for every domain length subset by measuring the variation in the end points of a polynomial regression. The color mapping in Figure 3a indicates an increase (blue), a decrease (yellow-red), or a non-significant change (green) of SMCO. Overall, 85% of the data returned a significant result according to the F-test. During early protein evolution (3.8–1.5 Gya), we found that 54%0.3% of all domains in each size subset optimized their foldability during evolution by decreasing their SMCO. Conversely, 37%0.4% of domains showed a slow-down in folding, i.e. a significant increase in SMCO. These results confirm the tendencies observed for the full data set (Figure 1a), and hold for different tresholds of identity, namely 95% and 40% (Figures S7, S8). As expected, due to the smaller data set, partitioning domains defined at F and SF levels according to size yielded results that were statistically not significant. In summary, even after dissecting the effect of chain length on changes in SMCO, the tendency of proteins to fold faster during evolution is confirmed.
After the “big bang”, the SMCO and thus foldability showed a overall increase in evolution (Figure 3b), in agreement with results from the total set (Figure 2a). Apparently, fast folding did not represent a major evolutionary constraint during this period. Instead, other constraints must have been optimized at the expense of foldability. We next discuss secondary structure as one factor influencing the impact of foldability on protein structure evolution.
Secondary structure composition is another factor reported to have an influence on folding kinetics [23], [27], [28]. We repeated the analysis of domains partitioned by size that was described above for domains in each secondary structure class of SCOP (all-, all-, /, and + domains) and thereby revealed differences in the evolution of foldability. As shown in Figure 4a, the tendency of a decreasing SMCO before the “big bang” is reproduced for all classes. This result was confirmed at the level 95% identity and 40% identity (Figures S9, S10), though with a significant decrease only for the + and classes at the 40% identity level, i.e. for a much smaller data set. Again, our analysis strongly supports an evolutionary constraint for fast folding of proteins appearing early in evolution, 3.8–1.5 Gya.
Interestingly, we here observe a specialization of protein classes, with all- proteins tending to fold faster and all- proteins tending to fold more slowly, all of which was supported at the 95% domain level (Figure 4b). Why should the all- class be under a stronger fast folding constraint than the all- class? Figure S12 shows the average SMCO for each secondary structure class. The all- and all- class show the highest and lowest SMCO, respectively, suggesting that all- proteins in general fold slower than all- proteins. This is in line with previous findings that containing all- proteins fold more slowly than all- proteins due to long range interactions between all- strands that increase contact order [23], [27], [28].
Protein aggregation damages cellular components and can lead to a variety of neuronal diseases [31]–[33]. A way of reducing aggregation is to enhance the kinetic and thermodynamic accessibility of the native fold of a protein. Incremental increases in kinetic or thermodynamic stability of a protein might therefore represent an evolutionary trace reflecting optimization of protein foldability [34].
Here, we confirm the hypothesis that foldability exerts a constraint in the evolution of protein domain structures, as we find a tendency of proteins to on average fold faster than their structural ancestors. As expected, shortening of protein chain length during evolution is an important factor leading to faster folding. However, the exclusion of this protein-size effect preserved the trend of decreasing folding times. Thus, faster folding is not a side effect of chain shortening, but likely acts as an evolutionary constraint in itself. An alternative reason for the decrease of folding times in evolution is the need of proteins for flexibility in order to optimize their function such as enzymatic catalysis or allosteric regulation [35]. Folding speed and flexibility are known to correlate, as the formation of the compact state with no or only minor native contacts is much quicker than the arrangement of the native – often long-range – contacts [36]. Fewer native contacts in turn result in lower stability and may increase conformational flexibility as required for some biological functions [37]. Our analysis of protein folding speed on an evolutionary time line can be similarly carried out for measures of flexibility to test this scenario.
Evolutionary constraints on folding are apparently not uniformly imposed onto the full repertoire of protein structures and during the entire protein history. Instead, our analysis revealed a bimodal evolutionary pattern, with folding speed increasing and decreasing before and after 1.5 Gya, respectively. The speed-up of folding was most pronounced for all- folds. The evolutionary inflexion point coincides with the previously identified protein “big bang”, which features a sudden increase in the number of domain architectures and rearrangements in multi-domain proteins triggered by increased rates of domain fusion and fission. We speculate that the slow down of folding that ensues could be due to cooperative interactions during folding of domains in the emerging multi-domain proteins [38]. Alternatively, the observed slow-down after the “big bang” could be related to the appearance of protein architectures that are known to help proteins to fold, such as chaperones [39], [40] Moreover, protein architectures specific to eukaryotes appeared at 1.5 Gya [16]. The Eukaryotic domain of life has the most elaborate protein synthesis and housekeeping machinery, including enzymes for post-translational modification. This machinery might have mitigated the constraints for fast folding, thereby increasing evolutionary rates of change [34], while preventing misfolding and aggregation prior to attaining the native fold [41].
Finally, we revealed striking evolutionary diversity in protein folding when comparing all- and all- fold classes from 1.5 Gya. Their average folding times diverged after the “big bang”, with the all- class further decreasing and the all- class instead increasing their folding times. This result can support the idea of an optimization of folding that increased the difference in folding time between all- and all- through evolution. As previously shown [22], all- folds have on average higher SMCO and fold slower than their all--counterparts. This simply results from their different topology and is also the result of our analysis (Figure S12). We here show that earlier in evolution, however, folding times have been more similar and only diverged from each other as late as after 1.8 Gya. But why would all- folds have been relieved from the evolutionary constraint of fast folding? Since the “big bang” is responsible for the discovery and optimization of many new functions, including an elaborate protein synthesis and folding machinery, we speculate that the divergence of averge folding times of all- and all- folds probably reflects an optimization of function. This optimization happens to be on the expense of foldability for only the all- class, the reasons of which remain unknown. One possible scenario would be that all- have the tendency to carry out functions that require high flexibility, a property that correlates with few long-range contacts, i.e. high foldability.
An important experimental study by Baker and colleagues [42] tested the idea that rapid folding of biological sequences to their native states does not require extensive evolutionary optimization. Using a phage display selection strategy, the barrel fold of the SH3 domain protein was reproduced with a reduced alphabet of only five amino-acids without any loss in folding rate. Despite extensive changes to protein sequence, experimental manipulation preserved contact order. While these results should not be generalized to the thousands of other fold topologies that exist in nature, they are revealing. They suggest that stabilizing interactions and sequence complexity can be sufficiently small and still enable evolutionary folding optimization. In other words, optimal folding structures can find their way through the free energy landscape without extensive explorations of sequence space. This property of robustness could be a recent evolutionary development, since the SH3 domain F appears very late in our timeline of protein history. Alternatively, it could represent a general structural property. The fact that we now see clear and consistent foldability patterns along the entire timeline supports the existence of limits to evolutionary optimization of folding that are being actively overcome in protein evolution. We conjecture that these limits were initially imposed by the topologies of the early folds, and that structural rearrangements (resulting from insertions, tandem duplication, circular permutations, etc [43]–[46]) offered later on opportunities for fast and robust folding as evolving structures negotiated trade-offs between function and stability.
We end by noting that we cannot exclude overlooking effects on folding times from cooperative folding. These could influence trends of folding times. The SMCO is known to show high correlations with folding times only for single-domain proteins [22]. Developing schemes for estimating folding times from structures comprising more than one domain is a challenge [38] but would enable a more general view onto protein foldability as a constraint throughout evolution. Moreover, our analysis is based on the sequence and structural data that is available. Results might therefore be biased by the choice of proteins and their accessibility. However, the structure of most protein folds and families have been acquired and will not exceed those that are expected [47]. Moreover, our approach allow us to steadily test if the predicted evolutionary trends of foldability are maintained upon inclusion of new sequences and protein folds into the analysis. Interestingly, multiple studies have found folding rates to correlate with stability rather than contact order [48]. Analyzing phylogenomic trends of stability might in this light be an important study to further elucidate evolutionary contraints on protein structure.
A most parsimonious phylogenomic tree of F domain structures was reconstructed from a structural genomic census of 3,513 Fs (defined according to SCOP 1.73) in the proteomes of 989 organisms (76 Archaea, 656 Bacteria and 257 Eukarya) with genomes that have been completely sequenced [49]. Similarly, a most parsimonious phylogenomic tree of SF structures (860,497 steps; CI = 0.0255, HI = 0.9745, RI = 0.780, RC = 0.020; g1 = −0.109) was derived from a structural genomic census of 1,915 SFs (defined according to SCOP 1.73) in the proteomes of 1,096 organisms (78 Archaea, 719 Bacteria and 299 Eukarya). The structural census scanned genomic sequences against a library of hidden Markov Models (HMMs) in SUPERFAMILY [50] with probability cutoffs E of 10-4, as described in detail in previous studies [15], [16]. Data matrices of domain abundances were normalized to genome size, coded as multistage phylogenetic characters with characters states ranging from 0 to 29, and used to build rooted trees using maximum parsimony (MP) as optimality criterion in PAUP* [51]. A combined parsimony ratchet and iterative search approach avoided traps in suboptimal regions of tree space. Finally, the age of each domain (nd) was derived directly from its relative position as taxa in reconstructed trees. A PERL script counted the number of nodes from the most ancient domain at the base of the tree to each leaf, providing it in a relative 0-to-1 scale. These relative ages (in nd units) were transformed to geological ages (in Gya) by using molecular clocks of SFs and Fs derived previously [17] and used to construct an evolutionary timeline of domain appearance. A general finding is a sudden explosion of diversity in protein architectures at 1.5 Gya [19].
As a measure for the folding time of each protein architecture, we evaluated the size modified contact order (SMCO) of domains indexed in the SCOP database. We used the ASTRAL repositories to download the 92,470 three-dimensional structures classified in SCOP 1.73. The phylogenomic tree was built at the F level on the basis of the same protein structures, i.e. the 1.73 SCOP version. We note that the SMCO calculations are based on single protein domains from SCOP, while many proteins consist of multiple domains. Some studies showed that interactions between domains might affect folding [52]. To test if the evolutionary trends also hold for the subset of domains excluding those which have been structurally solved in multi-domain proteins, we carried out the following steps. We first downloaded the CathDomainList from the website of CATH (http://www.cathdb.info/download), and removed the PDB chains with two or more CATH domains or NMR structures or obsolete PDB entries. We then eliminated redundancy using the PISCES webserver (http://dunbrack.fccc.edu/PISCES.php) [53] using the following cut-offs: Sequence percentage identity: < = 25%, resolution: 0.0 3.0, R-factor: 0.3, sequence length: 40 10,000, Non X-ray entries: excluded, C-only entries: excluded, cull PDB by chain. We detected SCOP families using HMMs on the PDB chains and removed chains with long non-domain segments, i.e. the length of a segments without any domain assignment should be less than 30. Finally, we removed the chains with two or more SCOP families and the chains with two or more CATH entries. Using this dataset, we revealed the same tendencies in SMCO (Figure S11) as those of the whole dataset (compare Figure 2).
We calculated the average SMCO for each F and SF, and mapped these averages, 3,513 of them for F, and 1,915 for SF, onto timelines derived from corresponding phylogenomic trees. Average SMCO of each F or SF as a function of node distance showed non-linear dependencies that were therefore analyzed using LOESS (locally weighted polynomial regression) [54], [55] to reveal global trends of foldability during evolution. A second-degree polynomial was fitted to the data at each point of the timeline, with a span of 0.7. LOESS resulted in regression function values for each of the 3,513 F or 1,915 SF data points. The results from LOESS revealed a drastic change in SMCO at 1.5 Gya, a time point of evolution that coincides with the “big bang” of protein domain rearrangements and the rise of Eukarya [19]. We therefore also analyzed our data by two independent linear regressions describing SMCO data points before and after the “big bang”. To validate our results, we repeated the phylogenomic analysis of SMCO using two subsets of protein structures, namely only SCOP domains with 40% of sequence identity (10,570 domains), and those with 95% identity (16,713 domains). In addition, we used one subset of single domain sequences (3,500,000 domains) from the TrEMBL [56] database with predicted SMCO [57] the results of which are shown in Figures S3, S4. Only results valid for all four different data sets and thus robust with respect to the selection of protein structures are presented here, if not otherwise noted. For the chain length analysis, we used all 92,000 domains to ensure enough data points for each length. The distributions of chain length are shown in Figure 3a, b. The analysis was repeated 100 times with varying data sample and every dataset (e.g: 95% and 40%). We obtained standard errors of the mean, which are included in Figure 3, 4 and Figures S7, S8, S9, S10.
|
10.1371/journal.ppat.1006995 | N6-methyladenosine modification and the YTHDF2 reader protein play cell type specific roles in lytic viral gene expression during Kaposi's sarcoma-associated herpesvirus infection | Methylation at the N6 position of adenosine (m6A) is a highly prevalent and reversible modification within eukaryotic mRNAs that has been linked to many stages of RNA processing and fate. Recent studies suggest that m6A deposition and proteins involved in the m6A pathway play a diverse set of roles in either restricting or modulating the lifecycles of select viruses. Here, we report that m6A levels are significantly increased in cells infected with the oncogenic human DNA virus Kaposi’s sarcoma-associated herpesvirus (KSHV). Transcriptome-wide m6A-sequencing of the KSHV-positive renal carcinoma cell line iSLK.219 during lytic reactivation revealed the presence of m6A across multiple kinetic classes of viral transcripts, and a concomitant decrease in m6A levels across much of the host transcriptome. However, we found that depletion of the m6A machinery had differential pro- and anti-viral impacts on viral gene expression depending on the cell-type analyzed. In iSLK.219 and iSLK.BAC16 cells the pathway functioned in a pro-viral manner, as depletion of the m6A writer METTL3 and the reader YTHDF2 significantly impaired virion production. In iSLK.219 cells the defect was linked to their roles in the post-transcriptional accumulation of the major viral lytic transactivator ORF50, which is m6A modified. In contrast, although the ORF50 mRNA was also m6A modified in KSHV infected B cells, ORF50 protein expression was instead increased upon depletion of METTL3, or, to a lesser extent, YTHDF2. These results highlight that the m6A pathway is centrally involved in regulating KSHV gene expression, and underscore how the outcome of this dynamically regulated modification can vary significantly between cell types.
| In addition to its roles in regulating cellular RNA fate, methylation at the N6 position of adenosine (m6A) of mRNA has recently emerged as a mechanism for regulating viral infection. While it has been known for over 40 years that the mRNA of nuclear replicating DNA viruses contain m6A, only recently have studies began to examine the distribution of this modification across viral transcripts, as well as characterize its functional impact upon viral lifecycles. Here, we apply m6A-sequencing to map the location of m6A modifications throughout the transcriptome of the oncogenic human DNA virus Kaposi’s sarcoma-associated herpesvirus (KSHV). We show that the m6A machinery functions in a cell type specific manner to either promote or inhibit KSHV gene expression. Thus, the KSHV lifecycle is impacted by the m6A pathway, but the functional outcome may depend on cell lineage specific differences in m6A-based regulation.
| The addition of chemical modifications is critical to many steps of mRNA processing and the regulation of mRNA fate. There are more than 100 different RNA modifications, but the most abundant internal modification of eukaryotic mRNAs is N6-methyladenosine (m6A), which impacts nearly every stage of the posttranscriptional mRNA lifecycle from splicing through translation and decay [1–6]. The breadth of impacts ascribed to the m6A mark can be attributed to its creation of new platforms for protein recognition, in part via local changes to the RNA structure [4,7–12]. The reversibility of m6A deposition through the activity of demethylases termed erasers adds a further layer of complexity by enabling dynamic regulation, for example during developmental transitions and stress [1,4,5,13–15]. Deposition of m6A occurs co- or post-transcriptionally through a complex of proteins with methyltransferase activity known as writers, which include the catalytic subunit METTL3 and cofactors such as METTL14 and WTAP [1,4,14,16,17]. The modification is then functionally ‘interpreted’ through the selective binding of m6A reader proteins, whose interactions with the mRNA promote distinct fates.
The best-characterized m6A readers are the YTH domain proteins. The nuclear YTHDC1 reader promotes exon inclusion [6], whereupon m6A-containing mRNA fate is guided in the cytoplasm by the YTHDF1-3 readers. Generally speaking, YTHDF1 directs mRNAs with 3’ UTR m6A modifications to promote translation [3], whereas YTHDF2 recruits the CCR4-NOT deadenylase complex to promote mRNA decay [18]. YTHDF3 has been proposed to serve as a co-factor to potentiate the effects of YTHDF1 and 2 [3,19,20]. Although the individual effects of YTHDF1 and 2 seem opposing, the YTHDF proteins may coordinate to promote accelerated mRNA processing during developmental transitions and cellular stress [1]. YTHDC2, the fifth member of the YTH family proteins, was recently shown to play critical roles in mammalian spermatogenesis through regulating translation efficiency of target transcripts [21]. Additional examples of distinct functions for m6A readers under specific contexts such as heat shock are rapidly emerging [13].
Given the prevalence of the m6A modification on cellular mRNAs, it is not surprising that a number of viruses have been shown to contain m6A in their RNA [22–29]. Indeed, a potential viral benefit could be a less robust innate antiviral immune response, as m6A modification of in vitro synthesized RNAs diminishes recognition by immune sensors such as TLR3 and RIG-I [30,31]. That said, the functional consequences of viral mRNA modification appear diverse and include both pro- and anti-viral roles. In the case of Influenza A, a negative sense ssRNA virus, m6A and the reader YTHDF2 have been shown to promote viral replication [32]. Furthermore, multiple studies have mapped the sites of m6A modification in the human immunodeficiency virus (HIV) genome, and shown that it promotes the nuclear export of HIV mRNA as well as viral protein synthesis and RNA replication [24,26,28]. Roles for the YTHDF proteins during HIV infection remain varied however, as Tirumuru and colleagues propose they function in an anti-viral context by binding viral RNA and inhibiting reverse transcription, while Kennedy and colleagues observe they enhance HIV replication and viral titers [24,28]. A more consistently anti-viral role for the m6A pathway has been described for the Flaviviridae, whose (+) RNA genomes are replicated exclusively in the cytoplasm and contain multiple m6A sites in their genomic RNA [23,25]. An elegant study by Horner and colleagues showed that depletion of m6A writers and readers or the introduction of m6A-abrogating mutations in the viral E1 gene all selectively inhibit hepatitis C virus (HCV) assembly [23]. Similarly, depletion of METTL3 or METTL14 enhances Zika virion production [25].
Despite the fact that m6A modification of DNA viruses was first reported more than 40 years ago for simian virus 40, herpes simplex virus type 1, and adenovirus type 2, roles for the modification in these and other DNA viruses remain largely unexplored [33–37]. Unlike most RNA viruses, with few exceptions DNA viruses replicate in the nucleus and rely on the cellular transcription and RNA processing machinery, indicating their gene expression strategies are likely interwoven with the m6A pathway. Indeed, it was recently shown that the nuclear reader YTHDC1 potentiates viral mRNA splicing during lytic infection with Kaposi’s sarcoma-associated herpesvirus (KSHV) [38]. Furthermore, new evidence suggests m6A modification potentiates the translation of late SV40 mRNAs [39], further indicating that this pathway is likely to exert a wide range of effects on viral lifecycles.
Here, we sought to address roles for the m6A pathway during lytic KSHV infection by measuring and mapping the abundance of m6A marks across the viral and host transcriptome. This gammaherpesvirus remains the leading etiologic agent of cancer in AIDS patients, in addition to causing the lymphoproliferative disorders multicentric Castleman’s disease and primary effusion lymphoma. The default state for KSHV in cultured cells is latency, although in select cell types the virus can be reactivated to engage in lytic replication, which involves a temporally ordered cascade of gene expression. We reveal that m6A levels are significantly increased upon KSHV reactivation, which is due to a combination of m6A deposition across multiple kinetic classes of viral transcripts and a concomitant decrease in m6A levels across much of the host transcriptome. Depletion of m6A writer and cytoplasmic reader proteins impaired viral lytic cycle progression in the KSHV iSLK.219 and iSLK.BAC16 reactivation models, suggesting this pathway potentiates the KSHV lytic cycle. Interestingly, however, the roles for the m6A writer and readers shifted to instead display neutral or anti-viral activity in the TREX-BCBL-1 reactivation model. These findings thus demonstrate that while KSHV mRNAs are marked by m6A, the functional consequences of this mark can vary significantly depending on cell context, reinforcing both the functional complexity and dynamic influence of m6A.
Epitranscriptome mapping has revealed significant roles for the m6A pathway in the lifecycle and regulation of several RNA viruses, but at the time we initiated these studies, similar global analyses had yet to be performed for a DNA virus. Given that herpesviral mRNAs are transcribed and processed in the nucleus using the cellular RNA biogenesis machinery, we hypothesized that these viruses would engage the m6A pathway. We therefore first quantified how KSHV reactivation impacted total cellular m6A levels in the KSHV-positive renal carcinoma cell line iSLK.219 (Fig 1). These cells are a widely used model for studying viral lytic events, as they stably express the KSHV genome in a tightly latent state but harbor a doxycycline (dox)-inducible version of the major viral lytic transactivator ORF50 (also known as RTA) that enables efficient entry into the lytic cycle [40,41]. Polyadenylated (polyA+) RNA was enriched from untreated (latent) or dox-reactivated iSLK.219 cells and the levels of m6A were quantitatively analyzed by liquid chromatography-tandem mass spectrometry (LC-MS/MS) (Fig 1A). Indeed, we observed a three-fold increase in total m6A levels upon KSHV lytic reactivation, suggesting that m6A deposition significantly increased during viral replication (Fig 1B).
We next sought to discern whether the increase in m6A during the KSHV lytic cycle favors host or viral mRNAs using high throughput m6A RNA sequencing (m6A-seq) [42]. This technique can reveal both the relative abundance and general location of m6A in KSHV and cellular mRNA. Total m6A containing RNA was immunoprecipitated from 2 biological replicates of latent or lytically reactivated iSLK.219 cells using an m6A-specific antibody. DNase-treated total mRNA was fragmented to lengths of 100 nt prior to immunoprecipitation and then subjected to m6A-seq. Total RNA-seq was run in parallel for each sample, allowing the degree of m6A modification to be normalized with respect to transcript abundance because the levels of many transcripts change upon viral lytic reactivation. Peaks with a fold-change four or higher (FC>4) and a false discovery rate of 5% or lower (FDR>5%) in both replicates were considered significant, although it is possible that additional transcripts detectably modified to lower levels or in a more dynamic manner may also be functionally regulated by m6A (complete list of viral peaks with FC>2 in S1 Table). In lytically reactivated samples, 10 transcripts comprising genes of immediate early, early, and late kinetic classes displayed significant m6A modification in both replicates (Figs 2A and S1). Within these KSHV mRNAs, m6A peaks were detected primarily in coding regions, although in some cases the location of a peak in a coding region overlaps with a UTR (S1 Fig). Furthermore, all but one peak contains at least one instance of the GG(m6A)C consensus sequence. While many of the modified viral transcripts contained only one m6A peak, multiple peaks were found in certain transcripts, including the major lytic transactivator ORF50 (Fig 2B). Of note, exon2 of ORF50 contained one m6A peak of FC>4 in replicate one, and three m6A peaks in replicate two, each of which have at least one m6A consensus motif, further increasing confidence that these peaks accurately represent m6A modified sites. Furthermore, the viral ncRNA PAN, which has been reported to comprise over 80% of nuclear PolyA+ RNA during lytic reactivation [43], contains FC>4 peaks in both replicates. Modification of PAN likely accounts for the marked three-fold increase in cellular m6A content observed upon lytic reactivation (Fig 1B). As anticipated given the restricted viral gene expression profile during latency, unreactivated samples had many fewer m6A containing viral mRNAs, with the only FC>4 peaks occurring in both replicates located in ORF4. Although ORF4 is a lytic transcript, its coding region overlaps with the 3’ UTR of K1, which is expressed during both the latent and lytic phases of the viral lifecycle (Figs 2A and S1) [44,45].
To validate the m6A-seq results, we performed m6A RNA immunoprecipitation (RIP) followed by quantitative real-time PCR (RT-qPCR) on six of the viral transcripts predicted to be m6A modified from the m6A-seq data. This technique allows determination of the relative level of m6A content in a given transcript compared to an unmodified transcript. As controls, we included primers for the cellular GAPDH transcript, which is known not to be m6A modified, and the DICER transcript, which is m6A modified [42]. The m6A RIP RT-qPCR confirmed modification of the vIL-6, K1, ORF50, ORF57 and PAN viral transcripts, in agreement with m6A-seq results (Fig 2C). In summary, we found m6A modification in approximately one third of KSHV transcripts upon lytic reactivation, consistent with the hypothesis that this pathway contributes to KSHV gene expression.
We next compared the distribution of m6A peaks in host mRNAs from unreactivated versus reactivated cells to assess whether lytic KSHV infection altered the m6A profile of cellular transcripts. Analyzing the two independent replicates for each condition, we found an average of 14,092 m6A modification sites (FC>4 and FDR>5%) in host transcripts pre-reactivation, compared to 10,537 peaks post-reactivation (Fig 2D and S2 Table). We observed that this >25% decrease in m6A deposition on cellular mRNA encompassed a wide spectrum of transcripts, and no notable patterns were apparent by GO term analysis for functional categories enriched in the altered population. Thus, while the functional impact of the altered host m6A profile remains unresolved, the observation that KSHV lytic infection increased the level of m6A in total poly A+ RNA despite decreasing its presence in cellular mRNA implies that m6A deposition during infection favors viral transcripts.
Given the significant deposition of m6A across KSHV transcripts, we reasoned that m6A might play an important role in potentiating the viral lifecycle. We therefore examined the effect of depleting the m6A writers and readers on KSHV virion production using a supernatant transfer assay. The KSHV genome in iSLK.219 cells contains a constitutively expressed version of GFP, which allows for fluorescence-based monitoring of infection by progeny virions. We performed siRNA-mediated knockdown of METTL3, the catalytic subunit responsible for m6A deposition, as well as the m6A readers YTHDF 1, 2 and 3 (Fig 3A). Cells were then treated with dox and sodium butyrate to induce lytic reactivation for 72 hr, whereupon supernatants were collected and used to infect 293T recipient cells. The number of GFP positive 293T cells at 24 hpi was measured by flow cytometry (Fig 3B). Notably, for virus generated from METTL3 depleted cells, only 7% of recipient cells were infected compared to 82% for virus generated during treatment with a control siRNA (Fig 3B). YTHDF2 depletion caused an even more pronounced defect, resulting in a near absence of virion production (Fig 3B). In contrast, YTHDF3 knockdown resulted in only modest changes in virion production, while virion production was unaffected by YTHDF1 knockdown (Fig 3B). The prominent defect in virion production in METTL3 and YTHDF2 depleted cells was not due to knockdown-associated toxicity, as we did not observe changes in cell viability in siRNA treated cells (representative experiment shown in S2 Fig). Furthermore, we validated the results for YTHDF2 and YTHDF3 using independent siRNAs (S2 Fig). Thus, the m6A writer METTL3 and the reader YTHDF2 play important roles in driving KSHV infectious virion production in iSLK.219 cells.
We then sought to determine the stage of the viral lifecycle impacted by the m6A pathway by measuring the impact of writer and reader depletion on the abundance of viral mRNAs of different kinetic classes. First, levels of representative immediate early, delayed early, and late viral mRNAs were measured by RT-qPCR following lytic reactivation for 72 hr. ORF50 and K8.1 transcripts contained at least one m6A peak, while ORF37 did not appear to be significantly modified in our m6A-seq data (see S1 Table). METTL3 depletion did not appear to impact accumulation of the ORF50 immediate early or ORF37 delayed early mRNAs at this time point, but resulted in a significant defect in accumulation of the K8.1 late gene mRNA (Fig 3C). Consistent with the virion production data, we observed a striking and consistent defect in the accumulation of each of the viral transcripts upon YTHDF2 depletion, suggesting that this protein is essential for lytic KSHV gene expression beginning at the immediate early stage (Fig 3C). Similar results were observed using an independent YTHDF2-targeting siRNA (S2 Fig). We also observed a prominent defect in accumulation of ORF50 and the delayed early ORF59 proteins by Western blot specifically upon YTHDF2 depletion (Fig 3D). In contrast, depletion of YTHDF1 or YTHDF3 did not reproducibly impact ORF50, ORF37, or K8.1 gene expression at 72 hr post reactivation (Fig 3C).
In agreement with the above findings, we also observed that iSLK.219 cells depleted of METTL3 and YTHDF2 displayed a prominent defect in viral reactivation, as measured by expression of red fluorescent protein (RFP) driven by the PAN lytic cycle promoter from the viral genome (Fig 3E). Similarly, ORF50 protein production was also markedly reduced upon METTL3 or YTHDF2 depletion at the 24 hr time point, which represents the early phase of the lytic cycle (Fig 3F).
To determine whether the effects of the m6A pathway on ORF50 were dependent on KSHV infection, we measured ORF50 protein in an uninfected iSLK cell line containing only the integrated, dox-inducible ORF50 gene (iSLK.puro cells) (Fig 3G). Similar to our findings with infected iSLK.219 cells, depletion of METTL3 or YTHDF2 strongly reduced ORF50 protein levels (Fig 3H). YTHDF3 depletion resulted in an increase in ORF50 expression, which we also observed to a more modest degree in the iSLK.219 cells (see Fig 3D). Collectively, these results suggest that m6A modification is integral to the KSHV lifecycle, and that YTHDF2 plays a particularly prominent role in mediating KSHV lytic gene expression in iSLK.219 cells. They further indicate that m6A modification can impact ORF50 expression in both uninfected and KSHV infected iSLK cells.
ORF50 is the major viral transcriptional transactivator, and its expression is essential to drive the KSHV lytic gene expression cascade [46]. The observations that ORF50 is m6A modified and that its accumulation is dependent on YTHDF2 indicate that the m6A pathway plays key roles in ORF50 mRNA biogenesis or fate in iSLK.219 cells, potentially explaining the lytic cycle progression defect in the knockdown cells. Deposition of m6A has been reported to occur both co-transcriptionally and post-transcriptionally [1,16,17,47]. To determine whether the m6A pathway is important for ORF50 synthesis or its posttranscriptional fate, we measured ORF50 transcription in reactivated iSLK.219 cells upon depletion of METTL3, YTHDF2, or YTHDF3 using 4-thiouridine (4sU) metabolic pulse labeling. 4sU is a uridine derivative that is incorporated into RNA during its transcription, and thiol-specific biotinylation of the 4sU-containing RNA enables its purification over streptavidin-coated beads [48,49]. At 24 hr post reactivation, RNA in the siRNA treated iSLK.219 cells was pulse labeled with 4sU for 30 min, whereupon the labeled RNA was isolated by biotin-streptavidin purification and viral transcripts were quantified by RT-qPCR (Fig 4A). Despite the defect in ORF50 accumulation observed upon YTHDF2 depletion (see Fig 3F), we observed no decrease in 4sU-labeled ORF50 mRNA upon depletion of any of the m6A writer or reader proteins (Fig 4B). However, in YTHDF2 depleted cells, there was a prominent defect in the level of 4sU-labeled ORF37, likely because its transcription is dependent on the presence of ORF50 protein (Fig 4C).
The ORF50 mRNA detected in the above experiments represents a combination of the mRNA transcribed from the dox-inducible cassette as well as from the KSHV genome [41]. While the dox-inducible promoter is constitutively active under dox treatment, ORF50 transcription from KSHV is sensitive to ORF50 protein levels because it transactivates its own promoter [50]. The decreased ORF50 protein levels observed in Fig 3 might therefore lead to a selective reduction in transcription from the native ORF50 promoter by interfering with this positive transcriptional feedback. Indeed, primers designed to specifically recognize ORF50 derived from the viral genome revealed a marked defect in transcription of KSHV-derived ORF50 upon YTHDF2 depletion, as well as a slight reduction upon METTL3 depletion (Fig 4D). Collectively, the above results suggest that m6A initially functions to post-transcriptionally regulate ORF50 mRNA abundance, but that when ORF50 protein levels fall upon YTHDF2 or METTL3 knockdown, the positive transcriptional feedback mechanism at the viral promoter also becomes restricted.
To independently validate the METTL3 and YTHDF2 phenotypes, we also evaluated their importance in the iSLK.BAC16 model. Although independently generated, this is the same cell background as iSLK.219, including the dox-inducible ORF50, but instead contains the viral genome in the context of a bacterial artificial chromosome (BAC16) [51]. Similar to our results with the infected iSLK.219 cells, depletion of METTL3 or YTHDF2 in iSLK.BAC16 cells led to a significant defect in virion production as measured by supernatant transfer assays (Fig 5A–5C). In addition, the total levels of ORF50 mRNA (from the dox-induced plus viral promoters) were unchanged between the different siRNA treated cells, while depletion of YTHDF2 led to a significant reduction in the level of BAC16-derived ORF50 and K8.1 mRNAs (Fig 5D). In contrast, METTL3 depletion did not significantly impact the level of ORF50, ORF37, or K8.1 transcripts. It should be noted that levels of METTL3 knockdown in excess of 80% have only been reported to reduce m6A levels in Poly A RNA by 20–30% [17]. Thus, at least some fraction of ORF50 (and other) transcripts may still be m6A modified due to residual enzyme activity of the remaining METTL3. In agreement with these observations, knockdown of METTL3 modestly reduced but did not eliminate the pool of m6A modified ORF50 or the cellular SON mRNAs in iSLK.BAC16 cells as measured by m6A RIP RT-qPCR (S3 Fig). Finally, we observed that although ORF59 protein levels were consistently reduced upon YTHDF2 knockdown, and to a more variable extent upon METTL3 depletion, we did not detect the same marked effects on ORF50 protein levels in iSLK.BAC16 cells as in iSLK.219 cells (Fig 5E). In summary, although iSLK.219 and iSLK.BAC16 cells exhibit a somewhat different gene expression profile in the context of YTHDF2 and METTL3 knockdown, depletion of these m6A pathway components restricts the KSHV lytic lifecycle in both models.
Given the diversity of functions reported for m6A in controlling cellular processes and virus infections [1,22,27], we also sought to evaluate the role of this pathway in mediating ORF50 expression in another widely used KSHV infected cell line of distinct origin, the B cell line TREX-BCBL-1 [52]. Similar to iSLK.219 and iSLK.BAC16 cells, TREX-BCBL-1 cells also contain a dox-inducible copy of ORF50 to boost reactivation. First, we evaluated whether the ORF50 transcript was m6A modified in TREX-BCBL-1 cells by m6A RIP, followed by RT-qPCR using control or ORF50 specific primers. Indeed, there was a clear enrichment of ORF50 in the reactivated, m6A-containing RNA population (Fig 6A). As expected, we detected the m6A modified DICER transcript in both reactivated and unreactivated cells, whereas the unmodified GAPDH transcript was present in neither sample (Fig 6A).
The m6A pathway components were then depleted from TREX-BCBL-1 cells via siRNA treatment, whereupon cells were reactivated for 72 hr with dox, TPA, and ionomycin. As knockdown efficiency for YTHDF1 was inconsistent in this cell type, we focused on the impact of METTL3, YTHDF2, and YTHDF3. We observed no significant changes in the level of ORF50 mRNA upon METTL3 or YTHDF3 depletion (Fig 6B and 6D). Although there was a consistent decrease in ORF50 mRNA in the YTHDF2 depleted cells, this may be due to the fact that YTHDF2 knockdown modestly decreased the viability of TREX-BCBL1 cells (S4 Fig). Surprisingly, however, METTL3 knockdown and, to a more variable extent YTHDF2 knockdown, resulted in increased ORF50 protein expression (Figs 6C, additional replicate experiments showing ORF50 levels in S4 Fig). YTHDF3 depletion did not significantly impact ORF50 or ORF59 protein (Fig 6C). Thus, unlike in iSLK cells, METTL3 and YTHDF2 appear to restrict ORF50 expression in TREX-BCBL1 cells. These phenotypic differences were not due to distinct virus-induced alterations in the abundance of METTL3, YTHDF2, or YTHDF3, as levels of these proteins remained consistent following lytic reactivation in TREX-BCBL1, iSLK.219, and iSLK.BAC16 cells (S5 Fig).
Finally, to determine whether the m6A pathway components impacted the outcome of the viral lifecycle in TREX-BCBL1 cells, we measured the impact of METTL3, YTHDF2, and YTHDF3 protein knockdown on virion production using a supernatant transfer assay. TREX-BCBL-1 cells lack the viral GFP marker, and thus infection of recipient cells was instead measured by RT-qPCR for the KSHV latency-associated LANA transcript. Again in contrast to the iSLK cell data, we observed that METTL3, YTHDF2, and YTHDF3 were dispensable for virion production in TREX-BCBL-1 cells (Fig 6E). Instead, METTL3 depletion consistently resulted in a modest, though not statistically significant, increase in the level of LANA mRNA in the recipient cells (Fig 6E). In summary, METTL3 and YTHDF2 appear to function in a pro-viral capacity and promote ORF50 expression in iSLK.219 and iSLK.BAC16 cells, but instead restrict ORF50 expression in TREX-BCBL-1 cells. These findings highlight how at least a subset of m6A pathway functions and targets may diverge between cell types.
Although m6A modification of viral RNAs has been recognized for more than 40 years, only recently are the contributions of this epitranscriptomic mark towards viral life cycles beginning to be revealed. Thus far, global epitranscriptomic analyses have documented m6A deposition during infections with KSHV, SV40, HIV, Influenza A virus and several members of the Flaviviridae, with a diverse set of resulting pro- and anti-viral roles [23–26,28,32,39,53]. The breadth and occasionally apparently contrasting functions for the m6A pathway during infection are perhaps unsurprising given the dynamic role for this modification in controlling mRNA fate and its ability to impact virtually every stage of host gene expression [1,27]. Our global analysis of the m6A epitranscriptome during lytic infection with the DNA virus KSHV showed the presence of m6A across multiple kinetic classes of viral transcripts and a general decrease in m6A deposition on cellular mRNAs. In the widely used KSHV-positive cell lines iSLK.219 and iSLK.BAC16, we found that depletion of several components of the m6A pathway inhibited the KSHV lytic cycle, most notably in iSLK.219 cells by restricting accumulation of the viral lytic transactivator ORF50. The YTHDF2 reader protein proved particularly important, as its depletion eliminated lytic entry and virion production. These observations are suggestive of a pro-viral role for m6A in the iSLK.219 and iSLK.BAC16 KSHV reactivation models. However, m6A marks on mRNA in a cell are widespread and contribute to a large variety of cellular and pathogenic processes that likely occur in a cell type or context-dependent manner. In this regard, it is notable that a distinct set of phenotypes was observed for m6A pathway components in the B cell line TREX-BCBL-1. Here, depletion of METTL3 and YTHDF2 increased ORF50 abundance, more suggestive of an anti-viral role. Thus, although KSHV engages the m6A pathway in multiple cell types, these findings underscore the importance of not broadly extrapolating m6A roles from a particular cell type, as this complex regulatory pathway can functionally vary in a cell type dependent manner.
What might be the basis for these phenotypic differences between cell types in the context of KSHV infection? m6A deposition was also recently reported in many KSHV mRNAs in BCBL-1 cells, including ORF50 [38]. Furthermore, while this work was in revision, Tan and colleagues documented extensive modification of KSHV transcripts during latent KSHV infection of multiple cell types, as well as upon lytic infection of iSLK.BAC16 and TREX-BCBL-1 cells [53]. Notably, while numerous differences were found in the cellular m6A profiles between the two cell lines, many peaks in viral transcripts were consistent across cell types, including two out of three m6A peaks in ORF50 [53]. In agreement with these studies, we also observed extensive modification of KSHV mRNAs, and observed that ORF50 is modified in iSLK.BAC16 cells, iSLK.219 cells and TREX-BCBL-1 cells. Thus, it is not the case that the viral mRNAs engage the m6A methyltransferase machinery in one cell type but not the other, although it is clear that site specificity of m6A deposition, particularly in host mRNAs, can vary between cell lines. The facts that m6A deposition is dynamic and does not strictly occur on consensus motifs render this possibility challenging to resolve. Indeed, how m6A deposition selectively controls gene regulation on particular transcripts or under particular stimuli remains a central unanswered question in the field [1]. We hypothesize that the distinct phenotypes derive either from how the viral modifications are ‘interpreted’ in each cell type and/or indirect effects driven by an altered m6A profile on cellular mRNAs. The recent finding that m6A modification of ORF50 in BCBL-1 cells contributes to efficient splicing through binding of YTHDC1 argues that modifications can have a direct cis-acting impact on KSHV mRNA fate [38]. However, herpesviral mRNAs are heavily reliant on host machinery at every stage of their biogenesis. Given that cellular mRNA fate is significantly altered upon depletion of METTL3 and the YTHDF reader proteins [1–3,18,54], it is possible that cell type specific changes in the abundance of a host factor(s) required for viral mRNA stability also contribute to the phenotypic differences. Furthermore, in HIV infected cells m6A modification and YTHDF proteins have been proposed to have a combination of pro-viral and anti-viral effects, including negatively impacting reverse transcription, enhancing mRNA export, and increasing viral protein production [24,26,28]. Therefore, the m6A pathway might similarly facilitate distinct phenotypes at different stages of the KSHV lifecycle.
Although our m6A-seq results are in agreement with the recent report from Tan and colleagues, our data on the role of YTHDF2 in iSLK.BAC16 cells differs from theirs [53]. They did not evaluate the impact of METTL3 depletion, but reported that YTHDF2 depletion increased KSHV replication in these cells. In contrast, we observed a significant reduction in virion production upon depletion of YTHDF2 in both iSLK.219 and iSLK.BAC16 cells. Given the similarity in approaches used to evaluate the impact of YTHDF2, the basis for these differences remains unclear. However, our experiments comparing the iSLK.219 and iSLK.BAC16 cells indicates that even in cell lines of the same origin there can be differences in the m6A-associated viral gene expression signatures.
As the ‘interpreters’ of m6A marks, the individual reader proteins play prominent roles in modulating gene expression. Generally speaking, in HeLa and 293T cells, YTHDF1 binding correlates with increased translational efficiency, YTHDF2 binding accelerates mRNA decay, and YTHDF3 may serve as a cofactor to assist the other reader protein function [1–4,19,20,54]. However, other roles for these factors are rapidly emerging, particularly in the context of cell stress, infection, or in the control of specific transcripts [7,13,15,23–26,28,54,55]. Furthermore, m6A is enriched in certain tissues, and different m6A patterns have been found depending on the tissue and developmental stage [42,56]. Intriguingly, a recent study showed that hypoxia increases global m6A content of mRNA, with many m6A modified RNAs exhibiting increased stability, raising the possibility that m6A deposition could also stabilize transcripts during other forms of cellular stress [57]. In KSHV-infected iSLK.219 cells, YTHDF2 appears essential for the post-transcriptional accumulation of ORF50, a role seemingly at odds with its more canonical mRNA destabilizing function. In this regard, it was recently revealed that SV40 late transcripts contain multiple m6A sites, and that YTHDF2 strongly promotes SV40 replication [39]. Thus, YTHDF2 has been shown to play a pro-viral role in the context of both DNA and RNA viruses. Although we observed less dramatic viral gene expression phenotypes upon METTL3 depletion, it nonetheless was required for WT levels of progeny virion production in iSLK.219 and iSLK.BAC16 cells. An important consideration may be that m6A factors differentially impact specific KSHV transcripts, or play different roles at distinct times during infection. However, dissecting these possibilities is likely to be complicated by the changes in ORF50 expression (either positive or negative), which will have ripple effects on the entire lytic life cycle. Another relevant question is the extent to which m6A mediates its effects on KSHV gene expression co-transcriptionally versus post-transcriptionally. A recent report indicated that m6A is primarily installed in nascent mRNA in exons and affects cytoplasmic stability, but not splicing [16,47]. It has also been demonstrated that m6A can be installed co-transcriptionally, and that slowing the rate of RNA Pol II elongation enhances m6A modification of mRNAs in a manner that ultimately decreases translation efficiency [16,47]. These add to a growing body of literature indicating that the position of m6A in a transcript is a key feature impacting the functional consequence of the modification [1,3,6,7,13,20]. For example, m6A in the 3’ UTR has been shown to recruit YTHDF1 and enhance translation initiation in HeLa cells, while deposition of m6A in the 5’ UTR has been shown to enhance 5’ cap independent translation [3,7,13]. Whether these position-linked effects on translation extend to viral transcripts remains to be tested, although there does not appear to be a consistent enrichment in a particular region of viral mRNAs for the viruses analyzed thus far. In KSHV, m6A sites are found throughout viral ORFs, some of which also overlap with untranslated regions of other viral transcripts. As KSHV transcription depends on the host RNA Pol II, the speed of transcriptional elongation on viral mRNAs likely impacts co-transcriptional deposition and positioning of m6A, and thus may ultimately regulate translation efficiency of a given mRNA. Thus, in the context of KSHV reactivation, a wide variety of mechanisms exist through which m6A modification could impact the transcription and translation of viral mRNA. Deciphering these remains an important challenge for future studies, as we are currently in the early stages of understanding how this and other viruses interface with the m6A RNA modification pathway.
The renal carcinoma cell line iSLK.puro containing a doxycycline-inducible copy of ORF50, and the KSHV infected renal carcinoma cell lines iSLK.219 and iSLK.BAC16 bearing doxycycline-inducible ORF50 [41] were cultured in Dulbecco’s modified Eagle medium (DMEM; Invitrogen) with 10% fetal bovine serum (FBS; Invitrogen, HyClone) and 100 U/ml penicillin-streptomycin (Invitrogen). The KSHV-positive B cell line TREX-BCBL-1 containing a doxycycline-inducible version of ORF50 [52] was cultured in RPMI medium (Invitrogen) supplemented with 20% FBS, 100 U/ml penicillin/streptomycin, and 200 μM L-glutamine (Invitrogen). HEK293T cells (ATCC) were grown in DMEM (Invitrogen) supplemented with 10% FBS. To induce lytic reactivation of iSLK.219 cells, 2x106 cells were plated in a 10 cm dish with 1 μg/ml doxycycline (BD Biosciences) and 1 mM sodium butyrate for 72 hr. Lytic reactivation of TREX-BCBL-1 cells was achieved by treatment of 7x105 cells/ml with 20 ng/ml 2-O-tetradecanoylphorbol-13-acetate (TPA, Sigma), 1 μg/ml doxycycline (BD Biosciences), and 500 ng/ml ionomycin (Fisher Scientific) for 72 hr (western Blot blots for viral gene expression), or for 120 hr (supernatant transfer experiments).
For iSLK.219 cells, 100 pmol of siRNA was reverse transfected into 5x105 cells plated in a 6-well dish using Lipofectamine RNAimax (Life Technologies). 24 hr post transfection, cells were trypsinized and re-seeded on a 10 cm plate. The next day, a second transfection was performed on the expanded cells with the same concentration of siRNA (400 pmol siRNA and 2x106 cells). The following day, cells were lytically reactivated in a 10 cm plate. 24 hr post-reactivation, cells were lysed in RIPA buffer (10 mM Tris-Cl (pH 8.0), 1 mM EDTA, 1% Triton X-100, 0.1% sodium deoxycholate, 0.1% SDS, 140 mM NaCl) to evaluate knockdown efficiency. siRNA experiments in iSLK.BAC16 cells were conducted with the same siRNAs and concentrations. For experiments to assess mRNA and protein levels at 24 hr post-reactivation, one round of siRNA knockdown was performed 48 hr prior to reactivation, and knockdown efficiency was evaluated at the time of cell harvest. For iSLK.BAC16 supernatant transfer experiments, two rounds of siRNA treatment were used, as described for iSLK.219 cells.
For TREX-BCBL-1 cells, 200 pmol of siRNA was nucleofected into 2x106 cells using Lonza Cell Line Nucleofector Kit V and a Lonza Nucleofector 2b set to Program T001. After nucleofection, cells were immediately resuspended in 2.2 ml of RPMI media in a 12 well plate. 48 hr later, 200 pmol of siRNA was added again to 2x106 cells using the same protocol. 48 hr after the second transfection, cells were lysed in RIPA buffer and knockdown efficiency was analyzed by Western Blot. Cell viability post-nucleofection was assessed using a Countess II Automated Cell Counter (Life Technologies) with Trypan blue staining. For RT-qPCR experiments, two rounds of siRNA knockdown were performed under the identical conditions, except using an Invitrogen Neon Nucleofector with a single pulse of 1350 volts and pulse length of 40 ms.
The following Qiagen siRNAs were used: SI00764715 and SI04279121 targeting YTHDF1, SI04205761 targeting YTHDF3, custom siRNA targeting METTL3 (sequence targeted: CTGCAAGTATGTTCACTATGA). The following Dharmacon siRNAs were used: SMARTpool siGENOME (M-021009-01-0005), targeting YTHDF2, and siGENOME Non-Targeting siRNA Pool #1 (D0012061305). These same siRNAs were used in all three cell lines for the experiments in Figs 3–6. In addition, independent siRNAs (Qiagen SI04174534 targeting YTHDF2, Qiagen SI00764778 targeting YTHDF3 and Qiagen SI03650318 (negative control siRNA)) were used in S2 Fig.
Assays were performed as previously described [58]. Briefly, for iSLK.219 and iSLK.BAC16 cells, viral supernatant was collected 72 hr post-reactivation, filtered, and added to uninfected HEK293T cells by spinfection (1500 rpm, 90 minutes at room temperature). 12 hr later, supernatant was removed and replaced with fresh media, whereupon the cells were assessed for the successful transfer of the GFP-containing KSHV BAC 24 hr post-infection using a BD Accuri C6 flow cytometer. Briefly, cells were trypsinized, fixed in 4% paraformaldehyde, washed twice in PBS and resuspended in FACS Buffer (PBS with 1% FBS). Uninfected HEK293T cells were used to define the GFP negative population. The percentage of GFP expressing cells was quantified using FlowJo Software (FlowJo LLC). For virus produced in TREX-BCBL-1 cells, supernatant transfers were performed as in iSLK.219 cells, except the virus was transferred to HEK293T cells at 120 hr post-reactivation. To quantify virus produced in TREX-BCBL-1 cells, RNA was extracted from HEK293T cells 48 hr post-supernatant transfer, and viral gene expression was quantified by RT-qPCR using primers specific for LANA.
Cell lysate was collected and analyzed as previously described [58]. Briefly, iSLK.219, iSLK.BAC16 or TREX-BCBL-1 cells were trypsinized, washed with PBS and lysed in RIPA buffer with protease inhibitors. After washing, 4X Laemmli sample buffer (Bio-Rad) was added to samples to elute bound proteins. Lysates were resolved by SDS-PAGE and western blots were carried out with the following antibodies: rabbit ORF50 (gift of Yoshihiro Izumiya, UC Davis), rabbit α-K8.1 (1:10000, antibody generated for this study), rabbit α-ORF59 (1:10000, antibody generated for this study), rabbit α-METTL3 (Bethyl, 1:1000), rabbit α-YTHDF1 (Proteintech, 1:1000), rabbit α-YTHDF2 (Millipore, 1:1000), rabbit α-YTHDF3 (Sigma, 1:1000), and goat α-mouse and goat α-rabbit HRP secondary antibodies (1:5000; Southern Biotech).
Following siRNA knockdown and 24 hr reactivation, iSLK.219 cells were pulse labeled with DMEM containing 500 μM 4sU (Sigma) for 30 minutes, followed by PBS wash and immediate isolation of total RNA with TRIzol. 4sU isolation was performed as previously described [59]. 4sU isolated RNA was analyzed by RT-qPCR.
Total RNA was harvested using TRIzol following the manufacturer’s protocol. Samples were DNase treated using Turbo DNase (Ambion), and cDNA was synthesized from 2 μg of total RNA using AMV reverse transcriptase (Promega), and used directly for quantitative PCR (qPCR) analysis with the DyNAmo ColorFlash SYBR green qPCR kit (Thermo Scientific). All qPCR results were normalized to levels of 18S or GAPDH as indicated, and WT or scramble control set to 1. RT-qPCR primers used in this study are listed in S4 Table.
Total RNA was isolated from iSLK.219 cells with TRIzol reagent. Dynabeads mRNA purification kit (Ambion) was used to isolate polyA(+) RNAs from 100 μg of total RNA according to the manufacturer. 100–200 ng of polyadenylated RNA was spiked with 10 μM of 5-fluorouridine (Sigma) and digested by nuclease P1 (1 U) in 25 μL of buffer containing 25 mM NaCl and 2.5 mM ZnCl2 at 42°C for 2–4 hr, followed by addition of NH4HCO3 (1 M, 3 μL) and bacterial alkaline phosphatase (1 U) and incubation at 37°C for 2 hr. The sample was then filtered (Amicon 3K cutoff spin column), and 5 μL of the flow through was analyzed by liquid chromatography (LC) coupled to an Orbitrap-XL mass spectrometer (MS) equipped with an electrospray ionization source (QB3 Chemistry facility).
Total cellular RNA (containing KSHV RNA) was extracted and purified by TRIzol and then DNAse treated with Turbo DNase (Ambion). 30 μl protein G magnetic beads (Invitrogen) were blocked in 1% BSA solution for 1 hour, followed by incubation with 12.5 μg affinity-purified anti-m6A polyclonal antibody (Millipore) at 4°C for 2 hr with head-over-tail rotation. 100 μg purified RNA was added to the antibody-bound beads in IP buffer (150 mM NaCl, 0.1% NP-40, and 10 mM Tris-HCl [pH 7.4]) containing RNAse inhibitor and protease inhibitor cocktail and incubated overnight at 4°C with head-over-tail rotation. The beads were washed three times in IP buffer, and then RNA was competitively eluted with 6.7 mM m6A-free nucleotide solution (Sigma Aldrich). RNA in the eluate was phenol chloroform extracted and then reverse transcribed to cDNA for Real-Time qPCR analysis.
High-throughput sequencing of the KSHV methylome (m6A-seq) was carried out following the previously published protocol [60]. In brief, 2.5 mg total cellular RNA was prepared from iSLK.219 cells that were either unreactivated, or reactivated for five days with doxycycline. RNA was isolated and DNAse treated as in the m6A RIP, except the RNA was first fragmented to lengths of ~100 nt prior to immunoprecipitation with anti-m6A antibody (Synaptic Systems). Immunoprecipitated RNA fragments and comparable amounts of input were subjected to first-strand cDNA synthesis using the NEBNext Ultra RNA Library Prep Kit for Illumina (New England Biolabs). Sequencing was carried out on Illumina HiSeq2500 according to the manufacturer’s instructions, using 10 pM template per sample for cluster generation, TruSeq SR Cluster kit v3 (Illumina), TruSeq SBS Kit v3-HS (Illumina) and TruSeq Multiplex Sequencing primer kit (Illumina). A reference human transcriptome was prepared based on the University of California, Santa Cruz (UCSC) and a reference KSHV transcriptome based on KSHV 2.0 annotation [61]. Analysis of m6A peaks was performed using the model-based analysis of ChIP-seq (MACS) peak-calling algorithm. Peaks were considered significant if their MACS-assigned fold change was greater than four and individual FDR value less than 5%. Sequencing data are available on GEO repository (accession number GSE104621). Raw reads and alignment to the viral genome are shown in S3 Table.
All results are expressed as means +/- S.E.M. of experiments independently repeated at least three times, except where indicated. Unpaired Student’s t test was used to evaluate the statistical difference between samples. Significance was evaluated with P values <0.05.
|
10.1371/journal.pntd.0001314 | Impact of Aetiological Treatment on Conventional and Multiplex Serology in Chronic Chagas Disease | The main criterion for treatment effectiveness in Chagas Disease has been the seronegative conversion, achieved many years post-treatment. One of the main limitations in evaluating treatment for chronic Chagas disease is the lack of reliable tests to ensure parasite clearance and to examine the effects of treatment. However, declines in conventional serological titers and a new multiplex assay can be useful tools to monitor early the treatment impact.
Changes in antibody levels, including seronegative conversion as well as declines in titers, were serially measured in 53 benznidazole-treated and 89 untreated chronic patients in Buenos Aires, Argentina with a median follow-up of 36 months. Decrease of titers (34/53 [64%] treated vs. 19/89 [21%] untreated, p<0.001) and seronegative conversion (21/53, [40%] treated vs. 6/89, [7%] untreated, p<0.001) in at least one conventional serological test were significantly higher in the benznidazole-treated group compare with the untreated group. When not only complete seronegative conversion but also seronegative conversion on 2 tests and the decreases of titers on 2 or 3 tests were considered, the impact of treatment on conventional serology increased from 21% (11/53 subjects) to 45% (24/53 subjects). A strong concordance was found between the combination of conventional serologic tests and multiplex assay (kappa index 0.60) to detect a decrease in antibody levels pos-treatment.
Treatment with benznidazole in subjects with chronic Chagas disease has a major impact on the serology specific for T. cruzi infection in a shorter follow-up period than previously considered, reflected either by a complete or partial seronegative conversion or by a significant decrease in the levels of T. cruzi antibodies, consistent with a possible elimination or reduction of parasite load.
| The main criterion for treatment effectiveness in Chagas Disease has been the seronegative conversion of previously reactive serology, generally achieved many years post-treatment. The lack of reliable tests to ensure parasite clearance and to examine the effect of treatment is the main difficulty in evaluating treatment for chronic Chagas disease. Decreases of conventional and non-conventional serological titers can be useful tools to monitor the early impact of treatment. We serially measured changes in antibody levels, including seronegative conversion as well as declines in titers in 53 benznidazole-treated and 89 untreated chronically T. cruzi-infected subjects. Seronegative conversion as well as decreases of titers was significantly higher in treated compared with untreated patients. A strong concordance was found between decreases of titers of conventional and non-conventional serologic tests post-treatment, reaffirming the findings. When seronegative conversion plus decreases of titers were considered altogether, the impact of treatment was higher, in a shorter follow-up period than previously considered. New tools for monitoring the effectiveness of treatment of chronic Chagas disease are necessary, and the results showed in this study is a contribution to researchers and physicians who assist patients suffering from this disease.
| The chronic form of Chagas disease, now considered to be globalized is prevalent in Latin America as well as in countries where T. cruzi-infection is not endemic [1]–[3]. In Argentina, T. cruzi infection is diagnosed by the conventional serologic tests indirect immunofluorescence assay (IFI), indirect hemagglutination (IHA) and ELISA that measure the level of circulating antibodies against Trypanosoma cruzi antigenic components (generally intact or whole parasite lysates), with 2 positive tests out of the 3 performed being required to confirm T. cruzi-infection [4]. This ‘best 2 out of 3’ approach is standard for diagnosis of T. cruzi infection in most endemic countries.
Even though long term follow-up of chronic Chagas disease patients treated with benznidazole showed that specific chemotherapy can prevent the progression of the heart-related pathology of Chagas disease in several [5]–[7], but not all studies [8], a wider use of the available drugs has not been achieved. One of the main limitations in evaluating treatment for chronic Chagas disease is the lack of reliable tests to ensure parasite clearance and to evaluate the impact of treatment on the evolution of lesions in target tissues. The main criterion for “cure” has been the conversion to negative serology on all tests performed [9]. However, this result is often not observed until 8 to10 years post-treatment and then only in approximately 15% of treated adult subjects [5]–[7], [10]. Clearly, the identification of better surrogate markers of parasite load decrease or complete parasite elimination is needed.
We have recently shown that changes in T. cruzi–specific T cell responses and in antibody responses to individual parasite proteins as determined using a multiplex assay can be used as early and effective predictors of the impact of drug treatment in human Chagas disease [11]. In addition, several studies have shown that a fall in conventional serological titers might be also useful to indicate treatment impact [5], [7], [9], [12]. However, these protocols have not been exhaustively explored in short post-treatment follow-up periods. In this study, the changes in antibody levels specific for T. cruzi were measured by conventional serologic assays and by a new multiplex assay during a relatively short follow-up period after treatment with benznidazole in chronically T. cruzi infected subjects.
Three hundred and twenty eight patients assisted in the first consultation at the Health Care Section for Chagas disease at the Hospital Eva Perón, Buenos Aires, Argentina were prospectively evaluated between 2004 and 2009. For inclusion in the study, a new serological evaluation by IFI, IHA and ELISA assays was conducted at the "Instituto Nacional de Parasitología “Dr. Mario Fatala Chaben”, the National Center of Reference for diagnosis of T. cruzi infection in Argentina. Of these 328 subjects who had a previous positive serology for T. cruzi in primary health care centers, 37 were excluded based upon negative tests at the Instituto Nacional de Parasitología “Dr. Mario Fatala Chaben”. Fifteen patients with discordant serology, 2 patients with associated coronary artery disease, 1 patient with HIV co-infection, and 1 patient living in an endemic area without vector control were also excluded from the present study. Adult individuals older than 21 years of age with 3 positive serological tests for T. cruzi infection, without heart disease (Group 0) or with mild heart disease (Group I), and serological follow-up of at least 24 months were included in this study. One hundred and thirty patients did not achieve the minimum follow-up period and were excluded for further analysis.
Aetiological treatment with benznidazole was offered to all patients, following the recommendations of the national and international guidelines for Chagas disease [13]–[15]. Benznidazole was administered at a dose of 5 mg/kg/day for 30 days. According to the principle of intention to treat, all patients, including those who did not complete the 30 day-treatment period (17%), were considered in the treated group.
A baseline electrocardiogram (ECG) and a 2-D echocardiogram were performed to stratify the patients according to a modified Kuschnir classification as follow: Group 0 has positive serology on 3 conventional serological tests, normal ECG and normal echocardiogram; and Group I includes subjects with positive serology on 3 tests, abnormal ECG (arrhythmias or conduction disturbances) and echocardiogram without left ventricular dilation or dysfunction [16]. Patients in group II (with dilation and left ventricular dysfunction) and in group III (with heart failure) were not included due to the low number of subjects with these conditions.
Continuous measures were expressed as mean and standard deviation or median and interquartile range of 25–75%, whereas dichotomous variables were expressed as a result/total and percentage. Chi square test or Fischer's exact test was performed to evaluate differences between discrete variables, whereas Student t test was applied in the study of continuous numerical variables. A Spearman rank correlation test and kappa index were applied to compare the decreases in antibody levels specific for Trypanosoma cruzi after treatment with benznidazole by the Multiplex and conventional serological tests.
Sensitivity, specificity and positive likelihood ratios (sensitivity/1-specificity) for detecting decreases of titers were determined for the Multiplex assay, and for the 3 conventional serological tests taken together or each serological taken separately. For the multivariate analysis, the logistic regression model was used, including all variables with significance (p<0.05) in the Chi square test and Student t test, as well as gender and years of residence in endemic areas for being considered clinically relevant.
The relative risk with a 95% confidence interval and the number needed to treat to achieve a change in post treatment serology was determined. All statistical analysis was performed using Analytical Software Statistix 8.0. The protocol was approved by the review Boards of the Eva Perón Hospital. Signed informed consent was obtained from all individuals before inclusion in the study.
We included 142 patients, 85 female and 57 male, mean ages 42.2±8.4 years. Fifty three patients (37%) were treated with benznidazole while 89 remained untreated (63%). Patients who agreed to receive treatment were marginally younger (mean age treated group = 39.7±8.4 vs. mean age untreated group = 43.7±8.1, p = 0.006) and had a higher likelihood of ECG abnormalities compared with untreated subjects (14/53, 26% of treated patients vs. 4/89, 4% of untreated patients, p<0.001). The median time and interquartile range (25–75%) of follow-up was 36 months (32–48) for the entire study population [48 months (36–60) for treated and 36 months (27.5–48) for untreated patients, p<0.001]; sex (male in treated group = 23/53, 43% vs. male untreated group = 34/89, 38%, p = 0.54) and years of residence in endemic areas (treated = 14.4±9.9 vs. untreated = 14.5±9.1, p = 0.94) were not significantly different between treated and untreated patients.
In order to evaluate treatment success, seronegative conversion and also decreases of titers in at least one test were determined in the present study. The frequency of decrease in titers as well as the conversion to negative results on one or more conventional serologic tests was significantly greater in the treated patient group compared with untreated individuals (Table 1 and Figure 1). Complete seroconversion was found in 4 out of 53 (8%) while seronegative conversion on 2 tests was observed in 7 out of 53 (13%) treated subjects. Because the standard for determining a diagnosis of T. cruzi infection as a positive result is based on at least 2 out of 3 conventional serological tests, we extended the criteria of cure to the seronegative conversion on 2 tests, as well. Thus, conversion to negative serology on 2 or 3 tests was observed in 11 out of the 53 treated subjects (21%).
IHA titers decreased in 22 out of 53 (41%) treated patients and in 8 out of 89 (9%) untreated subjects, p<0.001. Mean absorbance by ELISA decreased in 18 out of 53 (34%) treated and in 6 out of 89 (7%) untreated subjects, p<0.001. Finally, a decline in titers by IFI assays was observed in 18 out of 53 (34%) treated and in 7 out of 89 (8%) untreated subjects, p<0.001. In treated subjects, the median follow-up period to detect a decline in antibody levels was 27 months (interquartile range 25 to 75% 16.5–38.5), while a median of 24 months (interquartile range 25 to 75% 21–27.7) was required to identify seroconversion on 2 or 3 tests.
The aetiological treatment with benznidazole was the only variable that correlated with the decline of titers or seronegative conversion as determined by multivariate analysis (Table 2). None of the treated or untreated patients showed progression of the heart disease (new ECG changes or evolution to more severe clinical stages) during the follow-up period of this study.
Altogether, these findings show that taking into account either conversion to negative serology on 2 or 3 tests and decreases of titers on 2 or 3 tests, the impact of treatment on conventional serology might increase to 45%.
We have previously shown that treatment with benznidazole decreased antibody responses specific for a group of T. cruzi recombinant proteins as detected by a multiplex assay [11]. Decreases of antibody titers by conventional serology, as defined in Material and Methods, and by multiplex analysis were compared in a subset of individuals, 32 benznidazole-treated and 12 untreated subjects (50% male/female with a mean age of 40.7±7.7) during the follow-up period of the study. Seventy five percent of the subjects assessed showed similar trends with respect to the combined 3 conventional serologic tests and the multiplex assay (simple concordance = 75%; Kappa index = 0.60). In addition, a positive correlation was found between decreases in antibody titers measured by conventional serological tests and the multiplex (r = 0.59, p<0.001). Overall, 26/32 treated subjects (81%) and 1/12 untreated subjects (8%) showed a decline in antibody responses considering both conventional serological tests and the multiplex assay (Table 3). Although the overall results in this set of subjects were similar using the multiplex and combined conventional serological tests, the multiplex test was superior in detecting serological changes when compared to each conventional serologic test separately (i.e. 7 subjects showed titer decreases by multiplex while no alterations were recorded by ELISA, Table 3).
This study shows the unquestionable impact of aetiological treatment with benznidazole on conventional and multiplex serology for T cruzi-infection by three years postreatment. A significant proportion of treated patients reached the extended criteria of cure (conversion to seronegative on 2 or 3 tests) presented herein, while an even larger number of subjects showed declines in the titers on one or more tests compared with untreated adult subjects in this relatively short follow-up span. The strong correlation between the combined 3 conventional serologic tests and the multiplex assay, that measures antibody levels to individual recombinant proteins, also provides support for the impact of benznidazole treatment on chronic T. cruzi infection.
Cure rates of 8–40% in a 10–20 year-average of follow-up have been reported in the late chronic phase of adult patients who received etiological treatment for T. cruzi infection [18]. Work from our group showed a conversion to negative serology in only 15% of benznidazole-treated subjects after a 10-year follow-up study with serological tests performed every 3-year interval of time [6]. It is noteworthy that in the present study we were able to detect a 7% rate of seronegative conversion in less than half the time of our former studies. The more frequent sampling might likely improve the detection of both seronegative conversion and decreases in serological titers. In agreement with our findings, a recent publication has also described complete seroconvertion of 5% in a 3-year follow-up study after treatment with benznidazole [19]. Because the standard for determining a diagnosis of T. cruzi infection is based on at least 2 tests out of 3 conventional serological tests performed, we considered the conversion to negative serology on 2 or 3 tests as a possible indicator of cure. Although, spontaneous cure is a rare event documented in long-term untreated T. cruzi-infected subjects [20], [21], in this short-term study we have not recorded any case.
A reduction range of 28% to 34% in conventional serological titers in benznidazole-treated subjects was also demonstrated [7]. Likewise, declines in serological titers after treatment with benznidazole in chronically T. cruzi-infected subjects were shown by using modified conventional serological tests with recombinant antigens [12].
Herein, we found that a significant proportion of benznidazole-treated subjects showed changes in serological titers, from complete or partial conversion to negative serology by conventional tests to decreases in antibody levels determined by not only by conventional serological tests but also by the multiplex assay. The most plausible explanation for these observations is that those subjects who showed a decline of titers have achieved a reduction in parasite load or even complete parasite elimination, but a longer observation time is required to achieve complete seroconversion to negative. Concurring with this notion, we had previously documented that decreases in serological responses in benznidazole-treated subjects were associated with better clinical outcomes compared with untreated subjects, even when complete conversion to negative serology had not been achieved [5], [22]. More recently, we have shown that in the majority of benznidazole-treated subjects T. cruzi-specific IFN-γ- producing T cells significantly decreased or became negative along with decreases in antibody responses measured by the multiplex assay in a 5-year follow-up study, other likely indicators of treatment efficacy [11], [23]. Likewise, a reduction in parasitemia, as determined by PCR, was recently documented in chronic patients treated with benznidazole in a one year follow-up study, although serological changes were not reported in this short-term period. In this same study, PCR shifted to positive in around 10% of treated subjects showing that parasites were not completely eradicated in these subjects [24].
Serum antibodies have the capacity to be maintained for prolonged periods even in the absence of antigen making it challenging to correlate treatment efficacy with the measurement of changes in serology [25]. In this regard, antibody responses to viruses as varicella-zoster, measles and mumps in humans have been shown to be maintained for more than 50 years (in the case of smallpox even longer) [26]. Memory B-cell-independent antibody production by long-lived plasma cells has been proposed as the main mechanism to maintain long-term humoral immunity [27], [28]. The lifespan of long-lived plasma cells depends on external survival signals received in survival niches, mainly located in the bone marrow, although a small proportion of long-lived plasma cells persist in the spleen of both humans and mice [29]. However, under chronic inflammatory conditions, plasma cells are also found in inflamed tissues [29]. Therefore, a decrease in parasite burden in benznidazole-treated subject might be responsible for the elimination of survival niches in parasitized tissues limiting the lifespan of long-lived plasma cells with the subsequent decline in antibodies titers.
According to our and other author's findings, it can be proposed the notion of “impact of treatment” to extend the concept of successful treatment, by considering not only complete seroconversion but also conversion to negative serology in 2 tests, plus the decreases of titers in 2 or 3 conventional serological tests or as measured by the multiplex assay. Applying, the criteria proposed herein, when decreases of titers and negative seroconversion on 2 and 3 serological tests were taken into account (impact of treatment, Table 1), the effectiveness of treatment is significantly higher, reaching close to 45%.
Of note, the multiplex assay was able to detect the same level of decreases in serological titers as the overall decreases detected by the 3 conventional tests in combination, indicating the potential of this novel technique as a single and reproducible marker of treatment impact. Furthermore, the multiplex assay detected serological decreases following treatment with benznidazole in several cases in which conventional serological tests did not vary and some subjects showed decreases by multiplex as soon as 2–6 months post-treatment, as well.
While the main purpose of aetiological treatment is the complete elimination of T. cruzi, a diminution in parasite load might be relevant from the clinical standpoint. The available data support the conclusion that Chagas disease is the result of the failure of the immune system to completely clear this persistent infection, and the effect of decades of immune assault [30], [31]. One of the possible mechanisms is that the constant antigen stimulation might lead to a process of immune exhaustion which in turn might damper the ability of the immune system to control the infection with the subsequent disease progression [32], [33]. However, other regulatory pathways might also be involved in parasite control during the chronic infection [34]–[36]. Therefore, even a reduction in parasite load might restore, to some extent, the functionality of the immune system to keep the parasite under control with stable clinical conditions [37], [38].
Another scenario that emerges from the results presented herein is that completely unchanged serological findings after 3 years of treatment might be suggestive of treatment failure, while a decline of titers may indicate that a curative response is ongoing. Future studies might confirm whether unchanged conventional or multiplex serological assays or unaltered T cruzi-specific T cell responses following treatment reflect treatment failure, which might support the possibility of a new round of treatment, or the use of alternative drug treatments [39], [40].
The main limitation of this study is the non-randomized design of it, and thus rendering the findings of lesser strength compared with a randomized design.Our results should be also extrapolated with caution due to the possibility that diverse T cruzi lineages might have different susceptibility to benznidazole and thus treatment effectiveness may vary among endemic regions of Latin-American [41].
In summary, treatment with benznidazole in subjects with chronic Chagas disease has a major impact on the serology specific for T. cruzi infection in a shorter follow-up period than previously considered, reflected either by a complete negativization or by a significant decrease in the levels of T. cruzi antibodies, consistent with a possible elimination or reduction of parasite load.
|
10.1371/journal.pgen.1008244 | Congenital lipodystrophy induces severe osteosclerosis | Berardinelli-Seip congenital generalized lipodystrophy is associated with increased bone mass suggesting that fat tissue regulates the skeleton. Because there is little mechanistic information regarding this issue, we generated "fat-free" (FF) mice completely lacking visible visceral, subcutaneous and brown fat. Due to robust osteoblastic activity, trabecular and cortical bone volume is markedly enhanced in these animals. FF mice, like Berardinelli-Seip patients, are diabetic but normalization of glucose tolerance and significant reduction in circulating insulin fails to alter their skeletal phenotype. Importantly, the skeletal phenotype of FF mice is completely rescued by transplantation of adipocyte precursors or white or brown fat depots, indicating that adipocyte derived products regulate bone mass. Confirming such is the case, transplantation of fat derived from adiponectin and leptin double knockout mice, unlike that obtained from their WT counterparts, fails to normalize FF bone. These observations suggest a paucity of leptin and adiponectin may contribute to the increased bone mass of Berardinelli-Seip patients.
| Berardinelli-Seip congenital generalized lipodystrophy is a human disorder associated with increased bone mass suggesting that fat, per se, regulates the skeleton. To test this hypothesis we generated a murine model of congenital generalized lipodystrophy in which both brown and white adipose tissue are entirely depleted during embryogenesis. These “fat-free” (FF) exhibit a marked increase in bone mass throughout their body. The increased bone mass represents stimulation of bone formation and not retarded bone breakdown. Additionally, the increased bone mass of FF mice markedly increases skeletal strength and resistance to fracture. Like patients with congenital lipodystrophy, FF mice are diabetic but their metabolic state does not contribute to their increased bone mass. To identify the fat-produced molecules regulating bone mass we transplanted genetically modified adipose tissue into FF mice which established that absence of the fat-produced molecules, adiponectin and leptin, significantly enhances bone formation. These observations suggest that reducing the combined effect of adiponectin and leptin, on bone, will increase its abundance and fracture resistance.
| The past decades have witnessed elegant studies of the relationship of energy metabolism and bone, concluding they are mutually regulatory. By this scenario, selected adipokines influence skeletal mass by directly and indirectly targeting osteoblasts and osteoclasts [1–4]. These studies, many of which provide conflicting data, typically involved pharmacologically or genetically altering adipokine abundance. The realization that different depots of white adipose tissue (WAT) have distinct physiological effects provided insight into this enigma [5–7]. While subcutaneous fat, residing predominantly in thighs and buttocks, correlates with enhanced bone mass, increased visceral fat, which characterizes the metabolic syndrome, is associated with osteoporosis [1–3, 6]. Why this distinction occurs is, however, completely unknown nor is there mechanistic proof of a cause/effect relationship between fat and abundance of bone.
We reasoned that a mouse mirroring the virtual fat depletion characterizing Berardinelli-Seip syndrome, with a robust bone phenotype, normalized by adipocyte transplantation, would be a reasonable venue to establish how "fat talks to bone". This venue would enable determination of the influence of individual fat depots as well as adipokine-modified adipocytes on the skeleton. Understanding the means by which fat, in its various forms, impacts bone, may provide a foundation for preventing and treating the skeletal complications of the metabolic syndrome. To this end, we generated mice completely deficient in WAT as well as brown adipose tissue (BAT), also postulated to enhance bone mass [8]. While other forms of murine lipodystrophy, such as the A-ZIP/F1 mouse are associated with some degree of increased skeletal mass, our fat-free (FF) mice are extremely osteosclerotic due to profound osteogenesis, despite an abundance of osteoclasts [2]. Consistent with a sympathetic contribution in both circumstances, reduction of ambient temperature partially reduces the osteosclerosis of fat-depleted mice as well as their WT counterparts. Importantly, transplantation of WT adipocyte precursors or mature white or brown fat into FF mice completely rescues their trabecular skeleton. On the other hand, transplantation of adipose tissue derived from leptin and adiponectin double deficient mice fails to do so. Thus, due at least in part to a paucity of adipose derived leptin and adiponectin, congenital absence of fat induces severe osteosclerosis by stimulating bone formation. These observations provide insight into the mechanisms of enhanced skeletal mass in human congenital generalized lipodystrophy (CGL) and suggest that reducing the combined effect of adiponectin and leptin, on bone, will increase its abundance.
As described [9] generation of FF mice involved crossing those bearing a diphtheria toxin (DTA) transgene downstream of a floxed stop codon (DT-STOPfl/fl) to mice expressing adiponectin Cre (+/-). Cre- littermates serve as control. In keeping with the unique expression of adiponectin Cre in adipocytes and their precursors all Cre+ products of the mating contain no discernable WAT confirmed by virtually undetectable circulating adiponectin or leptin as well as visfatin and, resistin [10] (Fig 1). BAT is also absent and as a result, FF mice are cold intolerant as they require housing at 30°C to survive prior to weaning. Unexpectedly, the abundance of marrow adipocytes is unaltered in FF mice. Body weight of FF mice progressively diminishes with age, relative to WT, but the difference does not reach statistical significance until 20 weeks. There is no alteration of growth as evidenced by femoral length and FF mice have substantial splenomegaly (S1 Fig).
X-rays reveal the radiodensity of the FF skeleton is markedly increased (Fig 2A). This increase in trabecular bone mass is evident as early as 3 weeks post-partum. It maximizes at 2 months of age when BV/TV of FF mice is markedly greater than their WT littermates (Fig 2B and 2C, S2A Fig). The increase in FF trabecular bone mass is also evident histomorphometrically (Fig 2D). The osteosclerotic phenotype of fat-depleted mice is present in long bones and vertebrae (Fig 2E and 2F, S2B Fig).
Femoral cortical thickness of FF mice does not increase above control until 8 weeks and in fact, total area is reduced until that age (S3 Fig). The enhanced bone area and bone area/total area in 8 week old mice indicates that their decreased medullary area represents augmented cortical thickness due to accelerated endosteal MAR.
Augmented bone mass may represent enhanced osteogenesis and/or decreased resorption. Increased circulating osteocalcin as well as collagen1 α1, osteocalcin and osteopontin mRNAs, in bone, suggests stimulated osteoblast activity contributes to the osteosclerotic phenotype of FF mice (Fig 3A and 3B). To confirm such is the case, we administered time-spaced courses of calcein. Histomorphometric analysis reveals marked acceleration of trabecular bone formation in FF mice manifest by activity of individual osteoblasts (MAR) as well when expressed in the context of trabecular surface (BFR/BS) and total bone formation rate (BFR) (Fig 3C and 3D). In keeping with increased cortical thickness, osteoblasts lining the cortical endosteal surface of FF mice assume the columnar appearance of those actively synthesizing bone (Fig 3E). Supporting this conclusion, endocortical MAR is enhanced (Fig 3F) resulting in decreased medullary area (S3B Fig). Unlike endocortical, periosteal MAR is not significantly altered in FF mice (Fig 3G).
We next determined the quantity and percentage of marrow cells expressing leptin receptor (LepR) which are reported to provide a majority of osteoblast and adipocyte progenitors [11]. Predictably, in face of markedly decreased abundance of marrow, due to its replacement by bone, the number of LepR+ cells is substantially less in FF than WT mice (S4A Fig). On the other hand, the percentage of marrow cells expressing LepR is also markedly reduced in FF mice indicating selective diminution (S4B Fig). Unexpectedly, in face of total ablation of peripheral adipose tissue, the abundance of marrow adipocytes is unaltered in FF mice (S4C Fig). These observations posit that while adiponectin Cre/DT fails to target mature marrow adipocytes in FF mice, it diminishes LepR+ precursors. Preservation of marrow adipocytes and the abundance of bone forming osteoblasts, suggest that progenitors other than those expressing LepR are functional in FF mice. This conclusion is consistent with the observation that LepR+ MSCs serve as adipocyte and osteoblast precursors only in adult mice. Alternatively, recent single cell RNA-seq based evidence establishes LepR+ cells are heterogeneous with specific subsets representing osteogenic and adipogenic precursors [12]. Thus it is possible that germ-line fat ablation selectively depletes osteogenic LepR+ subpopulations while maintaining those which are adipogenic.
Bone remodeling is an ever-occurring event characterized by a tethering of osteoblast and osteoclast number. Thus, in keeping with the increase in bone formation in FF mice, circulating TRAP 5b, a classical marker of osteoclast abundance, is significantly higher than that of WT counterparts (Fig 4A). The increased number of osteoclasts is confirmed by histomorphometry (Fig 4B). Osteoclast specific mRNAs are also enhanced in FF bone (Fig 4C). Many FF osteoclasts fail, however, to attach to bone and have an irregular appearance suggesting cytoskeletal and resorptive dysfunction [13] (Fig 4D). While circulating CTx of WT and FF mice are equivalent (approximately 15 ng/ml each), serum CTx, normalized to the osteoclast abundance marker, TRAP5b, is reduced confirming the resorptive activity per individual cell is decreased (Fig 4E). Thus, osteoclastic bone resorption does not contribute to the skeletal phenotype of fat-depleted mice.
Increased bone mass normally amplifies the mechanical properties at the whole-bone (structural) level [14]. As such, three-point bending tests indicate that femora of 12week old FF mice are significantly stiffer (Fig 5A) with enhanced ultimate force (Fig 5B), the latter indicating superior whole-bone strength. These differences are consistent with increased bone area (Fig 5C) and cortical thickness (S3B Fig) although total area (S3B Fig) and moment of inertia (pMOI; Fig 5D) are not different. The morphological properties of FF femora reflect increased bone mass and are consistent with the enhanced endocortical but not periosteal bone formation noted above (Fig 3F and 3G). Bending tests also reveal FF femora have lower post-yield displacement (Fig 5E) and work-to-fracture (Fig 5F), indicating a more brittle phenotype. Thus, FF femora are stiffer and stronger at the whole-bone level, consistent with greater bone mass. On the other hand, FF bones have reduced post-yield displacement leading to reduced work-to-fracture and in keeping with more brittle material properties.
While our data confirm that fat depletion induces profound osteogenesis, we have not established whether the FF environment directly and/or indirectly stimulates osteoblasts. This is particularly relevant as regards the sympathetic nervous system (SNS) as its activation may dampen osteogenesis via β-adrenergic receptors [15]. Because heat regulates skeletal mass, most likely via SNS activity, we asked if the osteosclerosis of FF mice is impacted by the temperature to which they are exposed. Thus, at weaning, FF and control littermates were maintained as usual, at 30° or at room temperature (23°) for 3 months. Consistent with sympathetic regulation, bone mass of FF mice, maintained at 23°C, is diminished relative to those kept at 30°C (Fig 6A). On the other hand, comparable ambient temperature regulation of bone mass also occurs in control mice. Given these similar environmental effects in both genotypes it is unlikely that differences in sympathetic tone account for the profound osteosclerosis of FF mice.
Type 2 diabetes-associated humoral factors, such as insulin, have skeletal effects. In keeping with their lipodystrophic state, FF mice are insulin resistant as documented by abnormal insulin and glucose tolerance tests (Fig 6B–6D) and severe hepatic steatosis (see below) [9]. To determine if metabolic abnormalities likely contribute to their osteosclerotic phenotype, we fed metformin, standard therapy for lipodystrophy-associated diabetes, to 3 week old FF mice, maintained at 30°C [16]. The animals were sacrificed after 3 months. While the drug normalizes glucose tolerance and significantly reduces circulating insulin it has no effect on bone mass of FF mice (Fig 6B, 6E and 6F; S5 Fig). Interestingly, unlike ovariectomized rats housed at room temperature, whose skeletal mass increases in response to metformin, that of Cre- littermates of FF mice is unaffected [17]. Thus, the osteosclerosis of FF mice likely does not meaningfully reflect their metabolic dysfunction.
We hypothesized that the skeletal effects of various fat depots and genetically modified adipocytes may be clarified by their transplantation into mice with CGL. To this end we subcutaneously transplanted 3x106 mouse embryonic fibroblasts (MEFs) which differentiate into WAT, into 2 month old FF mice (Fig 7A and 7B) [18]. Within 4 months, transplantation of these adipocyte progenitors eliminates hepatic steatosis (Fig 7C) and completely reverses the trabecular osteosclerotic phenotype of the mutant mice (Fig 7D–7F; S6 Fig). To determine if the same occurs in the context of mature fat depots, we transplanted gonadal (visceral) or subcutaneous WAT or brown adipose tissue (BAT) into FF mice. Consistent with equivalent serum levels of adiponectin and leptin (S7A Fig), in each circumstance, trabecular bone mass, completely normalizes (Fig 7G and 7H, S7B and S7C Fig). Likely reflecting relative slow rate of bone remodeling, cortical thickness is unaffected after 4 months (S7D Fig). The rescue capacity of BAT likely does not reflect its effect on energy expenditure as mice with UCP1-Cre mediated BAT deletion exhibit no skeletal phenotype (S8 Fig). Thus, WT adipose tissue, regardless of origin, normalizes the FF skeleton suggesting mediation by commonly produced factors which, in the case of BAT, may reflect its “whitening” due to initial exposure to excess circulating lipids attending lipodystrophy [19] or transplanted BAT loss of sympathetic innervation.
Prior, albeit controversial, evidence indicates leptin and/or adiponectin exert skeletal effects [1, 2, 4]. Additionally, both adipokines are expressed by visceral, subcutaneous and brown adipocytes [20]. We therefore transplanted WT, leptin-/- (ob/ob), adiponectin-/- or double-deficient (DKO) WAT into FF mice. Three months later, the abundance of transplant-derived mutant fat was at least as great as WT (Fig 8A and 8B). Interestingly, in face of the capacity of transplanted WT WAT to completely normalize the FF skeleton, circulating leptin and particularly adiponectin, are substantially less in recipient mice than their naïve WT counterparts (Fig 8C). As expected, plasma of leptin-/- and adiponectin-/- grafted FF mice is completely devoid of their respective adipokine. While adiponectin-deficient grafts yield circulating leptin mirroring that of WT fat recipients, adiponectin is minimal in FF mice receiving leptin-/- fat. As leptin-/- fat is derived from obese ob/ob mice, the paucity of adiponectin is in keeping with the inverse relationship of the cytokine's expression and fat mass. μCT analysis established that absence of both leptin and adiponectin markedly reduces the capacity of WAT to normalize FF osteosclerosis (Fig 8D; S9A Fig). While a similar trend occurs in fat lacking either one of the cytokines, the distinction with WT fat transplantation is not significant. On the other hand, osteoclast number normalizes in FF mice regardless of adipokine expression (S9B Fig). Circulating TNFα is slightly but significantly increased in FF mice and normalized by fat transplantation (S9C Fig). Thus, absence of leptin and adiponectin, particularly when combined, contributes to the osteosclerotic phenotype of congenital generalized lipodystrophy.
Among the most controversial issues relating to the metabolic syndrome is the effect of obesity on bone, resolution of which will require an understanding of whether adipose tissue, per se, regulates skeletal biology and if so, what are the mechanisms? CGL is a rare disorder but potentially provides insights into the relationship of fat and the skeleton. For example, Berardinelli-Seip children grow rapidly and have advanced bone age. While not universal, bone mass of many CGL patients is substantially increased, a phenomenon which appears dictated by the specific causative mutation [21, 22].
We reasoned that generation of a completely fat-deficient mouse would permit use of adipose tissue transplantation to assess the effect of fat on the skeleton and most importantly, identify fat-residing bone-regulating molecules. To this end, we genetically targeted DTA to fat using adiponectin (Adipoq) Cre [7, 10]. All Cre+ mice are devoid of visible WAT and BAT. While other lipodystrophic mice, such as A-ZIP/F1, are osteosclerotic, we utilized FF mice because of their unique lack of any detectable peripheral fat depots [23]. Furthermore, for reasons unknown, marrow adipocyte abundance is unaltered in FF mice establishing that their osteosclerosis is mediated by peripheral adipose tissue, a conclusion confirmed by the complete rescue of their skeletal phenotype by gonadal, subcutaneous or brown adipose tissue. On the other hand, while FF mice mimic the phenotype of Berardinelli-Seip lipodystrophy, the pathogenesis of each differ as the human disorder is genetically based and that of the FF mouse represents adipocyte ablation [24].
Increased bone mass reflects enhanced osteoblast and/or retarded osteoclast function, an issue unresolved in lipodystrophic mouse models or patients. Bone formation is extremely robust in FF mice, a reflection of abundant osteoblasts and hyperactivity of the individual cell. This observation, taken with increased circulating osteocalcin, substantiates FF osteosclerosis is a manifestation of heightened osteogenesis.
Osteoclasts are abundant in FF mice. While circulating TNFα, a potent osteoclastogenic cytokine, is also significantly enhanced and normalized by transplanted fat, abundance of the cytokine, even in the naïve mutant mouse, is relatively low calling to question whether it contributes to increased osteoclast number [25]. While unproven, a more likely scenario holds that the abundant osteoblasts produce osteoclastogenic cytokines such as RANK ligand. Alternatively, the morphological features of cytoskeletal disorganization and diminished mobilization of CTx normalized to circulating TRAP5b, a marker of osteoclast number, indicate resorptive function of individual osteoclasts is compromised. Thus, net resorptive activity likely does not contribute to the FF bone phenotype. On the other hand, poorly resorbing osteoclasts may enhance skeletal mass by stimulating bone formation [26–28]. This event likely reflects the fact that osteoclasts promote osteogenesis not only by mobilizing matrix-residing osteoblast-activating growth factors, but also by secreting osteoblast-recruiting proteins independent of their resorptive activity. Thus, while the rate of bone formation of osteopetrotic mice, which lack osteoclasts, is suppressed, it is enhanced in those rich in dysfunctional polykaryons [28]. The same holds regarding therapy for pathological bone loss. Anti-resorptive drugs such as bisphosphonates which kill osteoclasts also suppress osteogenesis, whereas odanacatib, which impairs function of osteoclasts but does not diminish their abundance, substantially spares bone formation [29].
Among the most important contributions to understanding osteoblast biology is the discovery that their activity is sympathetically regulated and adipokines, at least in part, exert their skeletal effect by this mechanism [1, 15]. There is also evidence that maintaining mice at 23°C reduces their trabecular bone mass relative to those housed at 30°C [30]. Given that reduced ambient temperature activates the SNS, a reasonable hypothesis holds that FF mice, kept at room temperature, will have less bone than those maintained at thermoneutrality and we find such to be the case. On the other hand, the same occurs in WT mice indicating that while FF osteosclerosis may be SNS regulated, it likely occurs as a physiological event and not reflective of fat depletion.
As expected, given their lipodystrophic phenotype, FF mice are insulin resistant and steatotic. We propose these metabolic features add significance to the FF model as they replicate the complications of human CGL. Regarding a possible influence of the FF metabolic state on enhancing bone mass, evidence indicates the opposite. Specifically, fatty liver disease does not increase bone mass and in fact, promotes osteoporosis [31]. Moreover the bone mass of high fat-fed, insulin-resistant C57/BL6 mice is also decreased due to arrested bone formation [32]. Given its osteosclerotic properties and our observation that pharmacological modification of their diabetic state does not affect bone mass, it is unlikely that the metabolic syndrome of FF mice contributes substantially to their osteosclerosis and in fact may exert a negative influence superseded by robust osteogenesis.
The augmented bone mass in FF mice is contrary to the normal positive association between body size and bone mass. We recently showed in a large cohort of Large‐by‐Small advanced intercross (LG,SM AI) mice, which are WT but have a broad range of body mass due to genetic variation [33], that femur size and strength correlate positively with body and fat mass as well as serum leptin [34]. Thus, the osteosclerosis of FF mice is clearly counter to the normal physiology that links body size to bone mass. Nonetheless, biomechanical testing reveals that femora of FF mice are mechanically robust, with elevated stiffness and failure load in proportion to their increased mass. Thus, the excess bone in FF mice is mechanically competent. The only indication of possibly compromised bone properties in FF mice is decreased post-yield displacement and work-to-fracture, which reflects relatively brittle behavior of unclear origin.
We initiated this exercise to address two controversial issues relating to fat-mediated modulation of the skeleton. Both involved rescue of the CGL phenotype by fat transplantation which we first did using MEFs differentiating into adipocytes. Determining the capacity of subcutaneous and visceral WAT to alter FF osteosclerosis involved transplantation of WT individual depots into the affected mice. The fact that each depot completely normalizes FF bone suggests absence of a commonly-produced molecule(s) dampens bone formation. Moreover, reduction of bone mass by transplanted BAT challenges the commonly held position that these energy consuming adipocytes positively affect the skeleton [35]. The failure of UCP1 deletion, in adipose tissue, to impact bone mass, indicates that cure of FF osteosclerosis by BAT does not involve uncoupled energy expenditure.
A number of adipokines are proposed to impact the skeleton, most prominently, leptin and adiponectin which are virtually absent in CGL patients [36]. Despite great interest in these proteins, their skeletal properties are controversial. There is evidence that leptin diminishes bone mass by sympathetic activation via hypothalamic targeting [1, 15]. Other studies, however propose a direct effect on osteoblasts. In keeping with our observation that LepR+ marrow cells are diminished in FF mice, deletion of these cells promotes osteogenesis [37]. Alternatively, some argue that the cytokine actually increases bone mass [38–40]. The skeletal properties of adiponectin are equally complex with a majority of in vivo reports indicting osteoblast suppression while in vitro studies propose osteoblast stimulation [41]. Like leptin, both central and direct targeting of osteoblasts, by adiponectin, are postulated [4].
Transplanted WT fat completely normalizes FF bone in face of a paucity of circulating leptin and adiponectin relative to naïve WT mice. Thus, minimal expression of these adipokines may be sufficient to mediate the inhibitory effects of fat on osteogenesis and their absence likely contributes to lipodystrophy-associated osteosclerosis although lack of other adipocyte expressed factors, such as PPARϒ, may participate [2, 42, 43]. While surprising, this observation is in keeping with the capacity of small amounts of leptin to reverse the metabolic complications of lipodystrophy and the substantially greater bone mass in FF relative to other lipodystrophic mice with some residual fat [2]. On the other hand, individual deletion of either cytokine is not as effective as absence of both in maintaining the enhanced bone mass of FF mice. Whether this distinction reflects minimal expression of the other adipokine remains to be determined.
Animal work was performed according to the policies of Animal Studies Committee (ASC) at Washington University School of Medicine in St. Louis. Mice were analyzed under approved protocols and were provided appropriate care while undergoing research which complies with the standards in the Guide for the Use and Care of Laboratory Animals and the Animal Welfare Act.
Fat Free (FF) mice were generated by mating homozygous Lox-stop-Lox-ROSA-DTA mice to those expressing adiponectin-Cre. BAT-deficient mice were generated by mating homozygous Lox-stop-Lox-ROSA-DTA mice to those expressing UCP1-Cre. Although no gender differences exist in phenotype, male mice were exclusively used. Adiponectin-/-, leptin -/- mice were purchased from Jackson laboratory. Adiponectin/Leptin double knock-out (DKO) mice were created by mating Adiponectin-/- and Leptin +/- animals.
Metformin was purchased from MP Biomedicals (Santa Ana, California) and dissolved in mouse drinking water (2 g/L) for an equivalent dose of 300 mg/kg per day, based on estimates that mice drink 1.5 mL/10 g body weight per day [44]. Fresh drinking water with Metformin was changed daily for 3 months.
Blood was collected retro-orbitally under anesthesia immediately prior to sacrifice. Serum was obtained using serum separator tubes with lithium heparin (Becton Dickinson) and kept at -80C. ELISA kits were purchased from: Mouse adiponectin, leptin, adipsin, resistin and TNFα ELISA kits from R&D; Mouse Viafatin and Insulin ELISA kit from Ray biotech (Norcross, GA); Serum CTx-1 and TRAP5b kits from Nordic Bioscience; osteocalcin ELISA kit from Biomedical Technologies Inc.
Total RNA from fresh bone was extracted using Trizol following RNA purification with RNeasy RNA purification kit and RNase free DNase digestion (Qiagen). Complementary DNA (cDNA) was synthesized from 1 μg of total RNA using the iScript cDNA synthesis kit (Bio-Rad). Quantitative qPCR was performed using the PowerUp SYBR Green Master Mix kit (Applied Biosystems) according to the kit instruction and gene specific primers. All genes amplicon length is less than 150 nucleotides. PCR reactions for each sample were performed with 7500 fast Real-Time PCR System (Applied Biosystems, Foster City, CA, USA) using the comparative threshold cycle (Ct) method for relative quantification. The glyceraldehyde-3-phosphate dehydrogenase (GAPDH) gene was used as an endogenous control. The sequences of primers are: Itgb3: forward: TTCGACTACGGCCAGATGATT, reverse: GGAGAAAGACAGGTCCATCAAGT; Ctsk: forward: AGGCAGCTAAATGCAGAGGGTACA, reverse: AGCTTGCATCGATGGACACAGAGA; Trap: forward: CAGCTCCCTAGAAGATGGATTCAT, reverse: GTCAGGAGTGGGAGCCATATG; Dcstamp: forward: ACTAGAGGAGAAGTCCTGGGAGTC, reverse: CACCCACATGTAGAGATAGGTCAG; Nfatc1: forward: CCCGTCACATTCTGGTCCAT, reverse: CAAGTAACCGTGTAGCTGCACAA; Atp6v0d2: forward: CAGAGCTGTACTTCAATGTGGAC; reverse: AGGTCTCACACTGCACTAGGT; Bglap: forward: CTGACCTCACAGATGCCAAG, reverse: GTAGCGCCGGAGTCTGTTC; Opn: forward: GATTTGCTTTTGCCTGTTTGG, reverse: TCAGCTGCCAGAATCAGTCACT; Col1a1: forward: GAGCGGAGAGTACTGGATCG, reverse: GTTAGGGCTGATGTACCAGT; Gapdh: forward: AGGTCGGTGTGAACGGATTTG, reverse: TGTAGACCATGTAGTTGAGGTCA.
Bone marrow was isolated from the femur using the spin-flushing method, as previously described (Rohatgi et al., Blood Advances, 2018). Bone marrow cells from one femur/mouse were washed once in 200 μL PBS and then pelleted via centrifugation at 500 x g for 5 min at 4°C. Cells were resuspended in 50 μL of dead cell exclusion dye (ZombieUV, 1:600 in PBS, BioLegend) and incubated on ice for 10 min covered in foil. The ZombieUV stain was quenched with 200 μL FACS Buffer (PBS containing 2.5% heat-inactivated FBS and 2.5 mM EDTA), and cells were collected via centrifugation, as above. Cells were resuspended in 50 μL FcBlock (10 μg/mL of rat anti-mouse CD16/32 antibody, clone 2.4G2, BD Biosciences) and incubated on ice for 10 min covered in foil. An equal volume (50 μL) of 2X concentrate of primary antibody cocktail was added for a final staining volume of 100 μL. The primary antibody cocktail contained rat anti-mouse CD45-BUV395 (BD Horizon, clone 30-F11, final dilution factor 1:200), rat anti-mouse TER-119-APC (BioLegend, clone TER-119, 1:200), rat anti-mouse CD41-BV421 (BioLegend, clone MWReg30, 1:300), rat anti-mouse/human CD11b (BioLegend, clone M1/70, 1:400), and rat-anti mouse Leptin receptor (LepR)-biotin (R&D Systems, polyclonal, 1:50) in Brilliant Stain Buffer (BD Biosciences) containing 10 μg/mL FcBlock. Cells were stained for 30 min on ice and then washed 3 times in 200 μL FACS Buffer. The cells were resuspended in 100 μL of streptavidin-PE/Cy7 (BioLegend) and incubated on ice for 20 min before being washed 3 times in 200 μL FACS Buffer. Cells were resuspended in 200 μL FACS Buffer, and CountBright Absolute Counting Beads (Molecular Probes, 25 μL) were added to quantify total cell numbers using lot-specific bead concentration of 51,000 beads per 50 μL (Lot number 2014181). Flow cytometry data were analyzed with FlowJo version 10 (Becton, Dickinson & Company). LepR+ stromal cells were defined as singlet, live, CD45–, TER-119–, CD41–, CD11b–, LepR+ cells.
Glucose tolerance tests (GTT) were performed on 2 month old male mice in clean cages subjected to starvation with free access to water for 6 hours. Mice were weighed and a small amount of blood was obtained from tail vein for baseline (time 0) glucose measurement. Mice were then injected intraperitoneally with 50% sterile dextrose (1 mg/g body weight). Tail blood glucose was determined at 15, 30, 60, 90 and 120 min after challenge using a Bayer Contour glucometer. For insulin tolerance test (ITT), 2 month old male mice were placed in clean cages without food and free access to water. Following a 6 hr fast, the mice were weighed and baseline glucose reading was taken using Bayer Contour glucometer. Mice were injected intraperitoneally with human insulin (Humulin, Eli-Lilly) at a dose of 0.7U/Kg body weight and blood glucose measured at 15, 30, 45, 60, 90, 120 min after insulin injection.
Femur and Tibia were fixed in 10% neutral buffered formalin, followed by decalcification in 14% EDTA for 10 days, paraffin embedding, and TRAP staining. Static and dynamic histomorphometric parameters were measured using BioQuant OsteoII (BioQuant Image Analysis Corporation, Nashville, TN) in a blinded fashion.
2 month old male FF and control littermates were injected intraperitoneally with calcein (Sigma) (7.5 mg/kg of body weight) on days 6 and 2 before sacrifice. Non-decalcified histological sections of femur were analyzed using BioQuant OsteoII (BioQuant Image Analysis Corporation, Nashville, TN).
Trabecular bone was scanned using μCT40 scanner (Scanco Medical AG, Bassersdorf, Switzerland; 55 kVp, 145 μA, 300 ms integration time, 16 μm voxel size). A lower threshold of 250 was used for evaluation of all scans. The trabecular bone region of interest consisted of 100 slices that were drawn starting with the first slice in which condyles and primary spongiosa were no longer visible, constituting 1.6 mm in length. The region of interest of cortical analyses consisted of 50 slices covering a length of 0.8 mm at the femur midshaft.
Primary mouse embryonic fibroblasts (MEFs) were prepared from WT C57/BL6 E14 embryos as described [9, 45]. MEFs were injected subcutaneously at the sternum of 2 month old FF mice as reported [9, 46]. Mice were sacrificed 4 months after transplantation.
Mature fat depots were transplanted as described [47] with slight modification. 2 month old FF mice were anesthetized with isoflurane. Donor fat pads from 6–8 week old WT or adipokine deficient mice were placed into sterile PBS and cut into 100-150mg pieces. The grafts were implanted subcutaneously through small incisions in the shaved skin of the back, with 1 piece per incisions. 6 pieces of fat graft were implanted into each FF mouse. After surgery the mice were housed individually for a week and then 5 mice per cage. Mice were sacrificed 3 months after transplantation.
Femora (n = 5 per group) were scanned by microCT at the midshaft (Scanco uCT40; 70 kVp, 114 mA, 300 ms integration time, 10 um voxel size, 100 slices) to determine cross-sectional geometric properties. They were then mechanically tested to failure in three-point pending (Instron 8841; support span: 7 mm; displacement rate: 0.1 mm/sec). Failure occurred directly beneath the loading point, at the 50% length of the femur. Force-displacement data were collected and analyzed to determine whole-bone (structural) mechanical properties (stiffness, ultimate force, post-yield displacement, work-to-fracture) [14].
Statistical significance was determined using Student’s t test, one way or 2 way ANOVA test with Holm-Sidak post–hoc test with adjustment for multiple testing. Data are expressed as mean ± SD. *p < 0.05, **p < 0.01, ***p < 0.001 in all experiments.
|
10.1371/journal.pntd.0003659 | Pentoxifylline Reverses Chronic Experimental Chagasic Cardiomyopathy in Association with Repositioning of Abnormal CD8+ T-Cell Response | Chronic chagasic cardiomyopathy (CCC), the main clinical sign of Chagas disease, is associated with systemic CD8+ T-cell abnormalities and CD8-enriched myocarditis occurring in an inflammatory milieu. Pentoxifylline (PTX), a phosphodiesterase inhibitor, has immunoregulatory and cardioprotective properties. Here, we tested PTX effects on CD8+ T-cell abnormalities and cardiac alterations using a model of experimental Chagas’ heart disease.
C57BL/6 mice chronically infected by the Colombian Trypanosoma cruzi strain and presenting signs of CCC were treated with PTX. The downmodulation of T-cell receptors on CD8+ cells induced by T. cruzi infection was rescued by PTX therapy. Also, PTX reduced the frequency of CD8+ T-cells expressing activation and migration markers in the spleen and the activation of blood vessel endothelial cells and the intensity of inflammation in the heart tissue. Although preserved interferon-gamma production systemically and in the cardiac tissue, PTX therapy reduced the number of perforin+ cells invading this tissue. PTX did not alter parasite load, but hampered the progression of heart injury, improving connexin 43 expression and decreasing fibronectin overdeposition. Further, PTX reversed electrical abnormalities as bradycardia and prolonged PR, QTc and QRS intervals in chronically infected mice. Moreover, PTX therapy improved heart remodeling since reduced left ventricular (LV) hypertrophy and restored the decreased LV ejection fraction.
PTX therapy ameliorates critical aspects of CCC and repositioned CD8+ T-cell response towards homeostasis, reinforcing that immunological abnormalities are crucially linked, as cause or effect, to CCC. Therefore, PTX emerges as a candidate to treat the non-beneficial immune deregulation associated with chronic Chagas' heart disease and to improve prognosis.
| Chronic chagasic cardiomyopathy (CCC) is the main clinical manifestation of Chagas disease (CD), a neglected illness caused by the protozoan parasite Trypanosoma cruzi. More than hundred years after its discovery, CD continues to be a public health problem and millions of chronically infected people wait for an effective treatment. Chagasic cardiomyopathy is associated with CD8+ T-cell-enriched myocarditis, fibrosis and cardiac electrical and structural abnormalities, frequently progressing to heart failure. Presently, the available therapies only mitigate symptoms of CCC. Abnormalities in CD8+ T-cell compartment are present in CCC patients. Recently, we described the importance of CD8+ T-cells in the pathogenesis of CCC. Therefore, our proposal was to interfere with abnormalities of CD8+ T-cells glimpsing a better prognosis for CCC. Using PTX, an affordable drug with immunomodulatory properties on T-cells and cardioprotective effects in non-infections disease, we bring a therapeutic candidate for treating CCC. PTX therapy downmodulated detrimental CD8+ T-cells and promoted T. cruzi-specific interferon-gamma-producing T-cells. Importantly, chronic chagasic electrical and echocardiographic alterations were reversed by PTX therapy. Future studies may test the use of PTX combined with trypanocidal drug or as a vaccine adjuvant to improve the quality of life of chronic CD patients.
| Chagas disease (CD), a neglected tropical disease caused by the protozoan parasite Trypanosoma cruzi, affects 6 to 8 million people in Latin America [1]. The cardiac form, the most frequent clinical manifestation of CD, is characterized by fibrosis with remodeling of the myocardium and vasculature, which commonly progresses to heart failure [2]. The chronic chagasic cardiomyopathy (CCC) is a low-grade CD8+-enriched myocarditis occurring in an inflammatory cytokine-embedded milieu [3–5]. Abnormal CD8+ T-cell function may contribute to systemic inflammatory profile and cardiac tissue lesion in the chronic phase of T. cruzi infection [6–10]. Regardless their importance for T. cruzi host resistance [11], CD8+ T-cells gained particular attention as the major component of myocarditis in acute [12] and chronic [9,13] experimental T. cruzi infection and in chagasic patients with CCC [3,4,14]. Recently, we proposed that interferon-gamma (IFNγ)+ CD8+cells exert a beneficial role, whereas perforin (Pfn)+ CD8+ cells take part in T. cruzi-induced heart injury [9]. Additionally, we proposed that a proper therapeutic tool could interfere with distinct CD8+ T-cell populations hampering heart injury [9]. Indeed, CD8+ T-cell abnormalities and systemic inflammatory profile were reduced by administration of the anti-tumor necrosis factor (TNF) antibody Infliximab to a model of Chagas’ heart disease [15]. These findings unveiled that reversal of systemic immunological unbalance is a rational pathway to be explored to improve the prognosis of Chagas’ heart disease.
The methylxanthine pentoxifylline (PTX) is a phosphodiesterase inhibitor commonly used to treat peripheral vascular diseases. PTX also shows therapeutic potential as an anti-inflammatory and anti-tumor agent [16]. PTX has previously been proposed as an adjuvant therapeutic tool for leishmaniasis, a protozoan disease with an extensive inflammatory component [17]. Further, in non-infectious heart disorders PTX has shown cardioprotective effects in association with reduced plasma levels of TNF [18,19].
Given the lack of an effective specific therapy, CCC is treated similarly to all other heart failure syndromes using therapies to mitigate symptoms [2]. It is proposed that CCC pathogenesis relies on a parasite-driven systemic inflammatory profile, which may reverberate in the cardiac tissue and contribute to heart dysfunction [5,10,15,20,21]. Therefore, PTX arises as a therapeutic tool to interfere with immunological unbalance and to improve the progressive functional compromise of the heart in CD. Here we tested the effects of PTX on hallmarks of immunological and heart alterations detected in CD, using a model of CCC associated with high TNF expression and CD8+ T-cell abnormalities [9,15,22].
This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the Brazilian National Council of Animal Experimentation (http://www.cobea.org.br/) and the Federal Law 11.794 (October 8, 2008). The Institutional Committee for Animal Ethics of Fiocruz (CEUA-Fiocruz-L004/09; LW-10/14) approved all experimental procedures used in the present study. All presented data were obtained from three (D2–4) independent experiments (Experiment Register Books #41 and 49, LBI/IOC-Fiocruz).
Mice obtained from the animal facilities of the Oswaldo Cruz Foundation (CECAL/Fiocruz, Rio de Janeiro, Brazil) were housed under specific pathogen-free conditions in a 12-h light-dark cycle with access to food and water ad libitum. Five- to 7-week-old female C57BL/6 (H-2b) and C3H/He (H-2k) were intraperitoneally infected with 100 blood trypomastigotes (bt) of the Type I Colombian strain [23] of T. cruzi, and parasitemia was employed as a parameter to establish acute and chronic phases [13]. The chronically T. cruzi-infected C57BL/6 and C3H/He mice represent models of mild and severe CCC, respectively, paralleled to the degree of immunological abnormalities [22]. Sex- and age-matched noninfected (NI) controls were analyzed in parallel.
Chronically T. cruzi-infected C57BL/6 mice showing signs of CCC were intraperitoneally injected with saline (BioManguinhos/Fiocruz, Brazil) containing PTX (Trental, Sanofi-Aventis, Brazil) (20 mg/kg) or vehicle daily from 120 to 150 days postinfection (dpi).
For immunohistochemical staining (IHC), the polyclonal antibody recognizing T. cruzi antigens and supernatants containing anti-mouse CD8a (clone 53–6.7) and anti-mouse CD4 (clone GK1.5) were produced in our laboratory (LBI/IOC-Fiocruz, Rio de Janeiro, RJ, Brazil). Other antibodies included an anti-F4/80 polyclonal antibody (Caltag, USA); biotinylated rabbit anti-goat IgG cocktail (KPL, USA); polyclonal rabbit anti-connexin 43 (Cx43) (Sigma-Aldrich, USA), polyclonal rabbit anti-mouse FN (Gibco-BRL, USA), biotinylated anti-mouse CD54 (intercellular cell adhesion molecule-1, ICAM-1, BD Pharmingen, USA), biotinylated anti-rat immunoglobulin (DAKO, Denmark) and biotinylated anti-rabbit immunoglobulin and peroxidase-streptavidin complex (Amersham, UK). Monoclonal antibodies anti-mouse Pfn (CB5.4, Alexis Biochemicals, USA) and anti-IFNγ (R4–6A2, BD PharMingen, USA) produced in rat were also used in IHS. For flow cytometry studies, PE-Cy7-anti-mouse TCRαβ (clone H57–597), APC-conjugated anti-mouse CD8a (clone 53–6.7), FITC-anti-CD4 (GK1.5), PE-rat anti-mouse TNF (clone MP6-XT22), PerCP-anti-CD4 (clone GK1.5), FITC- conjugated anti-Pfn (11B11) and PECy-7-conjugated anti-IFNγ (clone XMG1.2) were purchased from BD Pharmingen (USA). PE-conjugated anti-CD107a (clone eBIO1D4B) was obtained from eBioscience. Anti-TNF receptor (TNFR)1 (TNFR1/p55/CD120a; clone 55R-286) conjugated to PE was purchased from BioLegend (USA). Appropriate controls were prepared by replacing the primary antibodies with the corresponding serum, purified immunoglobulin or isotype. All antibodies and reagents were used according to the manufacturers’ instructions.
Spleens were minced and the red blood cells were removed using lysis buffer (Sigma-Aldrich, USA). In a set of experiments, peripheral blood was also collected, as previously described [9]. The splenocytes and blood cells were labeled, events were acquired with a CyAn-ADP (Beckman Coulter, USA) and the data were analyzed with the Summit v.4.3 Build 2445 program (Dako, USA) as described elsewhere [9].
The ELISpot assay for the enumeration of IFNγ-producing cells was performed in triplicate as previously described [24]. Plates were coated with anti-mouse IFNγ (clone R4–6A2; BD PharMingen, USA) antibody diluted in PBS (5 μg/mL). Antigen-presenting cells were primed for 30 minutes at 37°C with total frozen extracts of epimastigote forms (Y strain) and amastigote surface protein 2 (ASP2) H-2Kb-restricted VNHRFTLV peptide [25]. After incubation, the freshly isolated splenocytes from experimental mice were seeded at 5 x 105 cells/well and incubated for 20 hours at 37°C and 5% CO2. Biotin-conjugated anti-mouse IFNγ antibody (clone XMG1.2; BD PharMingen, USA) was used to detect the captured cytokines. Spots were revealed after incubation of the samples with a solution of alkaline phosphatase-labeled streptavidin (BD PharMingen, USA) and a solution of NBT and BCIP (Sigma-Aldrich, USA) in Tris buffer (0.9% NaCl, 1% MgCl2, 1.2% Tris in H2O). The mean number of spots, in triplicate wells, was determined for each experimental condition. The number of specific IFNγ-secreting T-cells was calculated by estimating the stimulated spot count/106 cells using a CTL OHImmunoSpot A3 Analyzer (USA).
A mouse cytometric bead array (CBA) Inflammation Kit (Becton & Dickinson, USA) was used to quantify cytokines in the serum according to the manufacturer’s instructions. The fluorescence produced by the CBA beads was measured with a FACSCalibur instrument (Becton Dickinson, USA) and analyzed using FCAP Array software. Standard curves (1 pg/mL to 100 ng/mL) were generated in parallel. This method consistently detected concentrations above 10 pg/mL.
For real-time quantitative RT-PCR (RT-qPCR), the hearts were harvested, washed to remove blood clots, weighed and frozen in RNAlater (Life Technologies, USA). Total RNA (for gene expression studies) and DNA (for parasite detection) were extracted from the same sample using TRI-Reagent (Sigma-Aldrich, USA). For detection of TNF mRNA, the reverse transcriptase reactions were performed using a SuperScript III First Strand Synthesis Kit, and RT-qPCR was performed using TaqMan gene expression assays for TNF (# Mm00443258-m1) and the endogenous housekeeping control genes glyceraldehyde 3-phosphate dehydrogenase (GAPDH) (# Mm99999915-g1) and β actin (# Mm00607939-s1), purchased from Life Technologies (USA). Reactions were performed in duplicate according to manufacturer’s instruction, using cDNA template obtained from 2μg RNA. The conditions for the PCR were as follows: 95°C for 10 minutes, followed by 40 cycles at 95°C for 15 seconds and 60°C for 1 minute. Relative quantification of target gene levels was performed using the ΔΔCt method [26]. RT-qPCR data were normalized by the housekeeping genes GAPDH and β actin mRNA, using the Expression Suite Software V1.0.3 (Life Technologies, USA) and fold increase was determined in comparison with NI controls. For parasite detection 5 μL of purified DNA was analyzed by real time quantitative PCR (qPCR) using TaqMan system, with primers Cruzi 1 (5'-AST CGG CTG ATC GTT TTC GA-3'), Cruzi 2 (5'-AAT TCC TCC AAG CAG CGG ATA-3') and probe Cruzi 3: 6FAM-CACACACTGGACACCAA-MGB) targeting the T. cruzi nuclear satellite DNA, as previously [27]. As an internal amplification control, the TaqMan assay targeting mice glyceraldehyde 3-phosphate dehydrogenase (GAPDH) (# Mm99999915-g1, Life Technologies, USA) was used. Parasite load quantification was estimated by absolute quantification, following normalization by heart sample weight. The standard curve for the absolute quantification was generated by a 1:10 serial dilution of DNA extracted from the Colombian strain epimastigote culture stocks, ranging from 106 to 0.5 parasite equivalents.
In a flat bottom 96-well plate (Corning, Inc, USA) were distributed 100 μL of RPMI containing 5 x 106 bt/well of the T. cruzi. On parasites, were added 100 μL of different concentrations of PTX (0.3, 1, 3, 10, 30, 100 and 300 μg/mL), the positive (10 μM of the trypanocidal drug benznidazole) and negative (injection grade saline, Biomanguinhos/Fiocruz, Brazil) controls. After incubation for 24 hours at 37° C in an incubator containing constant tension of 5% CO2 (Shell Lab, USA), the number of parasites in the different treatment conditions was counted in a Neubauer chamber.
Eight to fifteen T. cruzi-infected and three to five NI animals were euthanized under anesthesia at 120 or 150 dpi and the hearts were removed, embedded in the tissue-freezing medium Tissue-Tek (Miles Laboratories, USA) and stored in liquid nitrogen. The phenotypes of the inflammatory cells colonizing the heart tissue and the T. cruzi parasitism were characterized and analyzed as previously described [9]. The ICAM-1-, FN- and Cx43-positive areas in 25 fields (12.5 mm2) per section (3 sections per heart) were evaluated with a digital morphometric apparatus. The images were digitized using a color view XS digital video camera adapted to a Zeiss microscope and analyzed with AnalySIS AUTO Software (Soft Imaging System, USA). According to the analyzed parameter, the data are shown as percent of positive area in the heart, as distance (μm) between stained gap junctions or as numbers of parasite nests or cells per 100 microscopic fields of view (400 X).
The activity of the creatine kinase cardiac MB isoenzyme (CK-MB) was measured using a commercial CK-MB Liquiform kit (Labtest, Brazil) according to the manufacturer’s recommendations as previously adapted for mouse samples [9].
Mice were tranquilized with diazepam (10 mg/kg) and transducers were placed subcutaneously (DII). The traces were recorded for 2 minutes using a digital Power Lab 2/20 system connected to a bio-amplifier at 2 mV for 1 second (PanLab Instruments, Spain). The filters were standardized to between 0.1 and 100 Hz and the traces were analyzed using Scope software for Windows V3.6.10 (PanLab Instruments, Spain). The ECG parameters were analyzed as previously described [9].
For analysis of cardiac function through echocardiography mice were anesthetized with 1.5% isoflurane gas in oxygen with flow 1L/minute, trichotomized in precordial region and examined with a Vevo 770 (Visual Sonics, Canada) coupled to a 30 MHz transducer. Cardiac geometry was made using two dimensional mode images acquired for measurement of internal area of heart cavities (right and left ventricles). M-mode images showed left ventricular (LV) muscle thickness was used for measurement LV mass. Heart and LV hypertrophy were measured by the ratios of heart weight (HW) and LV mass to body weight (BW), respectively. Left ventricular ejection fraction (LVEF) was determined using Simpson’s method and left and right ventricular (LV and RV) areas were obtained in B-mode using a short axis view at the level of the papillary muscles.
Data are expressed as mean ± SD. Analysis was performed using GraphPrism (GraphPad, USA). Comparison between groups was carried out by analysis of variance (ANOVA) followed by Bonferroni´s post-test or t-Student test when indicated. Differences were considered statistically significant when p<0.05.
To determine the effect of PTX on immune response in chronic experimental Chagas’ heart disease, PTX administration to C57BL/6 mice infected with the Colombian T. cruzi strain was initiated at 120 dpi (S1A Fig.). At this time-point, CD8-enriched myocarditis, splenomegaly, immune abnormalities, cytokine unbalance, electrical alterations and heart injury are already installed [9,15,22,28]. At 150 dpi, all PTX-treated infected mice were alive (S1B Fig.). Low parasitemia was detected in chronically infected mice administered with saline or PTX (6.8 ± 2.2 x 104 parasites /mL in saline-injected vs 6.4 ± 4.6 x 104 parasites /mL in PTX-treated; p>0.05). Further, PTX therapy had no effect on body weight (22 ± 1.3 g in NI controls; 21 ± 2.5 g in saline-injected vs 20.1 ± 0.8 g in PTX-treated infected mice; p>0.05). In comparison with sex- and age-matched NI controls, chronically T. cruzi-infected C57BL/6 mice presented splenomegaly (p<0.001), which is significantly (p<0.05) reversed by PTX therapy (S1C Fig.). Chronically T. cruzi-infected C57BL/6 mice have increased levels of TNF in the serum (S2A Fig.) and TNF mRNA in the heart tissue (S2B Fig.), corroborating previous data [15,22,28]. Further, at 150 dpi C57BL/6 mice have increased frequency of TNF-producing CD8+ T-cells in spleen (S2C Fig.). One of the proposed beneficial effects of PTX is its capacity of modulate TNF production [16]. However, PTX treatment from 120 to 150 dpi had no effect on T. cruzi-induced high TNF levels in the serum (S2A Fig.), TNF mRNA overexpression in the heart tissue (S2B Fig.) and TNF expression by CD8+ T-cells in spleen (S2C Fig.).
TNF signals via TNFR1/p55/CD120a and TNFR2/p75/CD120b [29]. In chronically T. cruzi-infected saline-injected C57BL/6 mice, there was a remarkable increase in the frequency of CD8+ TNFR1+ and CD8+ TNFR2+ T-cells in the spleen (S2D Fig.). PTX was previously shown to downmodulate TNFR1 expression by hepatic cells [30]. Importantly, PTX therapy (from 120 to 150 dpi) completely abrogated the elevated frequency of TNFR1+ CD8+ T-cells. Nevertheless, PTX therapy only partially reduced the high frequency of TNFR2-bearing CD8+ T-cells detected in chronically infected mice (S2D Fig.). These data support that PTX was selectively active in chronic experimental CD.
At 150 dpi, although a significant reduction in splenomegaly was noticed in PTX-treated mice (S1C Fig.), similar frequencies of CD8+ T-lymphocytes were detected in the spleen of NI controls and saline-injected and PTX-treated chronically infected mice (Fig. 1A). Further, CD8+ T-cells also express similar density of the CD8 molecule on cell surface in the studied groups (MFI in NI: 25.6 ± 0.6; saline-injected T. cruzi-infected: 20.8 ± 1.5; PTX-treated T. cruzi-infected: 23.7 ± 2.6). However, compared with NI controls, a considerable part of the splenic CD8+ cells of chronically Colombian-infected mice expressed TCRαβLow (Fig. 1B; p<0.001), corroborating previous data in a distinct model of chronic CD [6]. Importantly, PTX treatment significantly rescued the downregulation of TCR expression in CD8+ T-cells of chronically T. cruzi-infected mice (Fig. 1B), considering both frequency of TCRαβLow population (p<0.05) and density of TCRαβ on cell surface (p<0.05).
Next, we investigated whether PTX therapy influenced the expression of markers of naïve/memory/activation phenotypes of CD8+ T-cells during chronic T. cruzi infection using CD45, a tyrosine phosphatase essential for T-cell activation, and the expression of CCR7, a chemokine receptor associated with the homing of T-cells to lymph nodes [31]. Compared with NI controls, splenic TCR+CD8+ T-cells of chronically infected mice displayed low frequencies of CD45RA+CCR7+ (naïve; p<0.01) cells and CD45RA-CCR7+ (central memory; p<0.001) cells, but showed increased frequency of CD45RA-CCR7- (effector memory; p<0.01) cells (Fig. 1C). PTX treatment restored the frequencies of the minor CD45RA+CCR7+ T-cell subset (p<0.05), but did not interfere with the proportions of the major CD45RA-CCR7+ and CD45RA-CCR7- CD8+ T-cell populations (Fig. 1C). In an attempt to further dissect the immunoregulatory mechanism of PTX in chronic experimental CD, we analyzed the frequencies of naïve, memory and activated T-cells studying the expression of CD44, a chondroitin sulfate proteoglycan receptor associated with cell migration to peripheral tissues, and CD62L, a marker of T-cell homing to lymph nodes [31]. In chronic T. cruzi infection, there was a remarkable decrease in the frequencies of CD44-CD62L+ naïve (p<0.001) and CD44+CD62L+ central memory (p<0.01) but an increase in the frequency of CD44+CD62L- (p<0.001) CD8+ T-cells (Fig. 1D), in comparison with age-matched NI mice. Notably, PTX therapy partially restored these drastic alterations, significantly (p<0.05) increased the frequencies of CD44-CD62L+ and CD44+CD62L+ cells and decreased the proportion of CD44+CD62L- CD8+ T-cells in the spleen (Fig. 1D). However, there were no changes in the frequencies of CD8+ CD44+CD62L- (46.3 ± 3.3% in saline-injected vs 51.9 ± 5.5% in PTX-treated T. cruzi-infected mice; p>0.05) and CD8+ CD44+CD62L+ (15.4 ± 3.3% in saline-injected vs 10.7 ± 3.4% in PTX-treated T. cruzi-infected mice; p>0.05) T-cells in the blood of C57BL/6 mice.
Next, we explored the potential influence of PTX on the effector function of T. cruzi-specific total and CD8+ T-cells of chronically T. cruzi-infected C57BL/6 mice. Using ELISpot assay, we detected an increased (p<0.01) number of T-cells producing IFNγ after recognition of crude T. cruzi antigens (epimastigote extracts) in chronically infected mice. Also, the number of IFNγ-producing CD8+ T-cells specific for the immunodominant H-2Kb-restricted ASP2 VNHRFTLV peptide was increased (p<0.001) (Fig. 2A). PTX therapy did not alter the number of IFNγ-producing cells among splenocytes recognizing crude T. cruzi antigens, but upregulated (p<0.05) the number of IFNγ-producing ASP2-specific CD8+ T-cells (Fig. 2A). Having in mind the different CD8+ T-cell phenotypes, we evaluated the potential cytotoxic activity by CD8+ splenocytes of chronically infected mice, studying the expression of CD107a, a marker for T-cell degranulation [32], and inflammatory potential, studying intracellular IFNγ expression. At 150 dpi, in comparison with NI controls, there was a significant increase in the frequencies of IFNγ+ (p<0.01) and IFNγ+CD107a+ and CD107a+ (p<0.05) CD8+ T-cells in saline-injected infected mice (Fig. 2B and S3 Fig.). PTX therapy significantly increased the frequency of IFNγ+ CD8+ T-cells (p<0.01) and reduced the frequency of CD107a+ (p<0.01) CD8+T-cells (Fig. 2B and S3 Fig.). Considering the antagonistic roles for IFNγ+ and Pfn+ CD8+ T-cells in T. cruzi infection [9], we studied the effect of PTX on IFNγ and Pfn expression by CD8+ T-cells. PTX administration to chronically infected mice increased (p<0.05) the frequency of IFNγ+ cells, but reduced (p<0.05) the frequencies of IFNγ+Pfn+ and Pfn+ CD8+ T-cells (Fig. 2C). To investigate whether PTX effects on inflammatory IFNγ and cytotoxic Pfn+ cells were restricted to splenic compartment, we analyzed the numbers of IFNγ+ and Pfn+ inflammatory cells invading the heart tissue of chronically infected C57BL/6 mice. Saline-injected and PTX-treated chronically infected mice had similar numbers of IFNγ+ cells in the heart tissue (Fig. 2D). In comparison with saline injection, PTX therapy reduced (p<0.05) the number of inflammatory Pfn+ cells infiltrating the cardiac tissue of chronically infected mice (Fig. 2D).
Our previous data support that the formation of CD8-enriched chagasic myocarditis involves CCR1/CCR5-mediated cell migration [33,34]. Further, the CCR5 receptor and the cell adhesion molecule LFA-1 are co-expressed by peripheral blood mononuclear cells enabling them to migrate to heart tissue [33,35]. Here we described an increased (p<0.001) frequency of CD8+ T-cells co-expressing LFA-1 and CCR5 among CD8+ T-cells in chronically infected mice. Importantly, PTX treatment led to a significant reduction in the frequency of splenic LFA-1+CCR5+ CD8+ T-cells (Fig. 3A), when compared with saline injection. Moreover, a significant reduction in the frequency LFA-1+CCR5+ was also observed among circulating CD8+ T-cells after PTX therapy (3.14 ± 1.0% in PTX-treated vs 5.75 ± 2.2% in saline-injected T. cruzi-infected mice; p<0.05). Next, we analyzed the expression of ICAM-1, the LFA-1 ligand, on the cardiac endothelial cells [36]. ICAM-1 is upregulated (p<0.001) in the endothelial cells of heart blood vessels and cardiomyocytes of chronically T. cruzi-infected C57BL/6 mice (Fig. 3B and Fig. 3C). After PTX therapy, reduction (p<0.01) in ICAM-1 expression was noticed in cardiomyocytes and inflammatory cells infiltrating the heart tissue (Fig. 3C—upper panel) and, particularly, in the blood vessel endothelial cells (Fig. 3C—bottom panel). ICAM-1+ blood vessels with perivascular cuffs with several layers of inflammatory cells were commonly found in the heart of saline-injected but absent in PTX-treated T. cruzi-infected mice (Fig. 3B and Fig. 3C). At 150 dpi, the Colombian-infected C57BL/6 mice present myocarditis (Fig. 3D), mainly composed of CD8+ T-cells, corroborating previous findings [9]. Interestingly, the short term PTX therapy reduced (p<0.05) the intensity of the chronic T. cruzi-induced myocarditis (Fig. 3D).
The pathogenesis of Chagas’ heart disease is, at least in part, accounted to parasite persistence [2,37]. To bring further mechanistic insights into the beneficial effects of PTX in chronic infection, we analyzed a putative effect of PTX directly on T. cruzi trypomastigote forms and on heart parasitism. Contrasting with a significant (p<0.001) effect of the trypanocidal drug Bz (positive control), PTX showed no direct action on the survival of the trypomastigote forms of the parasite, in an in vitro assay (S4A Fig.). Meanwhile, PTX did not interfere with parasite control in chronically infected C57BL/6 mice, as rare T. cruzi amastigote nests (S4B Fig.) and low numbers of parasite DNA copies (S4C Fig.) were similarly (p>0.05) detected in the heart tissue of saline-injected and PTX-treated mice, at 150 dpi.
All the beneficial effects of PTX on the unbalanced immune response of chronically infected mice encouraged us to investigate the effects of PTX therapy on T. cruzi-induced chronic heart injuries. Chronically T. cruzi-infected mice showed a significant reduction (p<0.01) in Cx43 expression in the intercalary disc of myocardial cells, seen as increased distance of the Cx43-bearing gap junction plaques (Fig. 4A and Fig. 4B). Further, chronically infected mice presented FN overdeposition in the heart tissue (p<0.001, Fig. 4A and Fig. 4C), and CK-MB activity levels in the serum (p<0.05, Fig. 4D), compared with age- and sex-matched NI controls. In comparison with saline injection, PTX therapy ameliorated heart tissue injuries, improving Cx43 expression (p<0.05, Fig. 4A and Fig. 4B), decreasing FN overexpression (p<0.01, Fig. 4A and Fig. 4C) and reducing CK-M activity in the serum (p<0.01, Fig. 4D). Thus, PTX therapy hampered the progression of heart injury in chronically T. cruzi-infected C57BL/6 mice. Moreover, considering that significant increase in CK-MB activity is already detected in T. cruzi-infected C57BL/6 mice at 120 dpi [9], our data support that PTX therapy reversed cardiomyocyte injury.
After the demonstration that PTX restored major immunological abnormalities believed to be associated with the severity of Chagas’ heart disease [9,10,15,21,22], we examined the effect of PTX on electrical conduction in an experimental model of CCC. When compared with sex- and age-matched NI controls, saline-injected chronically infected C57BL/6 mice presented ECG alterations including prolonged P wave, PR interval and QRST complex (Fig. 5A). At 150 dpi, PTX-treated mice improved ECG alterations, compared with saline-injected mice (Fig. 5A). Notably, PTX had beneficial effects on heart rate (p<0.05), PR (p<0.001) and QRS (p<0.05) intervals in comparison with saline-injected animals (Fig. 5B). Actually, PTX therapy reduced the proportion of mice afflicted by arrhythmias (ART), second-degree atrio-ventricular block (AVB2) and other ECG abnormalities (Fig. 5C). At 120 dpi, ECG abnormalities are already detected in T. cruzi-infected C57BL/6 mice [9]; hence, our data support that PTX therapy reversed ECG alterations.
In parallel experiments, parasitemia, heart parasitism and inflammation were higher in Colombian-infected C3H/He compared with C57BL/6 mice, at 120 dpi [22]. In the model of severe infection (C3H/He), PTX therapy also ameliorated the expression of the biomarkers of heart injury Cx43 loss and FN deposition in the heart tissue, CK-MB activity levels in the serum (p<0.01, S5A Fig. and S5B Fig.) and ECG abnormalities (p<0.05, S5C Fig. and S5D Fig.). At 150dpi, the increased P duration, QTc and QRS intervals (p<0.05, S5C Fig.) and the proportions of mice afflicted by ECG abnormalities were diminished by PTX therapy (S5D Fig.).
At 150 dpi, all analyzed groups of C57BL/6 mice showed similar body weight (NI: 23.2 ± 1.3; saline-injected T. cruzi-infected: 20.6 ± 1.3; PTX-treated T. cruzi-infected: 22.5 ± 2.0). Chronic T. cruzi infection resulted in heart enlargement, shown as increased HW/BW ratio (Fig. 6A), corroborating our previous data [9]. The HW/BW coefficient of PTX-treated mice tends to decrease when compared with saline-injected infected mice and was similar to the HW/BW coefficient of sex- and age-matched NI controls (p>0.05; Fig. 6A). To assess whether chronic T. cruzi infection affects heart geometry and function, as well as the impact of PTX administration, all chronically infected mice underwent echocardiographic evaluation. At 150 dpi, compared with NI C57BL/6 controls, infected mice showed alterations in heart geometry, as increased LV mass (NI: 82.4 ± 2.9; saline-injected T. cruzi-infected: 100.7 ± 4.3; p<0.01). PTX therapy significantly reduced LV mass (PTX-treated T. cruzi-infected: 80.1 ± 4.3; p<0.01), restoring mass values akin to NI controls. The assessment of heart hypertrophy, as the ratio of LV mass to body weight, showed a remarkable LV hypertrophy (p<0.01) in chronically Colombian-infected mice (Fig. 6B). Importantly, PTX therapy significantly reduced the LV hypertrophy (p<0.01) of T. cruzi-infected mice (Fig. 6B). Additionally, compared with NI controls, saline-injected chronically infected mice showed higher right ventricular (RV) and LV areas (Fig. 6C). PTX-treated mice exhibited RV and LV areas similar to age-matched NI controls (Fig. 6C). When compared with NI age-matched controls, mice chronically infected by the Colombian strain showed significant decrease in LVEF (Fig. 6D). Notably, PTX therapy significantly (p<0.05) restored LVEF of chronically infected mice to values resembling NI controls (Fig. 6D).
Here we used PTX as a therapeutic tool in experimental CCC, exploring the effects of this immunomodulator upon key features of immunological and heart abnormalities. After PTX therapy, the overexpression of TNF remained unaltered, while TNFR1 and TNFR2 expression was reduced. PTX therapy also decreased the frequency of splenic cytotoxic (CD107a+ and Pfn+) CD8+ T-cells and the number of Pfn+ cells invading the cardiac tissue. Conversely, treatment with PTX increased the number of IFNγ-producing ASP2-specific CD8+ T-cells. Further, PTX-treated mice show reduced frequency of splenic and circulating LFA-1+CCR5+ CD8+ T-cells, decreased expression of ICAM-1 on cardiac tissue and less severe myocarditis. Importantly, PTX therapy ameliorated heart injury and dysfunction, without interfering with parasite control.
Splenomegaly is present in chronic CD patients [38]. In experimental T. cruzi infection, splenomegaly is associated with increased polyclonal T-cell activation, a hallmark of not regulated immune response [6,9, 38]. Notably, the splenomegaly seen in Colombian-infected C57BL/6 mice [9,15] was partially reversed by PTX therapy. This effect may result of the inhibitory effect of PTX on primary proliferative capacity of T-cells [39].
PTX ameliorates LVEF in patients with noninfectious heart failure in association with downmodulation of inflammatory biomarkers, such as TNF [18,19]. Elevated plasma TNF levels paralleled CCC severity in patients [21,40] and chronic experimental CD models [15,22]. During acute T. cruzi infection, PTX administration reduced the number of TNF+ cells in necrotic areas in the spleen [41]. Thus, a reduction in TNF expression was expected in our models of CCC subjected to PTX treatment. However, this was not the case as PTX did not affect TNF overexpression in the heart tissue or systemically (spleen and serum). The lack of effect of PTX on TNF expression in experimental CCC was not a surprise. Clinical trials assessing PTX in chronic heart failure showed no concordant action on downmodulation of TNF levels, despite clinical improvement and beneficial effects on biomarkers of heart lesion [16]. The mechanism by which PTX improves heart diseases remains unsolved [16,18,19]. TNF signals through two different receptors (TNFR1 and TNFR2) and affects diverse biological processes such as cell activation, proliferation, differentiation and survival [29]. Further, depending on the biological process TNF receptors may have opposite effects [42]. In CCC, PTX abrogated TNFR1 expression by CD8+ T-cells. Also, in ischemic-reperfusion hepatic injury the beneficial effect of PTX was associated with abrogation of the high expression of TNFR1 mRNA [30]. Thus, the downmodulatory effect of PTX on TNFR1 occurs independently of the trigger and target cell type. Additionally, in chronically infected mice PTX therapy partially reduced the upregulated expression of TNFR2 by CD8+ T-cells. Since TNFR2 receptor is involved in cell proliferation [29] and upregulated after T-cell activation [43], in reducing the frequency of TNFR2-expressing cells may reside, at least in part, the beneficial effect of PTX on splenomegaly during chronic T. cruzi infection. Therefore, akin the effect of anti-TNF therapy in CCC [15], PTX may emerge as a tool to disrupt TNF/TNFRs signaling pathway and restore immunological homeostasis and cardiac injury in experimental CCC.
CD8+ T-cells are the prominent inflammatory components in the cardiac tissue in Chagas’ heart disease [3,4,14], a feature reproduced in Colombian-infected mice [9,13,15,28,34]. In CD patients, CD8+ T-cells show abnormal activation phenotypes marked by low expression of CD8 and TCR [7]. Although in chronically Colombian-infected mice no changes in the frequency of splenic CD8+ T-cells and in the density of CD8 molecules on cell surface were detected, downmodulation of TCRαβ expression on CD8+ T-cells was noticed, corroborating previous data [6]. Importantly, in PTX-treated mice the frequency of TCR-bearing cells and the density of TCRαβ on cell membrane were restored to intensities similar to NI controls. In naïve T-cells, TCR complex is constitutively internalized and rapidly recirculate back to cell surface. However, by a molecular process not yet fully understood, antigen stimulation increases retention/degradation of TCR lowering the density of TCR on cell surface in association with reduced effector function [44, 45]. Therefore, in chronic T. cruzi-infection PTX therapy mice may interfere with the turnover of TCR and restore the capacity of CD8+ T-cells to respond to activation signals. This idea deserves to be explored.
In comparison with noninfected individuals, CD patients show significant increase in total effector/memory CD8+ T-cells (CD45RA−CCR7−), supporting a continuous stimulus by parasite antigens [7]. Similarly, increased frequency of effector/memory CD45RA−CCR7− CD8+ T-cells was detected in chronically infected C57BL/6 mice. PTX did not interfere with the frequency of these cells, supporting that they are potentially prone to control invasive pathogens [31]. Conversely, PTX therapy restored the naive CD45RA+CCR7+ CD8+ T-cell compartment, potentially seeking homeostasis and allowing immune response to new stimuli. Here we corroborated the findings that chronic T. cruzi infection reduced the frequency of CD62L+ (naïve/central memory) and increased the percentage of CD44+CD62L− (activated) cells among CD8+ T-cells [13,46]. Interestingly, PTX therapy decreased the frequency of CD44+CD62L− and increased the proportion of CD44−CD62L+CD8+ splenocytes. Therefore, PTX interferes with cell migration and increases retention of T-cells in secondary lymphoid tissues, major sites of antigen recognition [47]. In chronic T. cruzi infection, another remarkable disturb in cell migration scenario is the increased frequency of splenic and circulating LFA-1+CCR5+ CD8+T-cells [9,13,34]. These cells are potentially able to invade the heart tissue, where CCR5 ligands (CCL3/MIP-1α and CCL5/RANTES) are found [13,28,34]. As PTX therapy lowered the frequencies of splenic and circulating LFA-1+CCR5+ CD8+ T-cells in chronically infected mice, a reduction in myocarditis intensity was expected. Indeed, PTX decreased the number of inflammatory cells invading the heart tissue. T. cruzi infection increases the expression of ICAM-1 (ligand of LFA-1) on cardiac endothelial cells and cardiomyocytes [13,48], aiding T-cell entry into the heart tissue [35]. In chronically infected mice, PTX decreased the expression of ICAM-1on cardiac tissue. PTX was also shown to downmodulate ICAM-1 in a model of acute lung injury [49]. Therefore, in chronically infected mice PTX may reduce the CCR5-mediated chemotaxis and LFA-1/ICAM-1-mediated endothelial cell/lymphocyte interaction, explaining the decreased colonization of the heart tissue by inflammatory cells. CCR5 expression is related to CCC in patients [50] and myocarditis intensity and heart injury in infected mice [33,34]. In T. cruzi infection the majority of the detrimental Pfn+CD8+ cells are LFA-1+CCR5+ [9], hence we predicted a beneficial effect of PTX in experimental CCC in association with a selective reduction of the migration of Pfn+ cells to heart tissue.
In experimental T. cruzi infection, antigen specific Pfn+CD8+ T-cells may play a non-beneficial role, whereas IFNγ+CD8+ T-cells may exert a protective role in heart injury [9]. In CD patients, the frequency of IFNγ-producing CD8+ T-cells specific for T. cruzi antigens is inversely related with disease severity [7]. Therefore, we tested the impact of PTX on specific T-cell response in chronically infected mice. Notably, PTX therapy increased the number of IFNγ-producing anti-T. cruzi VNHRFTLV ASP2 effector CD8+ T-cells, a population shown to protect against T. cruzi infection [51]. Previously, PTX was shown to increase T-cell memory and protective immunity against Salmonella infection [52]. In refractory patients, PTX combined with antimonial drug ameliorated cutaneous leishmaniasis [17], supporting the use of PTX as immunological adjuvant. In chronically T. cruzi-infected mice, PTX therapy also increased the frequency of splenic IFNγ-producing CD8+ T-cells and reduced the proportion of CD8+ T-cells expressing CD107a, a marker for T-cell degranulation and cytotoxic activity [32]. Moreover, PTX treatment decreased the frequencies of Pfn+ and multifunctional Pfn+IFNγ+ CD8+ T-cells. Based on the possible antagonistic role of CD8+T-cells expressing IFNγ and Pfn, [9,15], we then analyzed the influence of PTX on the composition of chronic heart inflammation. PTX-treated mice had reduced number of Pfn+ cells, although the number of IFNγ+ cells remained unaltered. The heart infiltrating Pfn+ cells, probably CD8+ T-cells acting as CTLs, are involved in tissue damage in experimental CD [9,15]. Interestingly, in CD patients with severe cardiomyopathy the presence of cells expressing granzyme A (another component of the lytic machinery of CTL CD8+ T-cells) in heart lesions is in accordance with concepts that involve cytolysis in pathogenesis of CCC [9,53]. Further, there is a good correlation between the numbers of IFNγ+ and CD8+ cells infiltrating the heart tissue in CD patients presenting successful parasite control [53]. Actually, PTX did not influence cardiac or systemic parasite load, reinforcing that an effective immune response, which contributes to T. cruzi control, is preserved and disconnected from factors causing heart injury [9,15,33,34].
Altogether, these findings support that PTX therapy in experimental CCC reduced the ability of CD8+ T-cells to migrate and invade the heart tissue, which is less permissive to lymphocyte interaction. Further, PTX diminished the frequency of activated but increased the frequency of naïve CD8+ T-cells, which is paralleled by regain of TCR density on CD8+ lymphocytes and, apparently, restored the capacity to respond to new antigenic stimuli. Indeed, PTX therapy increased the number of CD8+IFNγ+ responsive to ASP2 T. cruzi antigen, while disfavored Pfn+ cells inside the heart tissue. Thus, PTX reestablished several aspects of the CD8+ T-cell alterations in chronically T. cruzi-infected mice. Considering that immunological abnormalities may contribute to cardiac alterations in experimental CCC [15, 54], we expected that the PTX-induced amelioration of the immunological unbalance in chronic T. cruzi infection would beneficially reverberate in the cardiac injury and dysfunction.
A loss of Cx43, the most abundant ventricular gap junction protein, is associated with arrhythmogenic disease [55]. The Cx43 loss may contribute to electrical conduction abnormalities in Chagas’ heart disease [56]. One important beneficial effect of PTX was the restoration of gap-junction Cx43 expression in chronically infected mice, therefore, indicating that Cx43 loss may be interrupted. In CD, overdeposition of FN discloses cardiac fibrosis [57]. PTX therapy hampered the progression of FN overexpression in experimental CCC; reinforcing the idea that in T. cruzi infection cardiac fibrosis can be improved, and even, reversed [15, 54, 58, 59]. The increased CK-MB activity in the serum, an important CCC feature and a biomarker of cardiomyocyte lesions [60], is increased in chronically infected mice before therapy, at 120 dpi [9]. PTX therapy also reduced and, moreover, reversed the increased CK-MB activity in the serum of experimental CCC. These findings support a broad beneficial effect of PTX on major features of T. cruzi-triggered heart injury.
To our knowledge, this study is the first demonstration that PTX improves electrical conduction and heart dysfunction in an infectious cardiomyopathy. In chronically infected C57BL/6 mice showing signs of CCC [9], PTX therapy ameliorated bradycardia, prolonged P wave duration, PR and QRS intervals, ART and AVB2. As previous shown in patients [2, 61] and T. cruzi-infected rhesus monkeys [59], higher ECG QRS scores directly correlated with the severity of heart fibrosis. Considering that electrical abnormalities and FN overdeposition were detected at 120 dpi [9,54], when therapy was initiated, our data support that PTX more than hampering progression is reversing electrical abnormalities and heart injury in experimental CCC. Moreover, the beneficial effects of PTX therapy were not restricted to a particularly experimental model, as amelioration of heart injury and electrical alterations were also observed in Colombian-infected C3H/He mice, a model of severe CCC [22]. Lastly, ECO studies revealed anatomical alterations with increased RV and LV areas, higher LV mass and decreased LVEF in chronically Colombian-infected mice, when compared with age-matched NI controls. Dilatation of the RV is a risk factor for sudden death in several cardiac diseases [62]. The enlargement of the RV, a marker for CCC in mice [63,64], was reduced by PTX therapy in chronic experimental CD. Increased LV internal dimensions emerged as a risk factor associated with morbidity and mortality in CD [65]. Further, increased LV mass, a marker of hypertrophy, is an independent risk factor of cardiovascular events [66]. Remarkably, PTX administration to mice with signs of CCC ameliorated the alterations in LV geometry and mass. Moreover, the reduced LVEF seen in chronically infected mice was improved to values detected in age-matched NI controls after PTX therapy. Similarly, PTX therapy improved LVEF in patients with heart failure due to ischemic cardiomyopathy [18,19]. Altogether, these data support that PTX therapy in chronic experimental CCC bettered cardiac tissue injury, electrical abnormalities and heart failure.
Here we bring evidence that immunological unbalance and Chagas’ heart disease are interconnected, involving multifactorial elements that may be working bidirectionally [9,15,22,33]. Further, whether the beneficial effects of PTX results of its immunomodulatory properties or direct action on heart tissue, particularly on cardiomyocytes, remains to be clarified. Therefore, our results opened a new avenue to be paved to explore PTX as adjuvant to immune protective response or cardioprotective tool in an infectious cardiomyopathy. More important, PTX emerges as a potent adjuvant to treat heart failure in CD. PTX might be a non fantasious strategy for CD immunotherapy, combined or not with trypanocidal drug, hampering the deleterious inflammation but preserving the beneficial anti-parasite immunity.
|
10.1371/journal.pntd.0005336 | Improving access to Chagas disease diagnosis and etiologic treatment in remote rural communities of the Argentine Chaco through strengthened primary health care and broad social participation | Rural populations in the Gran Chaco region have large prevalence rates of Trypanosoma cruzi infection and very limited access to diagnosis and treatment. We implemented an innovative strategy to bridge these gaps in 13 rural villages of Pampa del Indio held under sustained vector surveillance and control.
The non-randomized treatment program included participatory workshops, capacity strengthening of local health personnel, serodiagnosis, qualitative and quantitative PCRs, a 60-day treatment course with benznidazole and follow-up. Parents and healthcare agents were instructed on drug administration and early detection and notification of adverse drug-related reactions (ADR). Healthcare agents monitored medication adherence and ADRs at village level.
The seroprevalence of T. cruzi infection was 24.1% among 395 residents up to 18 years of age examined. Serodiagnostic (70%) and treatment coverage (82%) largely exceeded local historical levels. Sixty-six (85%) of 78 eligible patients completed treatment with 97% medication adherence. ADRs occurred in 32% of patients, but most were mild and manageable. Four patients showing severe or moderate ADRs required treatment withdrawal. T. cruzi DNA was detected by qPCR in 47 (76%) patients before treatment, and persistently occurred in only one patient over 20–180 days posttreatment.
Our results demonstrate that diagnosis and treatment of T. cruzi infection in remote, impoverished rural areas can be effectively addressed through strengthened primary healthcare attention and broad social participation with adequate external support. This strategy secured high treatment coverage and adherence; effectively managed ADRs, and provided early evidence of positive therapeutic responses.
| Less than 1% of patients infected with Trypanosoma cruzi have access to parasiticidal treatment with the two available drugs, including millions of patients in the early chronic phase who would greatly benefit from treatment. Furthermore, rural populations living under poor and marginalized conditions usually have little or no access to chemotherapeutic programs of Chagas and other neglected tropical diseases. Barriers to treatment range from misconceptions on medication-related risks to socio-cultural aspects, among others. This study addressed the challenge of diagnosis and treatment of T. cruzi infection in resource-poor, remote rural settings of the Argentine Chaco where 24% of residents up to 18 years of age were infected. The underlying premise was that participatory methods and multisector cooperation would increase program effectiveness. Our results show that diagnosis and treatment of T. cruzi infection there can be effectively addressed through strengthened primary healthcare attention and broad social participation with adequate external support, following an initial phase of intensified vector control and surveillance across the municipality. This strategy secured high diagnosis-and-treatment coverage and adherence; effectively managed adverse drug-related reactions; provided early evidence of a positive therapeutic response, and may stimulate healthcare services in the affected regions to aggressively bridge the treatment gap.
| Chagas disease ranks among the main neglected tropical diseases (NTDs) in Latin America and the Caribbean [1]. Trypanosoma cruzi, its etiologic agent, induces heart and digestive disease and reduces life expectancy in approximately 30–40% of the infected people [2,3]. The parasite infects 6–9 million people, the majority of which primarily were rural residents living in poverty with little access to healthcare services [4]. A well-known hotspot of Chagas disease and other NTDs is the Gran Chaco ecoregion which mainly extends over sections of Argentina, Bolivia, and Paraguay [5]. In rural villages across this region where Triatoma infestans is the only domestic vector, the seroprevalence of human T. cruzi infection in children younger than 15 years of age frequently ranged between 20% and 50% [6–12].
Prevention of human T. cruzi infection has traditionally relied on residual insecticide spraying campaigns and routine screening of blood-bank donors [2]. The two drugs (nifurtimox and benznidazole) registered for treatment of human infection with T. cruzi since the late 1960s and early 1970s were shown to be especially effective in young age groups during the acute and early chronic phase regardless of transmission mode [3,13–18]. Unfortunately, both nifurtimox and benznidazole cause adverse drug-related reactions (ADR) of various types, frequency and severity which increase with increasing patient’s age and reduce treatment compliance and effectiveness [19–22]. Benznidazole frequently causes mild or moderate dermatitis that respond well to antihistamines; low-dose oral glucocorticoids are less frequently needed. The rare cases presenting severe exfoliating dermatitis, dermatitis combined with fever and lymphadenopathy, and bone marrow suppression prompt immediate treatment discontinuation and intensive medical care [16,19,22,23]. Therefore, chemotherapeutic programs of human T. cruzi infection ideally should provide access to diagnosis and treatment as early as possible during the life course, minimize the occurrence of ADRs leading to reduced medication adherence, and avert eventual life threats posed by severe ADRs in the absence of timely medical attention.
Less than 1% of patients infected with T. cruzi have access to parasiticidal treatment [24]. Most populations living under poor and marginalized conditions often lack access to diagnosis and treatment of T. cruzi infection (and other neglected diseases) and are unaware of their condition [25]. They also ignore disease consequences and the opportunities and limitations of current therapies. Barriers to treatment are multiple and include lack of training on ADR management, misconceptions on medication-related risks, reluctance to provide treatment, wide fluctuations in medication availability, overburdened or distant healthcare services, socio-cultural aspects, and lack of effective vector control and surveillance [22,26–29].
The pioneering treatment programs of T. cruzi infection implemented by Médecines Sans Frontières (MSF) in various countries since 1999 demonstrated that the challenge was tractable with adequate resources and stringent procedures [21]. They proposed that “Etiological treatment of Chagas disease can and should be integrated at the primary health care level…” [21]. This recommendation has also been endorsed by others [3] and may be traced back to the Declaration of Alma Ata in 1978 [30]. The proposition is also related to the concept of innovative and intensified disease management (IDM) for NTDs that can be managed within the primary healthcare system through more intensive use of existing tools, as is the case of Chagas disease [31]. However, the challenge of how to address diagnosis and treatment of T. cruzi infection in resource-poor, remote rural settings through primary health care has yet to be developed and program effectiveness documented to meet the challenge of treating the sizable population of infected rural residents and correct health inequities.
The primary healthcare model focuses on community participation and social empowerment [30]. Broad social participation of multiple sectors may augment the feasibility and sustainability of control interventions, more so in disperse rural areas including various cultural groups [32–34]. Community participation is expected to increase the coverage, effectiveness and sustainability of vector and disease control actions of Chagas and malaria [35–39]. For example, in a remote rural area of Santiago del Estero (Argentine Chaco) under sustained community-based vector control in the mid-1990s, 17 (65%) of 26 T. cruzi-seropositive children aged up to 15 years of age treated with benznidazole or nifurtimox seroconverted to a negative status between 2 and 13 years posttreatment [39].
As part of a long-term program on the eco-epidemiology and control of Chagas disease in the Argentine Chaco, we developed, implemented and tested a strategy to increase access to diagnosis and treatment of human T. cruzi infection in sparsely populated rural sections of Pampa del Indio municipality including 13 villages. This strategy, based on strengthened primary healthcare attention and broad social participation, followed an initial phase of intensified vector control and surveillance across the municipality [40–42]. The underlying premise was that participatory methods and multisector cooperation combined with adequate external support would increase diagnosis-and-treatment coverage and adherence relative to historical local levels, manage ADRs effectively and achieve positive therapeutic responses, as we document in this paper.
The study protocol was approved and supervised by “Dr. Carlos Barclay Independent Ethical Committee for Clinical Research”, Buenos Aires, Argentina (Protocol N° TW-01-004). All clinical investigations were conducted according to the principles expressed in the Declaration of Helsinki. The Ministry of Health of Chaco province and local hospital authorities granted permission to conduct the activities herein described. All individuals participating in serosurveys and treatment accepted to do so and their parents or guardians provided written informed consent. When community meetings included indigenous residents, explanations were translated by an indigenous healthcare agent or by an appointed indigenous community member, and consent was obtained collectively and individually.
The intervention was conducted in Pampa del Indio (25°55’S 56°58’W), Chaco, Argentina. The study area included 353 houses and a few public buildings grouped in 13 rural villages distributed over a 450 km2 section as described elsewhere [40]. The study area was inhabited by 1,187 people in 2007 (1,318 people as of 2012, including 565 up to 18 years of age), most of which lived on a subsistence economy. The only existing medical facility was a first-level public hospital with four physicians; the primary healthcare system had 5 posts distributed across the study area, and there were 8 primary schools. The closest and farthest villages were at 6 and 37 km from the hospital through dirt roads, and there was no public transportation.
Initial selection of the study area followed the recommendations of the Chagas disease control program of Chaco: lack of insecticide spraying campaigns over the previous 12 years; reportedly high infestation levels with T. infestans, and requests of vector control interventions made by local district and healthcare authorities. Following an initial survey which provided evidence of large house infestation and active parasite transmission, all inhabited house compounds were sprayed with pyrethroid insecticides in November-December 2007 [40,43]. House infestation with T. infestans was monitored every 4–6 months from 2007 to 2010 (which revealed moderate pyrethroid resistance levels) and annually thereafter; all houses found to be reinfested were selectively re-sprayed with insecticides after each survey [40]. House infestation at the time of human diagnosis and treatment (September 2010-March 2011) was <1% and mainly occurred in peridomestic structures; none of the bugs collected were infected with T. cruzi as determined by microscopic analysis of feces at 400×. These conditions and an established vector surveillance system were taken as prerequisites to launch diagnostic and treatment activities.
The diagnosis-and-treatment program included five successive phases: preparatory, participatory planning, capacity strengthening of local health personnel, diagnostic surveys, and treatment and follow-up.
Treatment-related primary outcomes included treatment coverage and quality and therapeutic response. Treatment coverage was estimated as the percentage of seropositive patients up to 18 years of age at serodiagnosis that were treated with benznidazole relative to the number of seropositive patients in this age group who were eligible for treatment. Assessment of treatment quality was based on individual completion, medication adherence and ADR management. Adherence was evaluated at the periodic appointments through the percentage of benznidazole pills taken (i.e., number provided minus residual pills) relative to those provided for each specific time period, and then averaged over the treatment period. Patients with three or fewer pill counts and those who were withdrawn from or abandoned treatment were excluded from adherence estimates. Evaluation of ADR management included the percentage of patients who presented ≥1 ADR and were able to complete treatment; secondary outcomes included duration of the severest episode and proper notification of the event. Therapeutic response (i.e., treatment failure) was primarily evaluated through detection of T. cruzi DNA by kPCR and qPCR [3,13,47] in the subset of patients who completed the full treatment course and had ≥80% of medication adherence. Although conventional serodiagnosis was also performed, at least in the Gran Chaco region clearance of conventional anti-T. cruzi antibodies usually took several years even in young patients in the early chronic phase [12,16,17, 21]; therefore, we did not expect conventional serology would provide early evidence of seroconversion at 180 dpt. For additional comparisons, historical levels of treatment coverage with benznidazole or nifurtimox at the local hospital were estimated through a retrospective search of clinical histories of residents from the study area during the previous five years, and through householders’ reports.
We used Friedman’s two-way non-parametric analysis of variance to test for significant differences among repeated measurements of uremia, creatinine, AST, ALT and ALP performed before treatment and at 20 and 60 dpt, and Kendall’s K as an index of concordance. Fisher’s exact test or χ2 tests were used for investigating two by two contingency tables of independent data. Exact McNemar significance probabilities were calculated for paired data with small cell frequencies. The nominal level of statistical significance was set at a P value of 0.05. All tests were performed using Stata 12 [50].
The flow chart of the study population from census to follow-up is shown in Fig 1. The overall coverage of serodiagnosis was 70.3% (395 of 562) in children and adolescents aged up to 18 years old, with a peak in the age group 5–9 years old (Fig 2). The seroprevalence of T. cruzi infection in residents up to 18 years old was 24.1% (95 of 395). According to householders’ reports, the local health system had serologically examined for T. cruzi infection 16 (2.8%) local residents aged up to 18 years before the current intervention (Fig 2).
Ninety-five seropositive individuals were screened for eligibility (Fig 1). Reasons for lack of participation included refusal to initiate treatment, household emigration from the study area, impaired health conditions, and a previous treatment with benznidazole or nifurtimox. Treatment with benznidazole was initiated by all 78 eligible participants (Fig 1). Treatment coverage decreased with increasing age (Fig 2). Crude treatment coverage among all identified seropositive residents was 82% (78 of 95). The mean (±SD) age at treatment was 11.0±4.0 years old (range, 3–19), and 36 (46%) treated patients were females.
All seropositive children who initiated treatment were asymptomatic and displayed the indeterminate form of Chagas disease at baseline except one (of 71 patients examined by electrocardiography) with a congenital arrhythmia unrelated to Chagas. Hematocrit, hemoglobin, platelets, white cell counts and liver enzymes usually were within normal limits, except one case of anemia and 27 with eosinophilia (≥10%); these laboratory results did not prevent the initiation of treatment. Friedman’s test showed statistically significant changes over 0, 20 and 60 dpt in creatinine levels (P = 0.026, K = 0.502), AST (P = 0.012, K = 0.485), ALP (P < 0.001, K = 0.641), and marginally significant changes in uremia (P = 0.077, K = 0.420) and ALT (P = 0.082, K = 0.421). Although most laboratory tests remained within normal limits during treatment, three patients presented a ≥2× increase of ALP at 20 dpt coinciding with ADR episodes whereas other three showed elevated levels of ALT and ADRs. Eleven patients had a ≥2× increase of ALP at 60 dpt, but most of them had no ADR. No patient interrupted treatment due to laboratory abnormalities.
Treatment was completed by 66 (85%) of the 78 patients enrolled in the study (Fig 1). Completion rates decreased from 100% among children <5 years to 85% among young people aged 15–19 years (Fig 2). Individuals who did not complete treatment tended to be older (12.4±6.0 years, range 9–17) and had a balanced gender distribution (50%). Medication adherence across patients who completed the full treatment course averaged 97% (range, 80–100%) (Table 1). Among the 12 patients who did not complete treatment, four took medication for 25–28 days (range of adherence, 70–100%); other four took it for 4–17 days, and no data were provided by four patients. For comparison, the local health system had treated with benznidazole only two T. cruzi-seropositive children residing in the study area over the previous five years.
A total of 24 (32%; 95% CI: 21–43%) patients presented ≥1ADR (Table 1), and 20 (83%) reported it through the agreed mechanisms. Table 1 only includes the severest presentation for the six patients with 2 or 3 ADR episodes. The mean age of patients showing ≥1 ADR (12.0 years, 95% CI: 10.3–13.7) was not significantly different from that of patients showing no ADR (10.5 years, 95% CI: 9.4–11.6). On average, the first ADR appeared at 13.3 dpt initiation (95% CI: 10.8–15.8; range, 4–29 dpt).
Thirty-eight ADR episodes were recorded, including 22 mild, 9 moderate and 4 severe presentations, and 3 dermatological ADRs whose severity could not be established because of late reporting (Table 1). Twenty-one (88%) patients with ≥1 ADR displayed maculopapular exanthema, alone or combined with mild headaches or an increasingly severe arthralgia/myalgia. No Lyell or Stevens-Johnson syndromes and polyneuritis were recorded. Mild and moderate ADRs were managed in an out-patient basis; physicians evaluated the patients and eventually administered antihistamines, paracetamol and ibuprofen. Temporary dose reductions were indicated to four patients. Benznidazole was temporarily suspended for an average of 4 days in six patients showing mild or moderate exanthema and mild to severe arthralgia/myalgia or moderate headaches. Treatment withdrawal (followed by short-term hospitalization and administration of corticoesteroids and antihistamines) was prescribed to four patients displaying repeat ADR episodes that did not respond to symptomatic medication (age range, 6–17 years): three had a severe exanthema (one with fever) and one a moderate, prolonged exanthema and headaches. Duration of the severest ADR episode took on average 3.4 days (range, 1–7 days). Medication adherence varied very little among ADR types and levels.
Eighteen (75%) patients with ≥1ADR were able to complete treatment. Treatment completion rates significantly decreased with increasing ADR severity from 90–92% among patients with either no ADR or mild exanthema (including headaches and dizziness) to 0% among the three patients with a severe exanthema (Fisher’s exact test, P = 0.011), after pooling no or mild ADRs versus moderate or severe ADRs to avoid small cell frequencies (Table 1). One patient who failed to report an ADR (exanthema) abandoned treatment without medical indication and prompted other four family members under treatment to do so despite they had no ADR. Among patients having another household member under treatment, the relative odds of presenting at least one ADR was significantly and positively associated with having another household member with an ADR (OR = 2.57; 95% CI: 1.02–7.28; exact McNemar significance, P = 0.043).
The therapeutic response to treatment as determined by kPCR and qPCR is shown in Table 2. Before treatment, 40 (65%) patients were co-positive by both PCRs, 15 (24%) were co-negative, and discordant results occurred in 7 (11%) qPCR-positive and kPCR-negative patients; the performance of both PCRs differed significantly (exact McNemar significance, P = 0.016). During treatment, only one (2%) patient was kPCR-positive and qPCR-negative whereas the remainder was co-negative. Immediately after treatment at 60 dpt, 56 (97%) patients were co-negative, 1 (2%) was kPCR-negative and qPCR-positive, and 1 (2%) showed the reverse discordant pattern. This patient`s parasite burden decreased from 20 Pe/mL before treatment to 0.44 at 60 dpt and was negative thereafter. At 180 dpt, the only patient who was co-positive had shown T. cruzi DNA amplification at 20 dpt by kPCR (i.e., treatment failure) and refused the offer for new treatment with nifurtimox. The relative frequency of kPCR- or qPCR-positive results before treatment highly significantly declined by 180 dpt (exact McNemar significance, P < 0.0001). Median parasite load among kPCR- or qPCR-positive patients before treatment drastically fell from 1.4 Pe/mL to undetectable levels at 20 dpt and thereafter remained much lower than before treatment among the three patients positive by either PCR (Table 2). Of the 46 patients who were qPCR-positive before treatment, only two were positive at 60 (subsequently negative) or 180 dpt; the remainder was two (42) or three (38) times negative over 20–180 dpt. Of 16 patients initially qPCR-negative, none was positive over 20–180 dpt.
The treated patients who participated in the serological follow-up at 60 or 180 dpt showed virtually no decay in paired optical densities by the Chagatest ELISA between 0 and 60 dpt (mean percent absolute reduction, 5.5%; 95% confidence interval, CI, 1.3–9.8) or between 0 and 180 dpt (mean, 0.5%; CI, -1.6–1.7%). Only one patient had a drop in optical density levels below the Chagatest cutoff value. Using recombinant ELISA, mean percent absolute reductions between 0 and 60 dpt (6.6%; CI, 2.9–10.4%) and between 0 and 180 dpt (mean, 7.1%; CI, 2.3–11.8%) were slight at most, with only one patient having a >60% drop in optical densities.
Our study demonstrates that diagnosis and treatment of T. cruzi infection in remote, impoverished rural areas can be effectively addressed through strengthened primary healthcare attention and broad social participation combined with adequate external support. This strategy took advantage of locally available resources; secured high levels of diagnostic and treatment coverage and medication adherence, effectively managed ADRs at village level under carefully administered protocols, and provided early evidence of positive therapeutic responses except in one case. Although the efficacy of benznidazole has been firmly established in hospitals or specialty clinics [15,16], especially among young patients, the issues of treatment access, adherence and effectiveness in remote, hyperendemic rural settings have received little attention.
Broad social participation implied a multi-stakeholder agreement that included the affected communities and other sectors. The community workshops were essential to raise awareness of Chagas disease characteristics and consequences, diagnostic and treatment opportunities, and provided an appropriate context to reach an agreement on intervention details adapted to local circumstances. A key output was the two-sided commitment for implementing interventions and building of mutual trust. Compliance with dates for the expected return of serological results (unlike in other health interventions reported by householders) facilitated further community involvement with treatment and follow-up activities.
The local health system faced critical constraints (e.g., medical personnel and vehicle) at the time of the interventions. These facts, combined with considerable distances between rural villages and the local hospital, led to a centralized health service delivery model to which the rural population had poor access. The initial workshop outputs identified that most activities had to be conducted at the residents’ villages. Strengthening the capacity of rural healthcare agents and medical personnel was essential: the former undertook monitoring of medication adherence and ADRs, whereas the in situ participation of physicians was restricted to treatment indication, clinical exams and improved ADR management. In remote areas of Africa, transference of treatment initiation and monitoring of HIV patients to local nurses improved the access to and quality of health care and cost-effectiveness of integrated interventions [51,52]. Benznidazole was available at the time of our program and timely provision was carefully planned, but these usually are major issues elsewhere. As a direct repercussion of the current program, spontaneous treatment demand at the local hospital increased substantially over subsequent years and included adult patients.
The degree of diagnostic coverage attained among rural residents up to 18 years old was substantial (70.3%) and paralleled or exceeded levels recorded in rural communities of the Argentine Chaco [6,39] and by the local health system. The enhanced diagnostic coverage was partly related to the prospects of receiving treatment and eventual welfare benefits derived from being seropositive for T. cruzi.
Treatment completion rates were likely enhanced through the contribution of dedicated staff supervising the onset of benznidazole administration and ADR follow-up; no incentives for treatment initiation, completion and follow-up were given. Medication adherence was very high and similar to the levels recorded in urban or rural populations through other approaches [19,21,22]. The healthcare agent- and parent-based monitoring system was key to early identify patients showing ADRs and breaches in adherence. Likewise other benznidazole trials in children [19,21], most ADRs occurred during the first two weeks after onset of treatment; close patient monitoring during this period is therefore indicated. No significant age-related increase of the occurrence of an ADR was detected, unlike in studies covering a wider age range [20,26]. The significant household aggregation of ADRs suggests putative genetic or environmental factors acting on a familial level that may possibly modify the bioavailability of or response to benznidazole. Elevated levels of ALT or ALP during or immediately after benznidazole treatment occurred in several patients, but they were not as important as to interrupt treatment (rarely exceeded baseline levels by 3×) nor were laboratory abnormalities associated with the occurrence of an ADR.
The most frequent benznidazole-related ADRs were dermatological (usually of mild or moderate severity) as in several other studies and locations, although their frequency tends be quite variable [12,19,21,22]. Most mild or moderate ADRs were successfully managed with symptomatic medication, dose reduction and temporary suspension of benznidazole intake, whereas the four cases with severe ADRs required treatment withdrawal, short-term hospitalization and symptomatic medication. Treatment completion was significantly and inversely related to ADR severity, again supporting the need of close monitoring and adequate ADR management as a means of increasing completion rates and preventing treatment abandonment in other family members.
qPCR was significantly more sensitive than kPCR to detect T. cruzi DNA before treatment and evidenced the rapid fall of circulating parasites as early as three weeks after onset of treatment [20,48]. Evidence from several trials supports that qPCR is an early marker of treatment failure [13,31,46–48]. The proportion of T. cruzi-seropositive patients initially qPCR-positive (65%) was very close to the range recorded in several other studies targeting the same age group and early chronic infections [18,53,54]. Moreover, all children [22] and the great majority of adult patients [13] treated with benznidazole for 60 days remained qPCR-negative over 1–1.5 years of follow-up after effective treatment, hence suggesting that under such conditions recurrence of parasitemia appears to be rare but other patterns have been reported [e.g., 18]. In our study, the only patient with persistently positive kPCR or qPCR tests (before and at 20 and 180 dpt) was a 12-year-old child with rather limited adherence (80%) who had no travel history outside the study area and resided in a non-infested house (i.e., a treatment failure). However, a sizable fraction of adult individuals under a partial treatment course with benznidazole have shown positive therapeutic responses [55]. A second patient with very much reduced parasite burden from 0 to 60 dpt (but still qPCR-positive) was subsequently qPCR-negative, as recorded by others [46]. Most important, all patients but two were subsequently qPCR-negative on two or three occasions over 20–180 dpt. Although we cannot exclude whether longer follow-up times would allow some of the molecular tests to yield a positive result, the individual time patterns of qPCR including two or three negative results in a row are more compatible with a process of parasite clearance in young patients in the early chronic phase of infection [cf 14.17].
Intensified vector control and systematic surveillance was deemed a prerequisite for launching the diagnosis-and-treatment program because fast house reinfestation after insecticide spraying frequently causes new human infections in rural areas of the Gran Chaco [5,8,10] and elsewhere [18,56]. In the face of moderate pyrethroid resistance, suppressing house infestations demanded recurrent insecticide applications [40] and delayed treatment activities. However, this prolonged process laid the foundations for subsequent participatory activities; averted the chance of new vector-mediated human infections, and determined that child infections most likely had been acquired at least 3–4 years before treatment (i.e., early chronic infections).
Our study had several limitations. Generalizability of the current strategy to other areas depends on the existence of a primary healthcare service. Medication adherence was measured through pill counts in the absence of a practical assay for establishing serum benznidazole levels at the study setting. In the absence of a well-established biomarker of early cure [26,47], the time-limited follow-up of patients up to 180 dpt precluded us from using seroconversion to a negative status as a primary endpoint. Seroconversion usually takes several years even in young residents from the Gran Chaco in the early chronic phase [12,16,17,21], depending on the time span between primary infection and treatment and other ill-defined determinants, which explains the nearly stable pattern of ELISA optical densities before and after treatment. We note, however, that results elsewhere in central Brazil [14,15], Guatemala and Honduras [21] and Colombia [18] showed quite diverse patterns of conventional antibody clearance rates in response to benznidazole treatment in young patients <15 years old. Losses to follow-up via PCRs at 180 dpt included 17 (26%) of 66 patients who completed treatment and may bias estimates of treatment effect size. Cohort attrition over the follow-up is typical in trials conducted in rural locations, more so in remote areas with migrant populations. Whether T. cruzi-infected children treated with benznidazole will show improved long-term clinical outcomes is a crucial question that merits further research. Conclusions from this study may not be extrapolated to adult patients who have more frequent and severe ADRs with a different time pattern [22].
Our study links community participation to a health outcome improvement [57,58]. Community participation in remote rural settings is essential for treatment programs [32]. Increasing access to high-quality serodiagnosis and treatment of marginalized rural populations, combined with effective vector control and surveillance in the affected regions, is ethically imperative.
|
10.1371/journal.pbio.2000931 | A Ca2+ channel differentially regulates Clathrin-mediated and activity-dependent bulk endocytosis | Clathrin-mediated endocytosis (CME) and activity-dependent bulk endocytosis (ADBE) are two predominant forms of synaptic vesicle (SV) endocytosis, elicited by moderate and strong stimuli, respectively. They are tightly coupled with exocytosis for sustained neurotransmission. However, the underlying mechanisms are ill defined. We previously reported that the Flower (Fwe) Ca2+ channel present in SVs is incorporated into the periactive zone upon SV fusion, where it triggers CME, thus coupling exocytosis to CME. Here, we show that Fwe also promotes ADBE. Intriguingly, the effects of Fwe on CME and ADBE depend on the strength of the stimulus. Upon mild stimulation, Fwe controls CME independently of Ca2+ channeling. However, upon strong stimulation, Fwe triggers a Ca2+ influx that initiates ADBE. Moreover, knockout of rodent fwe in cultured rat hippocampal neurons impairs but does not completely abolish CME, similar to the loss of Drosophila fwe at the neuromuscular junction, suggesting that Fwe plays a regulatory role in regulating CME across species. In addition, the function of Fwe in ADBE is conserved at mammalian central synapses. Hence, Fwe exerts different effects in response to different stimulus strengths to control two major modes of endocytosis.
| The arrival of an action potential at the nerve end induces synaptic vesicle (SV) exocytosis to allow the release of chemical neurotransmitters and the rapid transmission of signals. SV endocytosis is in turn elicited in order to rapidly replenish the vesicle pool in neurons. Therefore, tight coupling between exocytosis and endocytosis within these cells maintains constant synaptic transmission. Exocytosis and intracellular Ca2+ elevation are known to be key prerequisites for the two main modes of SV endocytosis, Clathrin-mediated endocytosis (CME) and activity-dependent bulk endocytosis (ADBE), which are primarily triggered by moderate and strong nerve stimuli, respectively. However, how these two events cooperate to trigger endocytosis upon exocytosis remains unclear. In this study, we show that Flower (Fwe), an SV-associated Ca2+ channel, plays a significant role in these processes in Drosophila. Upon mild stimulation, we find that Fwe is transferred from the SV to endocytic zones to promote CME independently of Ca2+ channeling. In contrast, upon intense stimulation Fwe triggers Ca2+ influxes that elicit ADBE. In addition, we find that Fwe promotes CME and ADBE also in mammalian central synapses, revealing a conserved role for Fwe in endocytosis. We conclude that the Fwe channel exerts two different functions in response to two different stimuli to govern distinct modes of synaptic vesicle retrieval, thereby coupling exocytosis to endocytosis.
| In the presynaptic terminal, continuous release of synaptic vesicles (SVs) results in vesicle pool depletion, plasma membrane expansion, and SV protein overloading at the release site [1]. Endocytosis is therefore tightly coupled to exocytosis [2]. Among the different modes of SV endocytosis, Clathrin-mediated endocytosis (CME) and activity-dependent bulk endocytosis (ADBE) are well characterized [3,4]. In mild stimulation paradigms, CME is the prevalent mode of retrieving exocytic SVs in the form of a single SV [5,6]. In intense stimulation paradigms, however, ADBE promotes the uptake of large pieces of fused membranes in bulk endosomes or cisternae [7,8]. Small SVs are then formed from these membranous structures.
SV exocytosis is a prerequisite for CME and ADBE initiation [2], indicating that specific SV cargoes, an exocytic protein complex, or both are needed to trigger both modes of endocytosis. Indeed, Synaptotagmin, the SV Ca2+ sensor for exocytosis [9], and components of the soluble NSF attachment protein receptor (SNARE) complex play crucial roles in CME [10–14]. Moreover, recent studies have identified vesicle-associated membrane protein 4 (VAMP4), a vesicle-associated SNARE (v-SNARE) protein, as a selective SV cargo for ADBE. Interestingly, VAMP4 is responsible for the formation of ADBE as well [15]. Thus, SV proteins encode components that retrieve SV membrane in newly formed vesicles as well as coordinate the nature of the formation of SV endocytosis.
An increased local Ca2+ concentration in the presynaptic terminal is necessary not only for exocytosis but also for endocytosis [16–20]. Moreover, the Calmodulin/Calcineurin complex was proposed to function as a Ca2+ sensor for CME and ADBE [19,21,22]. Therefore, Ca2+/Calcineurin likely acts as a universal signal that elicits most forms of the SV retrieval [2]. At the rat Calyx of Held, in addition to a role in triggering exocytosis, a high, transient Ca2+ influx via a voltage-gated Ca2+ channel (VGCC) triggers CME [18,23]. However, the Ca2+ channel for triggering ADBE is unknown.
Our previous genetic screen in flies identified Flower (Fwe), an evolutionarily conserved transmembrane protein [24]. Fwe forms a Ca2+-permeable channel when expressed in heterologous cells or when incorporated into proteoliposomes. This protein localizes to SVs, and, upon SV release, Fwe is transferred to the periactive zone, where it triggers CME, thereby coupling exocytosis to CME. In the present study, we show that Fwe initiates ADBE as well. Intriguingly, the effects of Fwe on CME and ADBE depend on the strength of the stimulus. We found that the function of Fwe for regulating CME does not involve Ca2+ channeling. Instead, upon intense stimulation, Fwe triggers a Ca2+ influx that elicits ADBE. Lastly, when we removed ratFwe in cultured rat hippocampal neurons through clustered regularly interspaced short palindromic repeats (CRISPR)/ CRISPR-associated protein 9 (Cas9) technology, CME is impaired but not completely blocked, similar to the defect caused by the Drosophila fwe mutation. Furthermore, our data reveal that RatFwe is also involved in the induction of ADBE at mammalian central synapses. In summary, the Fwe channel exerts two different functions in response to two different stimuli that govern distinct modes of SV retrieval, thereby coupling exocytosis to endocytosis.
We previously reported that the Fwe Ca2+ channel promotes CME in the synaptic boutons of Drosophila neuromuscular junctions (NMJs) [24]. To investigate whether the Ca2+ influx via Fwe plays a direct role in CME, we utilized the FweE79Q mutant whose channel activity is severely impaired [24] and assessed the ability to rescue CME defects associated with fwe mutants. We expressed UAS-flag-fwe-HA and UAS-flag-fweE79Q-HA with nSyb-GAL4, a pan-neuron GAL4 driver, in a strong loss of fwe background (fweDB25/fweDB56) (S1A and S1B Fig) [24]. α-HA antibody staining was used to examine the distribution and expression of Fwe proteins in boutons. α-horse radish peroxidase (HRP) antibody staining labels the insect neuronal membranes, thereby outlining presynaptic compartments [25]. Both the wild-type Fwe and FweE79Q proteins are evenly distributed in boutons (S1A and S1B Fig). We then introduced a genomic HA-tagged fwe transgene to estimate the relative expression levels of UAS transgenes versus endogenous Fwe protein (S1C–S1F Fig). The proteins are expressed at ~50% of the endogenous Fwe protein level in type Ib NMJ boutons (S1L–S1N Fig), whereas their expressions in type Is NMJ boutons correspond to ~80% of the endogenous Fwe protein.
To estimate the efficacy of CME, we performed the FM1-43 dye uptake assay with moderate stimuli, i.e., 1-min 90 mM K+/0.5 mM Ca2+ and 10-min 60 mM K+/1 mM Ca2+ stimulations. The experimental paradigm is shown in S2A Fig. Both stimulation paradigms significantly elicit dye uptake in wild-type control larvae when compared to a resting paradigm (10-min incubation in 5 mM K+/0 mM Ca2+ solution) (S2B–S2E Fig). We then performed a transmission electron microscopy (TEM) assay to assess the formation of bulk cisternae, a hallmark of ADBE [2,8]. No bulk cisternae were induced under these conditions (S2F–S2I Fig), showing that the strength of these stimuli is mild, which predominantly promotes CME. Upon 1-min 90 mM K+/0.5 mM Ca2+ stimulation, loss of fwe impairs FM1-43 dye uptake (Fig 1A, 1B and 1H). It is possible that either a defect in SV endocytosis or exocytosis would reduce FM1-43 dye uptake in this case. To test the role of Fwe in SV exocytosis, we performed the FM1-43 dye loading/unloading assay. The experimental paradigm is indicated in S3A Fig. Both the control and fwe mutant boutons were subjected to 5-min 90 mM K+/0.5 mM Ca2+ stimulation, which labels the SV pool with FM1-43 dye (S3B and S3D Fig). Subsequently, 1-min stimulation with the same solution releases the SVs and unloads the dye from SVs (S3C and S3E Fig). The strength of SV exocytosis is correlated with the FM1-43 dye unloading efficiency ([(Fload—Funload) / Fload]). While the dye loading is significantly reduced when fwe is lost (S3B, S3D and S3F Fig), the dye unloading efficiencies of controls and fwe mutants are comparable (S3G Fig), indicating that Fwe plays a marginal or no role in SV exocytosis. Hence, the FM1-43 dye uptake deficit associated with fwe mutants mainly results from a defect in CME.
Reduced FM1-43 dye uptake in fwe mutants is completely rescued by the reintroduction of 50% Fwe protein (Fig 1C and 1H). However, a similar rescue was also observed when 50% FweE79Q is present (Fig 1D and 1H). Although the channel function of FweE79Q is mostly absent when analyzed in fly salivary glands [24], the remaining channel activity in this mutant might be sufficient to promote CME if enough proteins are present in the boutons. We therefore determined the minimal Fwe level required for CME to verify the channel function of Fwe. After surveying numerous GAL4 lines, elav-GAL4 and nSyb(w)-GAL4 were found to drive the expression of the transgenes at approximately 0% and 4% of the endogenous Fwe level, respectively (S1G–S1J, S1M and S1N Fig). As shown in Fig 1E, 1F and 1H, boutons expressing 10% or 4% Fwe can take up FM1-43 dye efficiently. We noted that the dye uptake with 4% Fwe expression is marginally reduced when compared to other Fwe-expressing larvae, indicating that this level of Fwe expression is near the minimal level for efficient CME.
Next, we assessed CME in 4% FweE79Q-expressing boutons (S1J, S1K, S1M and S1N Fig). Intriguingly, the efficiency of the dye uptake is not different between 4% Fwe and 4% Fwe E79Q-expressing boutons (Fig 1F–1H), suggesting that CME can occur despite the lack of a significant Ca2+ influx via Fwe. We previously showed that loss of fwe results in a reduced SV number and enlarged SVs [24]. To examine the changes in SV ultrastructure, we performed TEM. The wild-type control bouton under the resting condition contains numerous SVs (Fig 1I), whereas the number of SVs is decreased upon loss of fwe (Fig 1J and 1Q). This low SV number worsens following 10-min 60 mM K+/1 mM Ca2+ stimulation (Fig 1M, 1N and 1Q). Either 4% Fwe or 4% FweE79Q expression rescues this SV loss (Fig 1K, 1L and 1O–1Q). In addition, enlarged SV sizes associated with fwe mutants are normalized under both expression conditions (Fig 1R). Hence, these data indicate that Fwe triggers CME independent of Ca2+ channeling.
Upon mild stimulation, CME retrieves the membrane that corresponds in size to an SV. In contrast, in response to intense stimulation, ADBE takes up large quantities of fused SVs from the plasma membrane to form bulk cisternae. It has been shown that both SV exocytosis and intracellular Ca2+ elevation are essential for ADBE to proceed around the periactive zones [7,18,26]. This raises the possibility that Fwe may play a role in ADBE. High K+ and Ca2+-containing solutions have been widely used to elicit ADBE at several different synapses, including fly NMJ boutons [8,27–30]. To assess the role of Fwe in ADBE, we applied 10-min 90 mM K+/2 mM Ca2+ stimulation to induce ADBE and examine the formation of bulk cisternae using a TEM assay. The TEM image of control boutons reveals numerous cisternae (>80 nm in diameter, red arrows) elicited by this stimulation paradigm (Fig 2A, 2B and 2G). These processes, however, are dramatically suppressed by loss of fwe (Fig 2C, 2D and 2G). In unstimulated conditions, bulk cisternae are also less abundant in fwe mutants than in controls (Fig 2G). This ADBE defect indeed results from the fwe mutation, as 50% Fwe expression rescues this ADBE phenotype (Fig 2E–2G). Interestingly, the average size of the few bulk cisternae observed in fwe mutants is comparable to that observed in control boutons (Fig 2H), suggesting that Fwe acts at the initiation step of ADBE rather than during a late membrane invagination process. Furthermore, following high K+ stimulation, the accumulation of early endocytic intermediates was observed around the periactive zone in fwe mutant boutons (Fig 2D–2D1 and 2I, yellow arrows) when compared to wild-type controls and 50% Fwe-rescued larvae (Fig 2B–2B1, 2F–2F1 and 2I). Since optimal SV exocytosis is shown as a prerequisite for triggering ADBE [7,26], we therefore estimated the total SV area per bouton area under the resting condition. No difference between controls and fwe mutants was found (Fig 2J), showing that the ADBE defect associated with fwe mutants is not due to insufficient supply of exocytic SV membrane upon stimulation. Moreover, following 90 mM K+/2 mM Ca2+ stimulation, the strength of SV exocytosis determined by the FM1-43 dye loading/unloading assay is comparable between controls and fwe mutants (S3H–S3N Fig). Collectively, these results reveal that Fwe is responsible for initiating ADBE during intense activity stimulation.
Acute inactivation of the components involved in CME, such as Clathrin, AP180, and Dynamin, elicits bulk membrane invaginations [31–34], suggesting that CME suppresses ADBE or that ADBE is the result of membrane expansions. To assess the role of Fwe in this process, we treated larvae with 200 μM chlorpromazine to inhibit Clathrin coat assembly [31], followed by 10-min 90 mM K+/ 2 mM Ca2+ stimulation in the presence of FM1-43 dye. As shown in Fig 2K–2N, large membranous invaginations enriched with FM1-43 dye were detected in the controls. In contrast, these structures decrease upon loss of fwe. In summary, these results reinforce the functional importance of Fwe in ADBE.
To assess whether Fwe mediates an intracellular Ca2+ increase to initiate ADBE upon intense activity stimulation, we expressed the lexAop2 transgene of the fast-decay version of the genetically encoded Ca2+ indicator GCaMP6, GCaMP6f [35], in the presynaptic terminals with vglut-lexA, a glutamatergic neuron driver. We have shown previously that fwe mutant boutons display low resting Ca2+ levels [24]. Similarly, decreased GCaMP6f fluorescence was observed upon loss of fwe (Fig 3A, 3B and 3E, white arrows). This indicates a reduction in the resting Ca2+ levels, as the expression level of GCaMP6f in boutons is higher in fwe mutants than in controls (S4A, S4B and S4E Fig). Next, we stimulated boutons with 90 mM K+/2 mM Ca2+ solution for 10 min, which elicits ADBE (Fig 2A and 2B), and measured GCaMP6f fluorescence in the 6th and 10th min. In controls, the intracellular Ca2+ concentrations in response to stimuli are substantially increased (Fig 3A–3A3 and 3F), whereas loss of fwe significantly impedes these Ca2+ elevations (Fig 3B–3B3 and 3F). Hence, Fwe sustains presynaptic Ca2+ levels upon strong stimulation.
To assess the role of Fwe-derived Ca2+ influx in regulating presynaptic Ca2+ level, we traced GCaMP6f fluorescence in 4% Fwe- and 4% FweE79Q-expressing boutons under the resting and high K+ stimulation conditions. In both conditions, the levels of GCaMP6f are expressed similarly to that in controls (S4C–S4E Fig). We found that 4% Fwe but not 4% FweE79Q expression restores normal resting Ca2+ levels in fwe mutants (Fig 3C–3E). Furthermore, a defect in high K+-induced Ca2+ elevation associated with fwe mutants is partially reversed by 4% Fwe expression (Fig 3C–3C3 and 3F). In contrast, 4% FweE79Q expression fails to rescue this Ca2+ defect (Fig 3D–3D3 and 3F). However, when we examined the presynaptic Ca2+ changes following 1-min 90 mM K+/0.5 mM Ca2+ stimulation, which prevalently elicits CME (S2 Fig), the Ca2+ increases among all genotypes are quite similar (Fig 3G). Hence, these results indicate that Fwe triggers a Ca2+ influx specifically in response to strong stimuli.
To verify this stimulus-dependent Ca2+ channeling of Fwe, we measured the Ca2+ increase evoked by electric stimuli in wild-type and fwe mutant boutons. To this end, we expressed UAS-GCaMP6f with nSyb-GAL4. The levels of GCaMP6f in control and fwe mutant boutons are comparable (S4F, S4G and S4K Fig). The stimulation paradigms are shown in the top panel of Fig 3H. Upon 10–40 Hz train stimuli, intracellular Ca2+ increase in controls correlates with the stimulus strength (S5A–S5A4 Fig and Fig 3H, white circle). In fwe mutants (S5B–S5B4 Fig and Fig 3H, red circle), the Ca2+ increase at 10 Hz is slightly higher than that in controls, and the Ca2+ increase at 20 Hz is similar to that in controls. In contrast, at 40 Hz, loss of fwe significantly impairs the evoked Ca2+ increase (Fig 3H and 3I). Moreover, 50% Fwe expression normalizes this deficit (S4H, S4K and S5C–S5C4 Figs and Fig 3H and 3I, black circle), whereas a partial restoration by 50% FweE79Q expression was observed (S4I, S4K and S5D–S5D4 Figs and Fig 3H and 3I, orange circle). Furthermore, we observed similar effects on rescuing low resting Ca2+ levels associated with fwe mutants (S5F Fig). These results support the finding that Fwe does not mediate a Ca2+ influx under a moderate stimulation condition that predominantly induces CME. Instead, it conducts a Ca2+ influx when neurons undergo intense stimulation.
To rule out that the defect in evoked Ca2+ increase may be attributed to slow CME associated with fwe mutants, we applied the same stimulation protocol to dap160 mutants, which exhibit a similar CME defect [36–38]. As shown in S5E–S5E4 Fig and Fig 3H and 3I (green circle), in dap160 mutant boutons, the Ca2+ concentrations at 40 Hz are elevated to wild-type levels, although the Ca2+ increase at 10 and 20 Hz is higher than that observed in controls and fwe mutants. In addition, the fluorescence level but not the expression level of GCaMP6f under the resting condition is reduced upon loss of dap160 (S4J, S4K, S5E and S5F Figs), suggesting that dap160 mutants display low resting Ca2+ levels. Therefore, our results argue that a defective CME does not account for presynaptic Ca2+ dysregulation in fwe mutants. In addition, we found no evidence for the changes in the distribution and expression of Cacophony, the major VGCC located at the active zone (S6 Fig).
The above-mentioned results prompted investigations into the role of Fwe-driven Ca2+ influx in ADBE. We showed previously that 4% Fwe is sufficient for CME. We therefore addressed if this level of Fwe is sufficient to promote ADBE. When boutons are expressed with 10% or 4% Fwe, ADBE elicited by 10-min 90 mM K+/2 mM Ca2+ stimulation efficiently produces bulk cisternae (Fig 4A, 4B, 4D, 4E and 4K). Thus, a partial Ca2+ influx by 4% Fwe (Fig 3F) is sufficient for initiating ADBE. Consistently, 50% FweE79Q, which induces a fractional Ca2+ influx (Fig 3H and 3I), robustly triggers ADBE after high K+ stimulation (S7 Fig). However, high K+-induced bulk cisternae are significantly reduced in 4% FweE79Q-rescued larvae (Fig 4G, 4H and 4K). Notably, the number of high K+-induced bulk cisternae between 4% FweE79Q-rescued and fwe mutant larvae is comparable (fwe mutant larvae, 1.06 ± 0.4, n = 22, versus 4% FweE79Q-rescued larvae, 1.66 ± 0.42, n = 26, [Student’s t test, p = 0.31]). Similar to loss of fwe, there is an increase in the level of endocytic intermediates formed around the periactive zone in 4% FweE79Q-rescued boutons after high K+ stimulation when compared to 4% Fwe-rescued boutons (Fig 4L), thus supporting an important role of Ca2+ influx via Fwe in ADBE. At rest, the total SV membrane area per bouton area is also comparable between 4% Fwe- and 4% FweE79Q-rescued boutons (Fig 4M). Therefore, both expression conditions yield equal SV membranes available for SV exocytosis. These data suggest that Fwe triggers ADBE mainly through fluxing Ca2+. Furthermore, after chlorpromazine treatment, bulk membrane invaginations are less abundant in 4% FweE79Q-rescued boutons when compared to 4% Fwe-rescued boutons, also documenting a role of Fwe-derived Ca2+ influx in chlorpromazine-induced bulk membrane invagination (Fig 4N–4P).
If a suboptimal Ca2+ level in the presynaptic terminals results in impaired ADBE phenotype in 4% FweE79Q-rescued larvae, then we expected that increasing overall intracellular Ca2+ concentrations either via the remaining channel activity of FweE79Q or the other Ca2+ channels during stimulation might compensate for this low intracellular Ca2+ and rescue the defective ADBE. We therefore raised the Ca2+ concentration from 2 mM to 5 mM in our 90 mM K+ stimulation solution. When we applied a 10-min 90 mM K+/5 mM Ca2+ stimulation to 4% FweE79Q-rescued boutons, the bulk cisternae number is significantly rescued (Fig 4I and 4K). Similarly, 5mM Ca2+ also rescues the ADBE deficit associated with fwe mutants (Fig 4J and 4K). In contrast, this treatment does not increase the number of bulk cisternae further in 4% or 10% Fwe-rescued larvae when compared to the 10-min 90 mM K+/2 mM Ca2+ stimulation condition (Fig 4C, 4F and 4K). Hence, this rescue effect might be due to increased Ca2+ levels rather than enhanced ADBE in the presynaptic compartments. These data further support the role of Ca2+ influx via Fwe in triggering ADBE during intense activity stimulation.
If Ca2+ influx via Fwe triggers ADBE, one would anticipate that reducing channel activity will abolish ADBE. La3+ is a potent blocker of some Ca2+-permeable channels [39,40]. It may therefore inhibit the Fwe channel activity. We previously showed that heterologous expression of Fwe results in Ca2+ uptake by Drosophila salivary gland cells [24]. To determine the effect of La3+ on the Ca2+ conductance of Fwe, we applied 100 μM La3+ to the glands that carry a GAL4 driver only or overexpress Fwe and performed Ca2+ imaging. The cells were loaded with Fluo-4 AM Ca2+ indicator and bathed in 100 μM extracellular Ca2+ solution. Fwe-overexpressing cells display a slow but significant Ca2+ uptake over a 1-h period when compared to controls (Fig 5A, 5B and 5E), consistent with our previous observations [24]. However, application of 100 μM La3+ solution nearly abolishes the Ca2+ influx mediated by Fwe (Fig 5D–5E), although a mild suppression was observed in control cells as well (Fig 5C and 5E). These results indicate that, similar to other Ca2+-permeable channels, the channel pore region of Fwe has a high affinity for La3+ and is blocked by La3+.
Next, we assessed the impact of 100 μM La3+ on ADBE. Since La3+ impedes Ca2+ permeability of VGCCs [39,40], and VGCC-triggered exocytosis is essential for ADBE initiation [7,26], we first tested whether treatment with100 μM La3+ solution affects SV exocytosis. At 0.2 Hz in 1 mM Ca2+, 100 μM La3+ reduces the amplitude of excitatory junction potentials (EJPs) by ~60%–80% when compared to untreated ones, indicating that La3+ blocks a Ca2+ influx mediated by VGCCs. In contrast, 10 μM La3+ does not significantly influence 0.2 Hz-elicited EJPs in controls (Fig 5F, white column). As this low La3+ concentration might not block the Fwe-derived Ca2+ influx effectively, we used 10% Fwe-rescued larvae, which are more sensitive to 10 μM La3+ (Fig 5B). Under this expression condition, the application of 10 μM La3+ does not affect the EJP responses at 0.2 Hz (Fig 5F, green column) but largely reduces ADBE upon high K+ stimulation (Fig 5G and 5I). Notably, the number of bulk cisternae induced under La3+ treatment is almost identical to that observed in fwe mutants after high K+-stimulation (Fig 5I), suggesting that La3+ suppresses ADBE by selectively inhibiting the channel activity of Fwe. In support of this, in fwe mutants, 10 μM La3+ does not alter the 0.2 Hz-evoked EJP amplitude (Fig 5F, red column) or the level of high K+-induced bulk cisternae (Fig 5H and 5I). Overall, these data indicate a role of Fwe in Ca2+-mediated ADBE. In summary, Fwe governs two major modes of SV endocytosis to permit consecutive rounds of exocytosis of neurotransmitters following distinct activity stimuli.
Fwe homologs are found in most eukaryotes [24], but their role in SV endocytosis in vertebrates has not been established. The mouse Fwe (mFwe) gene can generate at least six alternative mRNA splicing isoforms [41], producing five different mFwe isoforms (S8 Fig). The mFwe isoform 2 (mFwe2) is the most similar to Drosophila Fwe. Moreover, the mFwe2 and rat Fwe isoform 2 (ratFwe2) share ~99% amino acid identity (170/172). In adult rat brain, ratFwe2 mRNA is widely expressed (Fig 6A). In the lysates of mouse neuroblastoma Neuro 2a (n2a) cells, our antisera against the C-termini of both mFwe2 and ratFwe2 (α-m/ratFwe2) recognize a ~18 kDa protein band, corresponding to the predicted molecular weight of mFwe2 (Fig 6B). This signal is significantly decreased when mFwe2 is knocked down by mFwe-microRNAi (miRNAi) (Fig 6B), showing antibody specificity. mFwe2 is expressed in postnatal as well as adult mouse brains (Fig 6B). Similarly, ratFwe2 was detected in rat brain and cultured rat hippocampal neurons (Fig 6C).
To determine if ratFwe2 is enriched in SVs, we purified SVs from adult rat brain using a series of centrifugations [42]. As shown in Fig 6D, ratFwe2 was specifically detected in the SV (Lysate pellet 2 [LP2]) fraction, marked by the presence of Synaptophysin (Syp), an abundant SV protein. ratFwe2 is also present in the SV fractions of adult rat brain separated with sucrose gradients (Fig 6E). To assess the subcellular localization of ratFwe2, we performed immunostaining in cultured neurons. Although staining with our antisera can visualize the expression of ratFwe2 in the cell bodies (S9B Fig), we failed to obtain specific staining in the presynaptic terminals. We therefore expressed HA-tagged mFwe2 in ratFwe knockout neurons (see below for details) and determined the SV localization of mFwe2-HA using α-HA staining. As shown in Fig 6F and 6G, mFwe2-HA protein is enriched in the presynaptic terminals and largely colocalized with Syp. The biochemical data combined with the in vivo localization data provide compelling evidence that ratFwe2 is associated with SV proteins, similar to Drosophila Fwe.
To determine whether rodent Fwe2 functions equivalently to Drosophila Fwe, we expressed UAS-flag-mFwe2-HA transgene in fwe mutants using nSyb(w)-GAL4. Overexpressed mFwe2 is localized to SVs in the boutons (Fig 6H–6H2) and rescues the FM1-43 dye uptake defect (Fig 6I–6K), as well as the reduced number of SVs (Fig 6L) in fwe mutants, showing that mFwe2 can promote CME in flies. Furthermore, the expression of mFwe2 corrects the ADBE deficit caused by loss of fwe (Fig 6M and 6N). Hence, mFwe2 promotes ADBE as well. We also observed that the early lethality of fwe mutant larvae is rescued by mFwe2 expression. Our results therefore suggest a conserved role of Fwe in SV endocytosis in mammals.
To verify the role of ratFwe2 in SV endocytosis, we knocked out ratFwe in cultured rat hippocampal neurons using CRISPR/Cas9 technology [43]. We designed a specific guide RNA (gRNA; m/ratFwe-gRNA) that targets the first intron/second exon junction of both mFwe and ratFwe genes. To estimate the knockout efficiency, we transfected the gRNA construct into mouse neuroblastoma n2a cells and established a mFwe knockout n2a cell line. While mFwe2 was detected in normal n2a cells, it is lost in mFwe knockout n2a cells (S9A Fig). At 14 days in vitro (DIV), ratFwe2 is present in the Golgi apparatus of the cultured rat hippocampal neurons (S9B–S9B3 Fig), consistent with the fact that ratFwe2 is an SV protein sorted from the Golgi. In neurons expressing Cas9 and m/ratFwe-gRNA, the expression of ratFwe2 in the Golgi apparatus is significantly diminished (S9C and S9D Fig). Thus, this gRNA can also efficiently remove ratFwe2 in cultured neurons when Cas9 is present.
To assess the efficacy of CME, we elicited exocytosis of Synaptophysin-pHluorin (SypHy) [5] by delivering 200 action potentials at 20 Hz and monitored its retrieval via SV endocytosis. It has been documented that this mild stimulation paradigm prevalently induces CME [12,15,44]. In control neurons (Fig 7A, black line) bathed at room temperature, repeated exocytosis in response to 20-Hz stimuli increases SypHy fluorescence, followed by a gradual fluorescence decay caused by the reacidification of SVs formed via CME. However, in ratFwe knockout neurons (Fig 7A, red line), the decay rate of SypHy fluorescence is much slower (Fig 7A and 7C). To verify whether this defect is specific to loss of ratFwe2, we expressed mFwe2-HA in ratfwe knockout neurons. mFwe2-HA properly localizes in the Golgi (S9E Fig) as well as in SVs (Fig 6F and 6G). This protein further normalizes the slow SypHy fluorescence decay (Fig 7A and 7C, blue line). Recent studies have revealed distinct properties of SV endocytosis under physiological conditions [15,45,46]. We found similar results when these recordings were performed at physiological temperatures (Fig 7B and 7D). A slow decay of SypHy fluorescence is possibly due to either impaired CME or inefficient SV reacidification or both. To distinguish these hypotheses, we performed an acidic quenching assay [47,48]. As shown in S10 Fig, upon perfusion of an acidic buffer, the newly recycled SVs in both control and ratFwe knockout neurons are efficiently acidified. Hence, our data suggest that ratFwe2 promotes CME at mammalian central synapses.
To assess the role of ratFwe2 in ADBE, we performed a dextran dye uptake assay in the control and ratFwe knockout neurons. We triggered ADBE with a stimulation of 1,600 action potentials delivered at 80 Hz in the presence of 40 kDa tetramethylrhodamine (TMR)-dextran [15,49]. As shown in Fig 7E–7H, in the control axons marked with green fluorescent protein (GFP) expression, the presynaptic terminals filled with dextran dye (red puncta) were observed frequently. In contrast, removal of ratFwe significantly diminishes dextran dye uptake. This phenotype is specific to loss of ratFwe2, as the reintroduction of mFwe2-HA corrects this dye uptake defect. Hence, ratFwe2 is indispensable for ADBE. In summary, Fwe promotes CME and ADBE in mammalian neurons, thereby coupling exocytosis to two major modes of endocytosis.
A tight coupling of exocytosis and endocytosis is critical for supporting continuous exocytosis of neurotransmitters. CME and ADBE are well-characterized forms of SV endocytosis triggered by moderate and strong nerve stimuli, respectively. However, how they are coupled with exocytosis under distinct stimulation paradigms remains less explored. Based on the present data, we propose a model as shown in Fig 7I. When presynaptic terminals are mildly stimulated, SV release leads to neurotransmitter release and the transfer of Fwe channel from SVs to the periactive zone where CME and ADBE occur actively [7,26,38]. Our data suggest that this channel does not supply Ca2+ for CME to proceed. However, intense activity promotes Fwe to elevate presynaptic Ca2+ levels near endocytic zones where ADBE is subsequently triggered. Thus, Fwe exerts different activities and properties in response to different stimuli to couple exocytosis to different modes of endocytosis.
We previously concluded that Fwe-dependent Ca2+ influx triggers CME [24]. However, the current results suggest alternative explanations. First, the presynaptic Ca2+ concentrations elicited by moderate activity conditions, i.e., 1-min 90 mM K+/0.5 mM Ca2+ or 20-s 10–20 Hz electric stimulation, are not dependent on Fwe (Fig 3G and 3H). Second, expression of 4% FweE79Q, a condition that abolishes Ca2+ influx via Fwe (Fig 3F), rescues the CME defects associated with fwe mutants, including decreased FM1-43 dye uptake, a reduced number of SVs, and enlarged SVs (Fig 1). Third, raising the presynaptic Ca2+ level has no beneficial impact on the reduced number of SVs observed in fwe mutants (Fig 4J). These data are consistent with the observations that a Ca2+ influx dependent on VGCCs triggers CME at a mammalian synapse [18,23]. Hence, Fwe acts in parallel with or downstream to VGCC-mediated Ca2+ influx during CME.
ADBE is triggered by intracellular Ca2+ elevation, which has been assumed to be driven by VGCCs that are located at the active zones [18,26]. However, our data strongly support a role for Fwe as an important Ca2+ channel for ADBE. First, following exocytosis, Fwe is enriched at the periactive zone where ADBE predominates [7,24,26]. Second, Fwe selectively supplies Ca2+ to the presynaptic compartment during intense activity stimulation (Fig 3), which is highly correlated with the rapid formation of ADBE upon stimulation [8,50]. Third, 4% FweE79Q expression, which induces very subtle or no Ca2+ upon strong stimulation, fails to rescue the ADBE defect associated with loss of fwe (Figs 3 and 4). Fourth, treatment with a low concentration of La3+ solution that specifically blocks the Ca2+ conductance of Fwe significantly abolishes ADBE (Fig 5). Lastly, the role of Fwe-derived Ca2+ influx in the initiation of ADBE mimics the effect of Ca2+ on ADBE at the rat Calyx of Held [7]. As loss of fwe does not completely eliminate ADBE, our results do not exclude the possibility that VGCC may function in parallel with Fwe to promote ADBE following intense stimulation.
Interestingly, Ca2+ influx via Fwe does not control SV exocytosis during mild and intense stimulations (S3 Fig). How do VGCC and Fwe selectively regulate SV exocytosis and ADBE, respectively? One potential mechanism is that VGCC triggers a high, transient Ca2+ influx around the active zone that elicits SV exocytosis. In contrast, Fwe is activated at the periactive zone to create a spatially and temporally distinct Ca2+ microdomain. A selective failure to increase the presynaptic Ca2+ level during strong stimulation is evident upon loss of fwe. This pinpoints to an activity-dependent gating of the Fwe channel. Consistent with this finding, an increase in the level of Fwe in the plasma membrane does not lead to presynaptic Ca2+ elevation at the Calyx of Held when the presynaptic terminals are at rest or subject to mild stimulation [23]. However, we previously showed that, in shits terminals, blocking CME results in the accumulation of the Fwe channel in the plasma membrane, elevating Ca2+ levels [24]. It is possible that Dynamin is also involved in regulating the channel activity of Fwe or that the effects other than Fwe accumulation associated with shits mutants may affect intracellular Ca2+ handling [51,52]. Further investigation of how neuronal activity gates the channel function of Fwe should advance our knowledge on the activity-dependent exo–endo coupling.
Although a proteomic analysis did not identify ratFwe2 in SVs purified from rat brain [53], our biochemical analyses show that ratFwe2 is indeed associated with the membrane of SVs. Our data show that 4% of the total endogenous Fwe channels efficiently promotes CME and ADBE at the Drosophila NMJ. If a single SV needs at least one functional Fwe channel complex during exo–endo coupling, and one functional Fwe complex comprises at least four monomers, similar to VGCCs, transient receptor potential cation channel subfamily V members (TRPV) 5 and 6, and calcium release-activated channel (CRAC)/Orai1 [24,40,54,55], then we anticipate that each SV contains ~100 Fwe proteins (4 monomers × 25). This suggests that Fwe is highly abundant on the SVs. It is unlikely that many SVs do not have the Fwe, as a 25-fold reduction of the protein is enough to ensure functional integrity during repetitive neurotransmission. Finally, our results for the SypHy and dextran uptake assays at mammalian central synapses indicate the functional conservation of the Fwe channel in promoting different modes of SV retrieval. In summary, the Fwe-mediated exo–endo coupling seems to be of broad importance for sustained synaptic transmission across species.
Detailed protocols are available at protocols.io (http://dx.doi.org/10.17504/protocols.io.hgbb3sn).
Most of the experiments used y w; FRT80B isogenized fly, which was used for the generation of the fweDB25 and fweDB56 mutations [56] as the controls. Larvae were reared in standard fly food or on grape juice agar covered with yeast paste at 22°C. The genotypes of flies used in the experiments are described below.
For GCaMP6f imaging and immunostaining, the genotypes that carry vglut-lexA and lexAop2-GCaMP6f are as follows:
The genotypes that carry nSyb-GAL4 and UAS-GCaMP6f are as follows:
The following UAS transgenes were made in this study:
For electrophysiology, the following were used:
For Cac-EGFP expression, the following were used:
Those used for Fluo-4 AM experiments in salivary glands are as follows:
Those used for TEM analysis, FM1-43 dye uptake assay, immunostaining of HA, and Fwe are as follows:
The pCasper4-genomic HA-fwe construct was constructed by inserting a HA sequence to the site after the translational start codon of fwe-RB in the context of the pCasper4-genomic fwe construct [24]. To obtain the pUAST-flag-mFwe2-HA construct, the fwe-RB fragment of the pUAST-flag-fwe-RB-HA construct [24] was replaced with the mFwe2 coding region, which was amplified from total mRNA of the adult mouse brain. P-element-mediated transgenesis was achieved by the standard procedure. The introduction of these genomic fwe transgenes to the fwe mutant background rescues the early lethality associated with fwe mutants, demonstrating that fused tags do not affect normal functions of Fwe. To generate the mFwe-miRNAi constructs, the sequences of miRNAs were designed according to Invitrogen’s RNAi Designer. mFwe-miRNAi-1 targets nucleotides 183–203 of the mFwe2 coding sequence (forward oligomer: TGCTGAAGG CGTTCATGATCATCCACGTTTTGGCCACTGACTGACGTGGATGAATGAACGCCTT; reverse oligomer: CCTGAAGGCGTTCATTCATCCACGTCAGTCAGTGGCCAAAACGTGG ATGATCATGAACGCCTTC). mFwe-miRNAi-2 targets nucleotides 236–256 of the mFwe2 coding sequence (forward oligomer: TGCTGTTG CAAACTCCACAAACTGGCGTTTTGGC CACTGACTGACGCCAGTTTGGAGTTTGCAA; reverse oligomer: CCTGTTGCAAACTCC AAACTGGCGTCAGTCAGTGGCCAAAACGCCAGTTTGTGGAGTTTGCAAC). Synthetic oligomers were annealed and subcloned to pcDNA 6.2-GW/EmGFPmiR vector (Invitrogen). To generate the pSpCas9(BB)-based plasmids, pSpCas9(BB)-2A-tagRFP was constructed by replacing the GFP region of the pSpCas9(BB)-2A-GFP plasmid (addgene#48138) [43] with a TagRFP coding sequence. m/ratFwe-gRNA was designed to target the first intron/second exon junction of the mFwe and ratFwe (forward oligomer: CACCGTTTGAAGCCTGTGCCATCTC; reverse oligomer: TTTGCTCTACCGTGTCCGAAG TTTG). Annealed synthetic oligomers were placed into the BbsI site of pSpCas9(BB)-2A-GFP and pSpCas9(BB)-2A-tagRFP to obtain pSpCas9(BB)-m/ratFwe-gRNA-2A-GFP and pSpCas9(BB)-m/ratFwe-gRNA-2A-tagRFP constructs. pSpCas9(BB)-m/ratFwe-gRNA-2A-GFP-2A-mFwe2-HA and pSpCas9(BB)-m/ratFwe-gRNA-2A-tagRFP-2A-mFwe2-HA were generated by inserting the DNA fragment of 2A-mFwe2-HA, which was amplified by PCR with the primers (forward primer: AAAAGCTTGGCAGTGGAGAGGGCAGAGGAAGTCTGCTAACATGCGGTGACGTCG AGGAGAATCCTGGCCCAAGCGGCTCGGGCGCC; reverse primer: CCCTCGAGTTA CGCGTAGTCGGGGAC), after GFP or tagRFP.
Total RNA of different regions of the adult rat brain were extracted with TRIZOL reagent according to the manufacturer's instructions. Five μg of total RNA was mixed with oligo-dT primer in 20 μl of reverse transcription reaction solution. One μl of this mixture was used to amplify the cDNA of ratFwe2 mRNA with specific primers (forward primer: GAAGATCTATGAGCGGCTCGGTCGCC; reverse primer: CGGAATTCTCACAGTTCCCCCTCGAATG). Twenty-five PCR cycles were used to allow exponential PCR amplification. The PCR products were sequenced to validate the identity of ratFwe2 mRNA.
GST-fused polypeptides comprising seven tandem repeats of the C-terminus of Drosophila Fwe-PB isoform [24] were injected in guinea pigs to obtain GP100Y antisera. To generate α-m/ratFwe2 antisera (GP67), the DNA fragment encoding seven tandem repeats of ratFwe2 C-terminus (a.a. 140–172) was subcloned to pET28a plasmid. His-fused polypeptides were purified and then injected into guinea pigs. Antibody generation was assisted by LTK BioLaboratories (Taiwan). Specific antibodies were further purified by antigen-conjugated affinity columns.
For immunostaining in fly NMJ boutons, larval fillets were dissected in ice-cold 1X PBS and fixed in 4% paraformaldehyde for 20 min at room temperature. The samples were permeabilized in 0.1% Triton X-100-containing 1X PBS solution for all staining, except staining with anti-Fwe (GP100Y) and anti-HA antibodies, which used 0.1% Tween-20-containing 1X PBS solution to prevent Fwe dissociation from the SVs. Primary antibody dilutions used mouse anti-Dlg (mAb 4F3) [57], 1:100 (Hybridoma bank) [58]; mouse anti-Brp (nc82), 1:100 (Hybridoma bank) [59]; rabbit anti-GFP, 1:500 (Invitrogen); mouse monoclonal anti-HA, 1:200 (Sigma); rabbit anti-HA, 1:200 (Sigma); rabbit Cy3 conjugated anti-HRP, 1:500 (Jackson ImmunoResearch); and guinea pig anti-Fwe-PB (GP100Y), 1:100. Secondary antibodies were diluted to 1:500 (Jackson ImmunoResearch and Invitrogen). For immunostaining of cultured rat hippocampal neurons, DIV14 neurons were fixed in 4% paraformaldehyde/4% sucrose for 10 min at room temperature and permeabilized and washed with 0.1% Tween-20-containing 1X PBS solution. To reduce nonspecific staining, GP67 antibodies were absorbed with paraformaldehyde-fixed n2a cells before staining. Primary antibody dilutions used rabbit anti-GFP, 1:500 (Invitrogen); guinea pig ant-m/ratFwe2 (GP67), 1:100; rabbit anti-GM130, 1:500 (Adcam); mouse anti-HA, 1:200 (Sigma); and mouse anti-SVP38, 1:200 (Sigma). Secondary antibodies were diluted to 1:500 (Jackson ImmunoResearch and Invitrogen). DAPI was used in the 1:2,000 dilution (Invitrogen). To compare the staining intensity of boutons among different genotypes, larval fillets used in the same graph were stained in the same Eppendorf tube. The images were captured using a Zeiss 780 confocal microscope, and the scan setup was fixed for the same experimental set. For data quantifications, single-plane confocal images were projected. The final staining intensity in boutons was calculated by subtracting the background fluorescence intensity in the surrounding muscles from the staining intensity in boutons. The staining intensities of all type Ib boutons from the same muscles 6 and 7 in one image were averaged to obtain each data value. Image processing was achieved using LSM Zen and Image J.
Mouse neuroblastoma n2a cells were transfected with pSpCas9(BB)-m/ratFwe-gRNA-2A-GFP plasmid. GFP-positive cells were sorted out using flow cytometry, and the cells were plated in a 96-well plate in which each well included approximately one cell. After 3-wk culture, single cell-driven colonies were subjected to immunostaining and immunoblotting for mFwe2 to verify the knockout of mFwe2. One of the confirmed mFwe knockout neuroblastoma n2a cell lines was used in S9A Fig.
For western blotting of n2a cell lysates, the cells lysed with RIPA buffer (50 mM Tris-HCl [pH 8], 150 mM NaCl, 1% NP-40, 0.5% sodium deoxycholate, and 0.1% SDS) were boiled in 1X SDS sample buffer for 10 min. To prepare subcellular fractions of adult rat brain, one adult brain (~1 g) was homogenized in 5 ml 0.32 M sucrose buffer (320 mM sucrose, 1.5 mM MgCl2, 1 mM EGTA, and 10 mM HEPES [pH 7.5]) using a Teflon glass homogenizer. The homogenates (H) were centrifuged at 800 × g for 15 min at 4°C to yield pellets (P1) and supernatants (S1). S1 supernatants were centrifuged at 9,200 × g for 15 min at 4°C to obtain pellets (P2) and supernatants (S2), which were centrifuged at 100,000 × g for 2 h at 4°C to obtain fractions of cytosol (S3) and light membrane (P3). The pellets (P2) were then lysed in ice-cold Mini-Q water, followed by equilibration with 4 mM HEPES. After 30-min mixing at 4°C, the lysates were centrifuged at 25,000 × g for 20 min at 4°C to yield the crude synaptic vesicle fraction (LS1) and lysed synaptosomal membrane fraction (LP1). The LS1 fraction was further centrifuged at 100,000 × g to obtain crude synaptic vesicles (LP2) and the synaptosomal cytosol fraction (LS2). A discontinuous sucrose gradient from 0.3–0.99 M was prepared by gradually layering the different concentrations of sucrose. The S1 supernatants were loaded on sucrose gradient solution and centrifuged at 33,000 × g for 3 h at 4°C. Fractions were collected from low- to high-density sucrose. These fractions were boiled in 1X SDS sample buffer and subjected to SDS-PAGE and western blotting. The primary antibody dilutions used were as follows: guinea pig anti-m/ratFwe2 (GP67), 1:500; rabbit anti-SVP38, 1:1,000 (Sigma); rabbit anti-GM130, 1:5,000 (Abcam); mouse anti-Tubulin, 1: 10,000 (Sigma); and mouse anti-α-Actin, 1:10,000 (Sigma). Secondary HRP-conjugated antibodies were diluted to 1:5,000 (Jackson ImmunoResearch).
To induce CME, the third instar larvae were dissected in 0 mM Ca2+ hemolymph-like (HL)-3 solution at room temperature (70 mM NaCl, 5 mM KCl, 10 mM MgCl2, 10 mM NaHCO3, 5 mM trehalose, 5 mM HEPES [pH 7.2], and 115 mM sucrose) [60] and subjected to 1-min 90 mM K+/0.5 mM Ca2+ stimulation (25 mM NaCl, 90 mM KCl, 10 mM MgCl2, 10 mM NaHCO3, 5 mM trehalose, 5 mM HEPES [pH 7.2], 30 mM sucrose, and 0.5 mM CaCl2) or 10-min 60 mM K+/1 mM Ca2+ stimulation (55 mM NaCl, 60 mM KCl, 10 mM MgCl2, 10 mM NaHCO3, 5 mM trehalose, 5 mM HEPES [pH 7.2], 30 mM sucrose, and 1 mM CaCl2) in the presence of 4 μM fixable FM1-43 (Invitrogen). Excess dye was extensively washed with 0 mM Ca2+ HL-3 solution for 10 min. Larval fillets were fixed in 4% paraformaldehyde for 10 min, washed, mounted, and imaged on a Zeiss 780 confocal microscope. The scan setup was fixed for all the sets of the experiments. For data quantifications, single-plane confocal images were projected, and the final FM1-43 dye intensity in the boutons was calculated by subtracting the dye fluorescence intensity in the surrounding muscles from the dye fluorescence intensity within the boutons. The dye fluorescence intensities of all type Ib boutons from the same muscles 6 and 7 were averaged to obtain each data value. For chlorpromazine treatment experiment, dissected larvae were incubated with 200 μM chlorpromazine (Sigma) in Schneider medium for 30 min. They were then stimulated with a solution of 90 mM K+/2 mM Ca2+/200 μM chlorpromazine (25 mM NaCl, 90 mM KCl, 10 mM MgCl2, 10 mM NaHCO3, 5 mM trehalose, 5 mM HEPES [pH 7.2], 30 mM sucrose, 2 mM CaCl2, and 200 μM chlorpromazine) in the presence of 4 μM fixable FM1-43 for 10 min. Bulk membranous invaginations were defined as the internalized structures labeled with high levels of FM1-43 dye. The areas of individual type Ib boutons and bulk membranous invaginations were measured using Image J. For FM1-43 dye loading/unloading assays, larvae were dissected in 0 mM Ca2+ HL-3 solution at room temperature and subjected to a stimulation of 90 mM K+/0.5 (or 2) mM Ca2+ HL-3 solution for 5 min to load boutons with the FM1-43 dye. Excess dye was removed by extensive washing with 0 mM Ca2+ HL-3 solution for 10 min. FM1-43 dye uptake by boutons was imaged to indicate “loading.” Subsequently, the dye loaded in SVs was unloaded by stimulation using 90 mM K+/0.5 (or 2) mM Ca2+ solution for 1 min. Released dye was removed by several washes with a 0 mM Ca2+ HL-3 solution. The remaining dye in boutons was imaged to indicate “unloading.” The final FM1-43 dye intensity in the boutons was calculated by subtracting the dye fluorescence intensity in the surrounding muscles from the dye fluorescence intensity within the boutons. The dye fluorescence intensities of at least ten type Ib boutons from the same muscles 6 and 7 were averaged to obtain each data value. The dye unloading efficiency was indicated as (Fload-Funload)/Fload. Images processing was achieved using Image J and LSM Zen.
The third instar larvae were dissected in 0 mM Ca2+ HL-3 at room temperature and then bathed in 1 mM Ca2+ HL-3 solution for 5–10 min before the recording. The mean value of the resistance of the recording electrode was ~40 MΩ when the electrode was filled with a 3M KCl solution. All recordings were obtained from muscle 6 of abdominal segment 3. Each larva was only used for one recording. Recordings from the muscles that hold resting membrane potentials at less than −60 mV were used for further data quantifications. EJPs were evoked by stimulating the axonal bundle via a glass capillary electrode with an internal diameter of ~10–15 μm (Harvard Apparatus Glass Capillaries GC120F-15) at 0.2 Hz. Stimulus pulses were fixed at 0.5 ms duration (pClamp 10.6 software, Axon Instruments). To obtain maximal EJP amplitude, 3–5 mV electric stimuli were applied. EJPs were amplified with an Axoclamp 900A amplifier (Axon Instruments, Foster City, California) under bridge mode and filtered at 10 kHz. EJPs were analyzed by pClamp 10.6 software (Axon Instruments). For the EJP amplitude at 0.2 Hz, the mean of the EJP amplitude was averaged from the amplitudes of 80 EJPs in one consecutive recording.
Larval fillets were dissected in 0 mM Ca2+ HL-3 medium at room temperature. To trigger CME, samples were stimulated with 90 mM K+/0.5 mM Ca2+ HL-3 solution for 1 min or 60 mM K+/1 mM Ca2+ HL-3 solution for 10 min. To induce ADBE, larval fillets were subjected to stimulation of a 90 mM K+/2 mM or 5 mM Ca2+ HL-3 solution in the presence or lack of 10 μM La3+ for 10 min. Subsequently, the samples were fixed for 12 h in 4% paraformaldehyde/1% glutaraldehyde/0.1 M cacodylic acid (pH 7.2) solution and then rinsed with 0.1 M cacodylic acid (pH 7.2) solution. They were subsequently fixed in 1% OsO4/0.1 M cacodylic acid solution at room temperature for 3 h. The samples were subjected to a series of dehydration from 30% to 100% ethanol. After 100% ethanol dehydration, the samples were incubated in propylene, a mixture of propylene and resin, and pure resin. Lastly, they were embedded in 100% resin. The images of type Ib boutons were captured using Tecnai G2 Spirit TWIN (FEI Company) and a Gatan CCD Camera (794.10.BP2 MultiScan) at ≥4,400 × magnifications. The size of the SVs and the bulk cisternae and the area of type Ib boutons were measured using Image J. We identified type Ib boutons by multiple layers of subsynaptic reticulum. The radius of the bulk cisternae was calculated from A(area) = πr2. Isolated membranous structures larger than 80 nm in diameter were defined as bulk cisternae.
For SypHy imaging, DIV7 cultured rat hippocampal neurons were transfected with pSpCas9(BB) and pCMV-SyphyA4 (addgene#24478) plasmids [5]. DIV13–15 neurons were bathed in the imaging buffer in a chamber (Warner instruments RC-25F) with two parallel platinum wires separated by 5 mm. The imaging buffer consisted of 136 mM NaCl, 2.5 mM KCl, 2 mM CaCl2, 1.3 mM MgCl2, 10 mM glucose, 10 mM HEPES (pH 7.4), 10 μM CNQX, and 50 μM AP-5. SV exocytosis was elicited with a train of 200 action potentials delivered with a 20-Hz electric field stimulation (50 mA, 1-ms pulse width). Single images were captured every 1 s using MetaMorph software and an ANDOR iXon 897 camera. The experiments were performed at either room or physiological temperatures controlled by a Warner temperature controller (TC-344B). For SV reacidification experiments, the imaging chamber was perfused with the imaging buffer (136 mM NaCl, 2.5 mM KCl, 2 mM CaCl2, 1.3 mM MgCl2, 10 mM glucose, and 10 mM HEPES [pH 7.4], 10 μM CNQX, and 50 μM AP-5) before perfusion with an acidic buffer (136 mM NaCl, 2.5 mM KCl, 2 mM CaCl2, 1.3 mM MgCl2, 10 mM glucose, and 10 mM 2-[N-morpholino] ethane sulphonic acid [pH 5.5], 10 μM CNQX, and 50 μM AP-5), which was prepared by replacing HEPES in the imaging buffer with 2-[N-morpholino] ethane sulphonic acid [47,48]. Next, the imaging buffer was perfused to allow surface SypHy to be fluorescent. Experimental temperatures were maintained at physiological temperatures. The final SypHy fluorescence intensities in the presynaptic terminals were calculated by subtracting the background fluorescence intensity on the surrounding coverslip from the SypHy fluorescence intensity within presynaptic terminals. Each data value was obtained from a single terminal. For 40 kDa TMR-dextran uptake assays, DIV13–15 neurons transfected with pSpCas9(BB) plasmids were stimulated by a train of 1,600 action potentials delivered with an 80 Hz electric field stimulation (50 mA, 1-ms pulse width) in the imaging solution (144 mM NaCl, 2.5 mM KCl, 2.5 mM CaCl2, 2.5 mM MgCl2, 10 mM HEPES [pH 7.5], 10 μM CNQX, and 50 μM AP-5) [49] in the presence of 50 μM 40 kDa TMR-dextran (Invitrogen). Subsequently, neurons were perfused with the same buffer for 5 min to remove excess dextran dye. Experiments were performed at room temperature. Imaging was achieved through MetaMorph software and an ANDOR iXon 897 camera. Image processing was achieved using Image J and LSM Zen.
Paired and multiple data sets were compared by Student’s t test and one-way ANOVA statistical analyses, respectively. All data analyses were achieved using GraphPad Prism 7.0. The numerical data used in all figures are included in S1 Data.
|
10.1371/journal.pcbi.1007264 | Machine learning-based microarray analyses indicate low-expression genes might collectively influence PAH disease | Accurately predicting and testing the types of Pulmonary arterial hypertension (PAH) of each patient using cost-effective microarray-based expression data and machine learning algorithms could greatly help either identifying the most targeting medicine or adopting other therapeutic measures that could correct/restore defective genetic signaling at the early stage. Furthermore, the prediction model construction processes can also help identifying highly informative genes controlling PAH, leading to enhanced understanding of the disease etiology and molecular pathways. In this study, we used several different gene filtering methods based on microarray expression data obtained from a high-quality patient PAH dataset. Following that, we proposed a novel feature selection and refinement algorithm in conjunction with well-known machine learning methods to identify a small set of highly informative genes. Results indicated that clusters of small-expression genes could be extremely informative at predicting and differentiating different forms of PAH. Additionally, our proposed novel feature refinement algorithm could lead to significant enhancement in model performance. To summarize, integrated with state-of-the-art machine learning and novel feature refining algorithms, the most accurate models could provide near-perfect classification accuracies using very few (close to ten) low-expression genes.
| Pulmonary arterial hypertension (PAH) is a serious and progressive disease, with only a roughly 50% of 5-year survival rate even with best available therapies. Accurately detecting/differentiating different forms of PAH and developing drugs that could directly target at genes involved in PAH pathogenesis are essential. We proposed a computational approach using low-cost microarray data collected from a clinical trial and had accurately predicted each PAH group. In particular, we considered the fact that there might exist some low-expression genes that were usually discarded by researchers but might function collectively and significantly controlling the disease in each case. Therefore, we had developed different filtering algorithms that intentionally selected those low-expression genes for constructing prediction model. Using a few highly informative low-expression genes that had never been extensively investigated before, our systematic approach had produced models that could offer prefect accuracy in predicting PAH. Additionally, our analysis also found that the composition of gene factors controlling the PAH etiology under each form are quite different from each other.
| Pulmonary arterial hypertension (PAH) is a fatal and progressive disease characterized by increasing pulmonary vascular resistance leading to heart failure and death [1–4]. A significant amount of research has been conducted previously, which greatly enhanced understanding of the molecular mechanisms and etiology involved in PAH. However, the underlying interactive effects of different genetic mutations and fundamental mechanisms of vascular dysfunction remains unclear [4]. For example, mutation in bone morphogenetic protein (BMP) receptor type II (BMPR2) was found to be significantly correlated with the development of both heritable (HPAH) and the idiopathic form of PAH (IPAH) [5]. In another study, a known BMP signaling regulator, transcription factor MSX1, was found to be strongly correlated with IPAH cases [1]. Additionally, genes expression signatures from IPAH patients were either tightly clustered with HPAH group or in an isolated cluster [4]. These findings suggest that different forms of PAH might share the majority of the molecular origins/signaling pathways but there might exist some distinct factors modulating the primary genetic expression in each case [1]. Furthermore, the majority of the PAH cases in human beings were found to be unassociated with BMPR2 mutation [1], and some other factors have been identified to be partially contributing to IPAH [6,7]. We attributed the difficulty in fully unraveling the genetic factors causing PAH largely to the lack of high-quality patient data in conjunction with advanced data processing algorithms, limited comprehension of the genetic etiology, and overlook of some of the important low-expression genes that might interactively affect PAH as a whole.
Microarray-based gene sequencing provides a fast and cost-effective screening technology, and has been used in many PAH studies to identify important signaling pathways that could impact PAH pathogenesis [3,8,9]. Particularly, more than 25 microarray studies have been conducted on human PAH in the past ranging from single-gene expression to more complex pathway analyses, providing large quantity of data pertaining to PAH pathogenesis. Following common data processing protocols, the majority of these microarray data analysis routines involve background noise reduction and normalization. Additionally, the differentially expressed genes are typically ranked by their logarithmically transformed fold change values or by moderated t-statistic values. The low-expression genes [e.g. log-transformed probe intensity values < 27 or 28, or the inter-quartile range (IQR) detected threshold values] are typically considered unreliable and treated as ‘noises,’ which are usually removed from the dataset manually [10–12] or by applying the IQR filtering algorithm provided by software packages such as the Bioconductor [13, 14]. However, we expect that these low-expression genes might have significant interactive effects in PAH etiology and might become very informative when functioning collectively. We hypothesized that using advanced data-science algorithms in conjunction with high-quality clinical research data would reveal significant coexisting controlling factors and interactive genes (including those with low-expression values) that could impact PAH pathogenesis.
The primary goal of this study was to identify a small group of genes (including low-expression genes) and construct classification models that could accurately predict each patient’s PAH status. To accomplish this, we ranked and selected genes according to their contribution towards the model construction processes. Particularly, our research group had formulated a novel recursive feature elimination algorithm integrated with conventional machine learning data analysis paradigm. Three popular supervised machine learning algorithms were selected for the modeling processes, including Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), and Artificial Neural Networks (ANN). All three algorithms rely on labelled training data to find a linear combination of features (e.g. LDA), a maximized margin of a decision boundary (e.g. SVM), or an optimized neuron-edge network structure (e.g. ANN) that could best separate data belonging to different groups. Additionally, this project investigated the possibility of constructing highly accurate prediction models in determining different forms of PAH, which, if worked, could provide great potential for future clinical usage and commercialization. The main dataset was obtained as Affymetrix array-based gene expression data from anticoagulated whole blood samples collected from 86 patients, which includes the healthy control group (22 patients), the IPAH group (20 patients, BMPR2 mutation excluded), the HPAH (17 patients) and the BMPR2 mutation carriers that have no clinical signs of disease (27 Unaffected Mutation Carriers, UMC). Our classification models indicated excellent performance in distinguishing each individual patient group (Control vs. IPAH vs. HPAH vs. UMC); and separating the control group vs. the combination of HPAH and IPAH, as well as the HPAH group vs. UMC. Additionally, different gene filtering criteria were used to test the assumption that there might exist clusters of important low-expression genes collectively controlling the forms of PAH in human beings.
The overall data analysis scheme/workflow is presented in Fig 1. Without using any filtering routine that is commonly embedded in many microarray software packages for duplication removal and high IQR probe retention, our preliminary filtering methods resulted in dramatic differences in the number of retained probe sets as indicated in Fig 2. Particularly, removing genes with at least one group average smaller than 256 (or synonymously AGA>256) yielded only 10,230 features, which is 18.7% of the entire gene count from the raw microarray results. Likewise, if only genes with at least one group average smaller than 256 were selected (ALOGA<256), 44,383 genes would be retained, accounting for 81.3% (equals 1–18.7%) of the entire gene count. For comparison purposes, the conventional IQR-detected threshold filtering (All>12, which corresponds to probe-sets with expression profile IQR > 0.3) allowed 38,597 genes pass the screening process [13,14].
For each combination of classification tasks and feature filtering methods, we investigated the modeling performance synthesizing results from repeated MCCV, including feature ranking, model training and testing. Furthermore, average classification accuracies were reported based on the number of ranked features (refined or unrefined) used by each model that could lead to the lowest error rate (or synonymously, the highest accuracy). Again, we anticipated that this number (number of ranked features providing the greatest accuracy) could be highly variable depending on the nature of the classification task per se as well as the feature ranking and classification algorithms used under each routine.
Fig 3 depicts the error rate (measured as the number of miss-predicted patients divided by the total number of patients) change when more genes from the ranked feature list were included in the prediction model. Solid or empty markers indicate algorithms incorporating SIRRFE-based refinement, or without, respectively. Edge color differences indicate different classification algorithms (black-ANN, blue-LDA, and red-SVM).
Regardless of incorporating SIRRFE or not, LDA-based classification algorithms always resulted in an error peak across all feature filtering methods. Using very few (less than 50) ranked genes, both SIRRFE-refined SVM and LDA algorithms produced lower error rates compared with other data processing routines. Generally speaking, the ANN-based algorithms were not as accurate as the other two. The enhanced modeling performance, particularly in the top 200 genes range, caused by SIRRFE were extremely significant under All Genes (Fig 3A), All>12 as identified by the IQR method (Fig 3B), ALOGA<128 (Fig 3H), and ALOGA<256 (Fig 3H) filtering methods; and moderately significant under ALOGA>128 (Fig 3E) and ALOGA>256 (Fig 3F) methods. Across all feature filtering methods, it was greatly evident that both ALOGA<128 (Fig 3G) and ALOGA<256 (Fig 3H) had resulted in incomparably superior performance than others. Both filtering methods provided very small error rates using very few ranked genes (less than 50). Additionally, the ALOGA<128 (Fig 3G) filtering method appeared to be the optimal. SIRRFE refined SVM routines provided stable and satisfactory performance across all filtering methods. Additionally, SIRRFE refined LDA routines produced superior results around the top 370 genes range compared with SVM, but used substantially greater number of features. Interestingly, both AGA>128 (Fig 3C) and AGA>256 (Fig 3D) filtering methods tended to produce the worst results (>0.24 error rates), followed by both ALOGA>128 (Fig 3E) and ALOGA>256 (>0.23 error rates, Fig 3F). It was also noteworthy that it seemed inclusion of more ranked genes in those non-SIRRFE refined routines could always boost performance; however, adding more than 200 SIRRFE refined ranked genes into different classification algorithms could lead to little or even compromised performance. This accented the efficiency and effectiveness of our SIRRFE-based feature ranking/selection algorithms.
Comparing all classification algorithms used in this task, both ANN and SVM yielded very similar performance, and both appeared to be better algorithms than LDA as indicated Fig 4. This was particularly obvious when very small quantity of ranked genes were used (<50). Regardless feature filtering methods, SIRRFE provided little improvement when LDA algorithm was used. Furthermore, under AGA>256 (Fig 4D), no improvement was observed across all three classification algorithms. However, SIRRFE generally enhanced model classification performance of both ANN and SVM-based algorithms. We also found similar error peaks while using LDA-based algorithms as reported in the previous session. Likewise, we observed that different feature filtering methods could have tremendous impact on the modeling results. In this binary classification case, we found that ALOGA<128 (Fig 4G), ALOGA<256 (Fig 4H), and no filtering (Fig 4A) could all provide very low error rates. Particularly, ALOGA<128 (Fig 4G) and 256 (Fig 4H) generated perfect classification (zero error rate) using less than 20 genes. Additionally, the majority of classification routines reached optimal accuracy within the top 100 genes ranking range, and the performance enhancement for including more genes diminished rapidly.
Overall, the impacts of various feature filtering methods on the model performance significantly outweighed that attributed to different classification algorithms as indicated in Fig 5. Apparently, any kind of “denoising” filtering method [e.g. AGA>128, 256 (Fig 5C and 5D); All>12, (Fig 5B); or ALOGA>128,256 (Fig 5E and 5F)] had caused increased error rates. Additionally, both ALOGA<128 (Fig 5G) and ALOGA<256 (Fig 5H) as well as All-Genes-Included (Fig 5A) methods provided superb results (zero error) under all classification algorithms. This effect became even more pronounced when SIRRFE was incorporated. In general, all algorithms worked equally well under ALOGA<128 (Fig 5G), ALOGA<256 (Fig 5H), and All-Genes-Included methods (Fig 5A); except for the similar error peak induced by LDA as reported in the other classification tasks described preciously. Again, we observed significant performance boost using SIRRFE. One finding that was very unique for this classification task was that inclusion of more (>50) ranked genes in our classification routine had significantly increased error rates except for those high-performance filtering methods [e.g. ALOGA<128, 256, and All (Fig 5A, 5G and 5H)].
As expected, the multiclass classification for distinguishing each individual PAH patient group was the most challenging classification task, which required larger number of genes for achieving acceptable accuracies and the final outcome depended greatly on the type of classification algorithms used. Additionally, the performance enhancement using SIRRFE was very significant. The binary classification of the healthy control vs. the combination of HPAH and IPAH seemed to be much easier, with the majority of models yielding zero error rates using less than 20–30 genes. The contributions of different classification algorithms and SIRRFE refinement were not as great as those filtering methods. Finally, the binary classification task involving HPAH vs. UMC appeared to be the most responsive towards different feature filtering methods, with significant advantages observed under ALOGA<128, 256, or All-Genes-Included methods.
To further examine the optimal models that provided the lowest error rates (highest accuracy) using the least number of ranked genes, we provided a detailed summarization in Table 1. Additionally, considering that SIRRFE generally resulted in improved performance as indicated in Figs 3–5, we only reported the optimal SIRRFE-refined results for simplicity purposes. For the multiclass classification (Control vs. IPAH vs. HPAH vs. UMC), our best routine (ALOGA<128 with ANOVA+SIRRFE+LDA) achieved an accuracy value of 0.98 using 199 ranked genes. On average, ALOGA<128 appeared to be the optimal filtering method (average accuracy = 0.96) and LDA algorithm seemed to be the best classification algorithm without considering the numerical stability and the error peak observed in Fig 3. For the binary classification of the healthy control vs. the combination of HPAH and IPAH, the majority of the routines yielded perfect accuracies. Therefore, judging from the number of ranked genes for achieving the perfect prediction, both SIRRFE-refined LDA and SVM algorithms with ALOGA<128 filtering provided perfect accuracy using only eight genes. Overall, the ALOGA<128 filtering method appeared to be the optimal (average accuracy = 1, average number of ranked genes used = 11) and SVM seemed to be the best classification algorithm (average accuracy = 0.99, average number of ranked genes used = 19). Finally, the optimal model for the binary classification of HPAH vs. UMC was obtained using the ALOGA<128 filtering method and the SIRRFE-refined SVM algorithm, providing a perfect classification accuracy using only 10 ranked genes. On average, the ALOGA<128 filtering method appeared to be the optimal (average accuracy = 1, average number of ranked genes used = 18) and both LDA and SVM seemed to be good classification algorithm (average accuracy = 0.97, average number of ranked genes used = 14–16). Again, we thought SVM should be considered superior over LDA due to the error peak observed under LDA as indicated in Fig 5.
Finally, similar to what was reported in Figs 3–5, the binary classification of the healthy control vs. the combination of HPAH and IPAH seemed to be the easiest task, providing a very high accuracy (overall average = 0.99) using a small number of ranked genes (overall average = 49). The multiclass classification (Control vs. IPAH vs. HPAH vs. UMC) was the most challenging task with an overall accuracy of 0.81 using 247 genes on average. Across all classification goals/tasks, the ALOGA<128 filtering method consistently provided the highest accuracy with the least number of genes utilized.
To further analyze the refined list of highly informative genes (Table 1) used for constructing the optimal classification models, and to evaluate the possibilities of distinguishing patient groups without relying on advanced machine learning algorithms for comparison purposes; we also performed conventional unsupervised hierarchical clustering that is commonly used in microarray data analysis. Using this method, genes that share similar expression patterns are more likely to be clustered together. Particularly, we selected the identical groups of genes used for constructing the optimal machine learning classification model under each task (Table 1). For the first classification task (Control vs. IPAH vs. HPAH vs. UMC), SVM-based algorithm achieved an accuracy value of 0.97 using only 31 filtered (ALOGA<128) and SIRRFE-refined genes (Table 1). The LDA-based model provided slightly better accuracy (0.98) under the same filtering method, but used substantially larger number of genes (199), thus, was not considered optimal. Instead of using SVM-based algorithms, unsupervised hierarchical clustering identified five groups of patients based on the expression pattern of these 31 genes. As indicated in Fig 6A, Group I patients primarily belonged to the Control group. Group II, III, and V consisted mainly of patients from the IPAH, UMC, and HPAH groups, respectively. Group IV had a mixture both IPAH and HPAH patient groups. From a genetic expression perspective, it appeared that certain protein coding genes such as the KIAA1217, TNFRSF25, ADCY5, NFIA (FLJ39164), and LHFP; as well as some non-protein coding genes, such as the LINC01181 (FLJ10489) typically had relatively higher expression values in the control patient groups than others. Genes such as the CARD19 (c9orf89), PIK3C3, LINC00308 (C21orf74), SLC7A5P1 (LAT1-3TM), BDNF, NECAB1 (EFCBP1), ZNF221 usually had higher expression values in the IPAH patient group. The LINC00461 (LOC645323), DACH1, TXLNGY (CYORF15A), CADPS, SSTR2, ZNF335 genes all had relatively higher expression values in the UMC patient group. According to the KEGG pathways database [15], PIK3C3 is greatly involved in the phosphatidylinositol signaling system, SSTR2 is involved in the cAMP signaling pathway and the neuroactive ligand-receptor interaction processes, and BDNF is involved in the neurotrophin signaling pathway. For the second classification task (control vs. the combination of HPAH and IPAH), both LDA and SVM-based algorithms under the ALOGA<128 filter provided perfect accuracies using only eight SIRRFE-refined genes (Table 1). Using the identical list of genes and unsupervised hierarchical clustering method, three major patient clusters could be identified as Group I dominated by Control-group patients; Group II, predominately IPAH patients; and Group III, mainly HPAH patients as indicated in Fig 6B. Additionally, each individual gene behaved differently among different patient groups. Both LHFP and FTCD genes seemed to have greater expression values in the Control patient group than the other two. For the last classification task (HPAH vs. UMC), SVM-based model achieved perfect accuracy using 10 genes filtered by the ALOGA<128 method (Table 1). Unsupervised hierarchical clustering method identified two patient clusters using the same 10 genes, including Group I consisting of mainly HPAH patients, and Group II mainly UMC patients as indicated in Fig 6C. Both PLLP and SGCA genes had greater expression values in HPAH patient groups than UMC. LEF1 genes indicated drastically increased expression levels in certain UMC patients than HPAH. Other genes lacked consistency in expression values between patient groups, such as CYP21A2 and SYNE2. To summarize, all unsupervised hierarchical clustering models yielded very limited performance compared to machine learning-based models, which typically offered near-perfect/perfect accuracies.
Finally, Fig 7 summarized the average gene expression values and their corresponding standard errors within each patient group under the optimal feature filtering method (ALOGA < 128) as indicated in Table 1. The most remarkable finding was that out of the total 46 genes used by the optimal models across all three classification tasks, only four genes had at least one group average expression value larger than 128 (SSTR2, MEST, LEF1, and FAM38B). In other words, more than 91% of the highly informative genes used for constructing the optimal models had all group averages less than 128. Additionally, each classification routine tended to generate a separate list of features. Only three genes were shared between tasks 1 and 2, including LHFP, MRNA, and MEST. Last but not the least, 27 out of the total 46 genes (60%) had probe intensity values smaller than 12, a threshold identified by the IQR filtering algorithm.
Original array data are available at S1 Appendix and S2 Appendix. IQR-based threshold detection report, codes and detailed modeling results files are made openly available at S3 Appendix. Detailed clustering of patient expression profiles (Fig 6), including each patient’s PAH status and gene probe identifications information, is presented in S1 Fig.
In this study, we designed and implemented a novel and efficient data analysis paradigm using high-quality clinical data obtained from cultured, unaffected tissues that are free of contamination from drug effects with well-controlled confounding influence from other pathogenesis/molecular factors. The reason for using cultured blood samples as a proxy for evaluating PAH etiology is three-folds. First, the genetic expressions of fresh blood cells are largely impacted by drug effects, the inflammatory state in the lungs of PAH patients, and by other factors induced by individual variations. These factors could entirely override the impacts of baseline genetic differences (particularly on those low-expression genes), and eventually make them invisible. On the other hand, variations in gene expression in these cultured peripheral blood mononuclear cells are free of contamination and also reflect the key underlying differences in gene expression, particularly related to PAH, which is true across tissue types in while-body metabolism and has been verified by our group in several previous studies [16,17]. Second, bone-marrow derived cells are found to be more essential to many disease processes. Particularly, transplantation of patient-derived bone marrow has been enough to induce PAH in mice [18], and transplantation of mutant mice with wild-type bone marrow ameliorates it [19] in more than one model [20]. Therefore, these cultured blood cells actually have important etiological roles in identifying PAH.
To directly test the hypothesis that low-expression genes might be greatly informative in differentiating different forms of PAH patient groups, we proposed a series of feature filtering methods including a traditional IQR-based filtering method and others that are rarely used in conventional microarray-based clinical studies. As a matter of fact, information relating to adopting similar approaches in other disease studies of human being or other organisms is extremely limited, even non-exist. Again, this is because the majority of these low-expression genes are typically removed manually by the researchers or by certain “denoising” algorithms or feature filtering functions offered by many software packages (e.g. IQR) as part of the standard data analysis pipeline. Furthermore, in this study, we also formalized a novel feature ranking/selection algorithm (SIRRFE), which provided excellent modeling performance when incorporated with state-of-art machine learning algorithms. Several findings could provide important insights into designing innovative algorithmic approaches for analyzing microarray data and enhance our understanding on the molecular pathogenesis of PAH and its interrelationship with BMPR2 mutation.
Some highlights from this research finding are described in the following. First, as indicated in Fig 2, our most efficient gene filtering method (ALOGA < 128) only resulted in a small reduction in the number of probe sets selected (26% less). However, the performance boost was quite remarkable as presented in Figs 3–5. This was consistently observed across all filtering methods as indicated in Figs 3–5, but the most obvious effect was found in Fig 5. This indicated that there exist a decent portion of genes that have high expression values across all patients but are indeed less informative in identifying different PAH patient groups. Likewise, genes with at least one PAH patient group average less than 128 (or 256), including those with all group averages less than 128 (or 256), could actually serve as important signature features for constructing accurate and robust classification models. Removing these genes could cause significant losses in feature number (Fig 2) and increased error rates as observed under AGA>128 and AGA>256 (Figs 3–5). Generally speaking, the more low-expression genes (<12 or 128) removed from the dataset, the less accurate the models became. This was consistently observed in Figs 3–5 as indicated by the increased error rates from All genes, All>12, AGA>128 to AGA>256. This, again, indicated that low-expression genes are indeed important. Particularly, even the conventional IQR-based low-expression gene filtering could result in elimination of highly informative low-expression genes. Therefore, extra caution should be used while adopting various feature filtering methods/algorithms. To date, the main-trend microarray-based data analysis still focuses on the post-filtering stage (e.g. t-test, correction methods, p-value, etc.), but our study highlighted that the importance of pre-analysis filtering methods/algorithms should never be overlooked. Second, our SIRFE-based feature refining algorithm resulted in a significant reduction in error rate across several filtering methods and classification algorithms. Particularly, it became extremely effective once incorporated with more complex non-parametric algorithms, such as the ANN and SVM. The general superior performance of SVM than ANN could be caused by the incorporation of kernel methods and the uncertainties associated with the optimal ANN structure setting (e.g. number of hidden layers, number of neurons, types of excitation functions, etc.). This finding could potentially lay the essential groundwork for future work in the bioinformatics domain, where information abounds with using state-of-the-art algorithms such as ANN, SVM, deep learning, etc.; but efficient feature ranking/selection algorithms is always lacking. Third, the LDA-based algorithms always resulted in an error peak around the top 50 to 100 ranking feature region. We attributed this primarily to the functionality limitation of LDA itself because no other classification methods had produced similar trends. Last but not the least, all SIRRFE refined algorithms had produced very impressive optimal accuracy values (overall average 0.93) across various combinations of gene filtering methods and classification algorithms as indicated in Table 1. However, the number of genes used for achieving these optimal accuracy values varied greatly. This finding again accented the effectiveness of SIRRFE. More importantly, the ability for generating highly accurate classification/prediction models with only 8 to 31 features based on cross validated datasets could lead to the development of clinical analysis software packages for future use. We expect that similar methodology, model construction and data analysis routines could be extended to other pathological domains besides PAH.
The clustered heat maps (Fig 6A–6C) of all the three classification tasks provided an overview of the general patterns of the gene expression and sample grouping across the 86 patients using conventional unsupervised hierarchical clustering methods. It was noteworthy that the gene lists used in the clustering analysis were filtered and refined by SIRRFE, and were identical to those used in advanced machine learning models (e.g. SVM and LDA) indicated in Table 1. However, unlike SVM or ANN, which transformed original data using kernel tricks or abstract neural network structure to achieve near-perfect classification; conventional unsupervised hierarchical clustering failed to separate patient groups accurately even with filtered and SIRRFE-refined genes (Fig 6). Therefore, the chances for achieving great classification performance using unsupervised clustering methods with unfiltered/unrefined gene lists are extremely low. Additionally, it was noteworthy that none of the filtering nor SIRRFE algorithms depended on any classification model. Therefore, there were no biases towards any classification method. As indicated in this study, the overall performance of conventional heat map-based microarray data analysis was extremely limited, particularly when solving complex data analytic problems. Due to the fact that many informative genes identified in each task were low-expression genes, information relating to their functionality and molecular pathways are extremely scanty. Some highlights included KIAA1217 genes identified to be highly informative for completing the first, also the most challenging classification task. It was recently found as a novel Rearranged-during-Transfection fusion gene detected in a small portion of lung adenocarcinomas and was known as an oncogenic driver factor [21]. Another important gene identified in task 1, Dachshund 1 (DACH1) was found as an inhibitor that could prevent proliferation and invasion of lung adenocarcinoma cells as well as growth of tumor cells through repression of other factor genes [22]. For both task one and two (control vs. the combination of HPAH and IPAH), the mesoderm-specific transcript (MEST) gene, also known as the paternally expressed gene 1, was recognized as of high importance. It was found to be correlated with frequent loss of imprinting in lung adenocarcinoma cases [23,24]. For the last classification task (HPAH vs. UMC), the lymphocyte enhancement factor 1 (LEF1) gene, which plays important roles in mediating lung tumor occurrence [25,26], had relatively greater expression values than others. Additionally, the 21-hydroxylase instruction gene (CYP21A2), another highly informative gene for task three, was recently found in the developing distal epithelium of the human developing lungs, potentially with its products binding to the glucocorticoid receptor and exert certain intracrine actions [27].
There are some potential limitations associated with this study. First of all, only a limited number of machine learning algorithms were used in the model construction and validation processes. We expected this shouldn’t affect the result interpretation much because those selected ones represent some of the most popular methods in data science domain. Additionally, the main goal of this study was to test the hypothesis stating that clusters of low-expression genes might collectively control PAH, thus, model construction procedure per se was not as important as obtaining the optimal models and the identification of the most informative gene list/ranking. Third, because the primary goal of this study was to identify a comprehensive ranking list of genes that could be uniformly evaluated using different gene filtering and classification algorithms, no separate initial feature ranking method was used for each data analysis routine. However, our SIRRFE-based gene refining algorithms was carried out independently within each classification method. Fourth, our optimal model analysis was based on the minimal MCCV error. We noted that the MCCV process involves randomness. Other models, especially those with gene numbers similar to the optimal models, might have MCCV errors not statistically different from the optimal models. In other words, one may derive a different optimal model by running the same program with different random seeds. However, provided the models were built and selected by using the same gene ranking list, we would expect they share sufficient overlap and the change of optimal models should not affect our primary findings much. Fifth, our approach used kernel method-based classification algorithms as a proxy for evaluating the importance of all genes used in each task. The underlying methods considered the interactions and correlations among all features, but these interactive relationships were not explicitly modeled or presented in this work, emphasizing the necessity for using interactive component modeling/selection methods in future research [28]. Sixth, adding another set of independent validation cohort could greatly strengthen the study. However, due the lack of publicly available high quality datasets, it is not possible to validate these machine learning models on other datasets at the present time. The particular dataset used in this study is unique; because it is obtained from cultured, unaffected, and uncontaminated tissues. The underlying genetically based differences in gene expression are easily overwhelmed by effects of end-stage diseases and treatments (e.g. prostacyclin) as those obtained from fresh isolated blood samples, making them unusable for this kind of tasks. Meanwhile, we also recognize that the cell immortalization processes, which should keep cross-group comparisons valid, might mask or enlarge certain differences in gene expression. However, the baseline genetic differences should be preserved and captured by our data analysis. We felt that the standard cross-validated modeling paradigm (e.g. MCCV) should provide adequate independency and control over model over-fitting. Last but not the least, our work relied on the hypothesis that gene expression differences are essentially caused by genetic functional differences, thus, the captured differences were the cause of PAH instead of the consequences.
To date, microarray analysis has become an effective and powerful tool widely used by many scientists in PAH research, providing reliable detection of genome-wide expression differences among patient groups [6]. Information abounds with use of standard microarray data analysis pipeline, which involves background noise reduction and normalization followed by gene expression analysis using fold change values or t-statistics, as well as co-expression patterns detection using different clustering/scaling algorithms (e.g. principal component analysis). Additionally, the IQR-based filtering methods are still dominating, which involves great amount of empiricism and subjectivity in the q-value threshold selection processes [13,14]. Only a few studies focused on investigating the signal threshold of gene expression when analyzing microarray data, and almost all of them endorsed the removal of low-expression data. For example, Li et al. [12] adopted novel signal threshold algorithms, which greatly reduced false positive/negative rates compared with two-fold change methods. The algorithms incorporated a two-step filtration strategy, which was very similar to our feature filtering methods but using slightly different cutoff values (100 and 200). In another study, Aris et al. [11] performed sensitivity analysis on different probe set intensity extraction methods and found that the minimum intensity cutoff thresholds were the most influential parameter affecting performance, and low intensity genes should be eliminated for achieving better sensitivity. More importantly, the study explicitly emphasized the removal of low-expression noisy data using cutoff intensity value of 100, which produced relatively stable performance with reasonable false positive rate. The most important finding of this study was that a large portion of the highly informative genes were low-expressed, regardless of cutoff threshold identification methods (IQR-based or simple probe intensity value-based approaches). Thus, if conventional IQR-based methods were adopted, 60% of highly informative genes would be eliminated before reaching the phase for identifying differentially expressed genes. To date, information related to the evaluation and investigation of low-expression genes from microarray data is still extremely limited. Oftentimes, attention was given to how different filtering algorithms/methods could be used for effectively removing low-expression genes [29] or how signal intensity of these genes be enhanced [30]. For example, a previous study indicated that microarrays with long oligonucleotide probes could greatly improve signal intensity of low-expression genes, which might greatly influence the results [31]. Little research has evaluated the importance and/or the collective effects of low-expression genes on various animal and human diseases. Particularly, systematic modeling work focusing on PAH pathobiology is almost nonexistent. We believe similar molecular pathways and/or genetic mechanisms should also exist in the etiology of other types of diseases, and explicitly modeling the genetic signaling network and interplay could provide important insights into better understanding of the underlying mechanisms of the pathogenesis of PAH. These are not the main focuses of this study, but are indeed important future directions.
In conclusion, this research indicated that low-expression genes, typically removed during the background de-noising processes when analyzing microarray data, possess important biological information for controlling PAH. Conservatively speaking, even though each individual low-expression gene might be of little importance, the clusters of many should provide significant synergetic impacts on many pathophysiological processes. Integrating the information provided by these clusters of highly informative low-expression genes with advanced machine learning algorithms and novel feature ranking/refining methods (e.g. SIRRFE) could help construct robust classification models that can accurately predict each patient’s PAH status using very few genes. Finally, we acknowledge that using other advanced sequencing method such as RNA-seq could lead to more important findings in novel pathway/biomarker identification and risk evaluation related to PAH [4,31,32], thus, should also be investigated using similar approaches as proposed in this study in the future.
Human subject research ethnic statement is included in the manuscript and approved by the Vanderbilt University School of Medicine. Study protocols were approved by the VUMC Institutional Review Board. All participants had given written consent to participate in clinical studies and underwent genetic counseling according to guidelines established by the American College of Cardiology Foundation and the American Heart Association prior to blood sampling.
All patient data were obtained from the Vanderbilt Pulmonary Hypertension Research Cohort (VPHRC), which houses more than 40 year of clinical and biologic specimens of patients, including those with IPAH and HPAH and their detailed medical history and family pedigree. Various forms of PAH mutations were detected particularly in the HPAH patient group, including frameshift, insertion/deletion, missense, and nonsense mutations. All PAH patients were diagnosed by specialist physicians at Vanderbilt University Medical Center (VUMC) or other regional hospitals. PAH was determined either by autopsy evidences indicating plexogenic pulmonary arteriopathy without alternative causes or by cardiac/clinical criteria widely accepted internationally [33]. Study subjects of VPHRC were recruited via the Vanderbilt Pulmonary Hypertension Center, the Pulmonary Hypertension Association, and the NIH Clinical Trials website (http://clinicaltrials.gov). Study protocols were approved by the VUMC Institutional Review Board. All participants had given written consent to participate in clinical studies and underwent genetic counseling according to guidelines established by the American College of Cardiology Foundation and the American Heart Association prior to blood sampling [34]. Ehylenediaminetetraacetic acid (EDTA) anticoagulated blood samples were carefully sampled at the time of hospitalization or clinical visits and then mailed via commercial blood shipping kit. Genomic DNA was isolated using the Puregene DNA Purification Kits (Gentra, Minneapolis, Minn.). Mutations of BMPR2 gene were performed by RT-PCR described previously, and the results were reported in a recent paper published by our research group [35]. We performed lymphocyte sampling/culturing from all patient groups using exactly the same protocols established in previously published article [16]. In particular, lymphocytes were isolated from anticoagulated whole blood within 48 hr of collection and were then exposed to Epstein-Barr Virus (EBV) to induce cell immortalization. Two ml blood was diluted with 2 ml Phosphate-Buffered Saline (PBS) solution, layered on top of 3 ml of Lympho Separation Medium (MP Biomedicals) and centrifuged for 10 minutes at 1,000 × G at room temperature. Using a Pasteur pipet, the lymphocytes were removed from the serum/Lympho Sep Media interface, washed in 10 ml PBS and then resuspended in 3 ml lymphoblast media (RPMI 1640 media containing L-glutamine, and 20% fetal bovine serum) containing 2 μg/ml cyclosporine. The lymphocytes were then infected with 3 ml Epstein-Barr virus (EBV) and transferred to a T-25 vent capped flask. The cells were incubated at 37°C/5% CO2 and fed weekly with lymphoblast media plus cyclosporine until signs of growth occurred. RNA was isolated from cultured lymphocytes using a Qiagen RNeasy mini kit (Valencia, Calif.). Complimentary DNA was synthesized and biotin-labeled complimentary RNA was produced by in-vitro transcription reaction. Affymetrix HGU133 Plus 2 microarrays (Affymetrix, Foster City, Calif.) were hybridized with 20 μg cRNA. Hybridization, washing, staining, and array probe scanning were carried out according to the protocol specified in the Affymetrix GeneChip Expression Analysis Manual.
The detailed patient characteristics data and microarray expression results analyzed using the Robust MultiArray Average algorithm (RMA) and R2.13/Bioconductor2.8 analysis were reported in a previous study conducted by our group [1]. In that study, we deliberately removed genes that have low-expression values according to common practices. For this study, we used the original background-corrected, normalized/summarized expression data processed by RMA function in the Affy Package. The summarized data contained 54,613 features (gene probes) for each of the 86 samples (patients), including 22 healthy control, 20 IPAH, 17 HPAH, and 27 UMC patients.
One specific goal for the current research was to construct classification tools/models that could provide the maximum accuracy while predicting patients’ PAH status using the least amount of genes. Considering the fact that the complexity of the nature for the proposed tasks could greatly affect the performance, we established classification targets using the following grouping regime: 1) multiclass classification with each PAH group forming its own target group, 2) binary classification including the healthy control vs. the combination of HPAH and IPAH, 3) binary classification including HPAH vs. UMC; with the anticipated computational difficulty levels arranged in a descending order.
Again, we expected that some low-expression genes might collectively affect certain PAH pathogenesis pathways and were usually removed manually by the researchers or by the preprocessing algorithms (e.g. IQR) embedded in microarray software packages. In light of this, we used eight different feature filtering methods based on the RMA-normalized microarray data, including: using all genes; or genes with average expression values larger than 12 (All>12), a value detected by the IQR threshold detection method [13,14]; or using genes with all PAH group average expression values larger than 128 (AGA>128) or 256 (AGA>256); or genes with at least one PAH group average expression value larger than 128 (ALOGA>128) or 256 (ALOGA>256); or genes with at least one group average expression value smaller than 128 (ALOGA<128) or 256 (ALOGA<256). The rationale for this feature filtering regime is that: method 1 could preserve all expression data including those “noises”; method 2 serves as a comparison method commonly used by researchers, which removes low-expression genes based on IQR expression profile values; methods 3 and 4 closely imitated popular manual preprocessing algorithms, which treat low-expression genes as undependable/unreliable features; methods 5 and 6 would remove consistently low-expressed genes across all patient groups; and method 7 and 8 should capture the clusters of low-expression genes that might be greatly informative at distinguishing different PAH groups when functioning collectively.
For each classification task and preliminary filtering method combination (3 groupings × 8 filtering methods = 24 combinations) as indicated in the previous session, a standard workflow of feature ranking, selection, model construction and validation was carried out independently.
Firstly, we performed the one-way Analysis of Variance (ANOVA) for each gene and ranked the genes according to their p-values. A small p-value indicated that the mean gene expression values were significantly different between two groups, so that the gene could potentially serve as a good biomarker for distinguishing one group from the other. The smaller the p-value was, the more significant the differences were. Particularly, for each classification task, a different set of p-values were generated accordingly (four, two, and two groups for task one, two, and three; respectively). Note that one-way ANOVA was somewhat analogous to the correlation methods in two groups setting. It is simple and effective methods widely used for feature ranking. However, its major drawbacks is that it fails to generate compact gene sets and it may miss some complementary or highly correlated genes that have little contribution when functioning alone [36]. Thus, in our study, we retained the top 1000 genes based on the p-value (p-values < 0.001), which conserved the majority of highly informative genes for each classification task and effectively controlled the computational complexity before entering the feature refining step.
Secondly, we proposed a novel Sliced Inverse Regression-based Recursive Feature Elimination algorithm (SIRRFE) in this study to further refine the feature ranking list. Sliced inverse regression (SIR) is a popular dimension reduction method in multivariate statistics and has been used in many fields of science [37, 38]. It can be used in both regression and classification problems. For classification problem, it typically reduces the dimension of the data with k target groups (classes) down to the most k-1 relevant dimensions by solving a generalized eigen-decomposition problem:
Γβ=λΣβ,
Where Γ is the between group covariance matrix and Σ is the covariance matrix of the whole data. The eigenvectors β associated to large eigenvalues are called effective dimension reduction (EDR) directions and could be used for transforming the original high dimension data to low dimensional data. Each reduced dimension becomes a linear combination of the original features. In this paper, we proposed an innovative recursive feature refining algorithm based on SIR, which used the magnitude of the coefficient associated each gene to rank its importance. Particularly, since our dataset has a very high dimensionality (54,613 in the case of using all genes or 1000 after preliminary selection after ANOVA) but limited sample size (86 patients), using SIR as a direct feature ranking method might greatly compromise the accuracy of the feature ranking; which is a classic problem called “curse of dimensionality,” well known in the machine learning community. However, the least informative feature could always be considered as less influential/informative and a recursive feature elimination (RFE) procedure could well overcome this problem [36,39,40]. Therefore, we proposed the following SIRRFE Algorithm:
Inputs:
Training samples: X0 = [xij]i = 1:N,j = 1:G with N being the number of patients and G0 the number of genes, class labels: y = [y1,y2,…yN]T, k = number of groups, feature ranking list r = [],
Initialize: S = 1: G0, X = X0,G = G0
Updates:
• Run SIR algorithm to obtain the EDR directions associated to the largest l = min(G,k−1) eigenvalues β1 = [β11,β12,…,β1G],β2 = [β21,β22,…,β2G],…, βl = [βl1,βl2,…,βl,G]
• Find g*=argmin1≤g≤Gβ1,g2+β2,g2+⋯+βlg2
• r = [S(g*),r]
• S = S\S(g*)
• X = [xij]i = 1:N,j∈S
• G = G−1
Repeated until: S = [] (or equivalently G = 0)
Outputs:
Updated feature ranking list r
Finally, we tested the diagnosis power of the genes that were refined by SIRRFE or not using three popular machine learning algorithms (LDA, SVM, and ANN) that have been widely used in a broad array of problem domains [41–46]. Specifically, we used the Statistics and Machine Learning Toolbox of Matlab Programming Language (The MathWorks, Inc., Natick, Massachusetts) for implementation. Additionally, we used 0.0001 to slightly regularize the covariance matrix in LDA to guarantee its invertibility when it is singular. For SVM, we adopted the one-versus-one coding design and applied linear kernel SVM for each binary classification problem. For ANN, we used five hidden layers in setting the neural network structure. Monte-Carlo cross validation (MCCV) technique was used in the model training and validating processes. Specifically, the entire dataset was randomly divided into two sections, with 80% of the samples retained for training the remaining 20% for testing. The classification model was then built using the training set and the testing set was used for prediction and performance evaluation. This process was repeated 100 times and the average values of performance indices were reported. For performance evaluation, we only evaluated the average accuracy defined as the total number of correctly predicted patients divided by the total number of patients. We ignored other evaluation metrics (e.g. precision, recall, area under the ROC curve, etc.) because the structure of the dataset was generally balanced, meaning the number of patients from each group was relatively similar to each other. The second task, which combines two PAH groups (HPAH plus IPAH) into one target group, provided a slightly unbalanced structure; but should not be considered as highly skewed.
In summary, for each target grouping and filtering method combination, we performed ANOVA-based gene ranking/selection, SIRRFE-based refinement, and three different machine learning algorithm-based training and prediction routines; namely, ANOVA+SIRRFE+LDA, ANOVA+SIRRFE+SVM, and ANOVA+SIRRFE+ANN. Additionally, we also run the experiments without using SIRRFE (ANOVA+LDA, ANOVA+SVM, and ANOVA+ANN) to verify its effectiveness.
|
10.1371/journal.ppat.1003528 | Distinct Binding and Immunogenic Properties of the Gonococcal Homologue of Meningococcal Factor H Binding Protein | Neisseria meningitidis is a leading cause of sepsis and meningitis. The bacterium recruits factor H (fH), a negative regulator of the complement system, to its surface via fH binding protein (fHbp), providing a mechanism to avoid complement-mediated killing. fHbp is an important antigen that elicits protective immunity against the meningococcus and has been divided into three different variant groups, V1, V2 and V3, or families A and B. However, immunisation with fHbp V1 does not result in cross-protection against V2 and V3 and vice versa. Furthermore, high affinity binding of fH could impair immune responses against fHbp. Here, we investigate a homologue of fHbp in Neisseria gonorrhoeae, designated as Gonococcal homologue of fHbp (Ghfp) which we show is a promising vaccine candidate for N. meningitidis. We demonstrate that Gfhp is not expressed on the surface of the gonococcus and, despite its high level of identity with fHbp, does not bind fH. Substitution of only two amino acids in Ghfp is sufficient to confer fH binding, while the corresponding residues in V3 fHbp are essential for high affinity fH binding. Furthermore, immune responses against Ghfp recognise V1, V2 and V3 fHbps expressed by a range of clinical isolates, and have serum bactericidal activity against N. meningitidis expressing fHbps from all variant groups.
| Neisseria meningitidis is a major cause of sepsis and meningitis in young children and adolescents. Although vaccines are currently available against several serogroups, a broadly effective vaccine against serogroup B is still needed. Factor H binding protein (fHbp) can bind the human complement regulator factor H (fH) and is an important meningococcal immunogen. fHbp is divided into three variant groups (V1, V2 and V3) and immunisation with V1 fHbp does not elicit cross-protection against meningococcus expressing fHbp V2 or V3, and vice versa. Here, we investigate a homologue of fHbp in Neisseria gonorrhoeae which we named Gonococcal homologue of factor H binding protein (Ghfp). We show that in contrast to fHbp, Ghfp is not expressed on the bacterial surface and is unable to bind to factor H. Surprisingly, we found that antibodies raised against Ghfp have the capacity to mediate protective immunity against N. meningitidis expressing any of the three variant groups of fHbp, and could provide a broadly protective vaccine against N. meningitidis.
| The Gram negative bacterium Neisseria meningitidis is part of the normal human nasopharyngeal flora in up to 40% of healthy individuals [1], [2] and a leading cause of sepsis and meningitis worldwide, with a case fatality rate from septicaemia of approximately 10% [3], [4]. Because of the non-specific early symptoms and rapid progression of meningococcal disease, there is an urgent need to develop vaccines to protect individuals from this important infection [4], [5]. N. meningitidis is classified into 12 different serogroups based on its polysaccharide capsule, although only six serogroups are responsible for the majority of disease. Currently there are vaccines based on the polysaccharide capsule of four of these serogroups (i.e. A, C, W, and Y) [5]. However, the capsule of serogroup B N. meningitidis (MenB) is structurally identical to a modification of a cell adhesion molecule present in the foetal brain, and is thus weakly immunogenic and could induce autoimmunity if used as a vaccine [6]. Vaccines based on outer membrane vesicles have proven to be effective against MenB but only in combating epidemic disease caused by a single clone [7]; the most effective approach to produce a broadly protective vaccine against all N. meningitidis serogroups (including MenB) will be the use of protein based vaccines [8].
Factor H binding protein (fHbp) of N. meningitidis is an important component of MenB vaccines currently under advanced clinical development [8], [9]. Immunisation with fHbp elicits serum bactericidal antibodies [8], [9], a marker of protection, and the protein provides an important mechanism for immune evasion for the meningococcus by recruiting the negative complement regulator, factor H (fH), thereby protecting N. meningitidis against complement-mediated lysis [10], [11]. fHbp is a surface expressed lipoprotein consisting of two β barrels [12], [13]. Based on sequence alignments, fHbp has been categorised into three different variant groups, V1, V2 and V3, or two families, A and B [8], [9]. However, immunisation with V1 fHbp (family B) does not elicit bactericidal responses against V2 and V3 (family A) fHbp-expressing strains and vice versa [8], [9], [14]. In addition, immunisation with one V1 peptide does not provide cross-protection against all strains expressing V1 fHbps [14]. This suggests that a broadly protective vaccine should include multiple fHbps, or fHbps which elicit cross-protection. Current vaccines contain V1.1 fHbp together with other antigens, or a combination of V1 and V3 fHbps [8], [9].
Although Neisseria gonorrhoeae binds fH to its surface, the receptor on the bacterium is Por1A which is not related to fHbp [15]. However, inspection of gonococcal genome reveals a homologue of fhbp (annotated as ngo0033 in N. gonorrhoeae strain FA1090); we designated the predicted protein Gonococcal homologue of fHbp (Ghfp), because it is approximately 90% identical to V3 fHbps. In contrast to meningococcal fHbp, Ghfp is highly conserved with three alleles described which only differ by one or two amino acids [16]. Furthermore, Ghfp is not predicted to contain a signal sequence or a lipid modification motif (LXXC) suggesting it is unlikely to be expressed on the bacterial surface [16], [17]. Here, we investigate the location, the fH binding capacity and the vaccine potential of Ghfp.
Analysis of the genome sequence of N. gonorrhoeae strain FA1090 identified the presence of a fhbp homologue [16], [17] which we designated Gonococcal homologue of fHbp , Ghfp. Sequence alignment of Ghfp with available fHbp sequences (www.neisseria.org) reveals that Ghfp has between 60–67%, 81–89% and 86–94% amino acid identity with V1, V2 and V3 fHbps, respectively (Supplementary Figure S1). To investigate the cellular location of Ghfp, sera were raised against recombinant Ghfp from N. gonorrhoeae strain FA1090. By Western blot analysis, sera recognised a protein with an estimated molecular weight of 30 kDa (corresponding to Ghfp) in lysates of N. gonorrhoeae strains FA1090 and F62; no protein was detected in lysates from F62Δghfp (Figure 1A). Sera raised against Ghfp also recognise V3.28 fHbp expressed by N. meningitidis strain M1239 (Figure 1A). Moreover, Ghfp was expressed by 20 clinical N. gonorrhoeae strains isolated in the UK (Figure 1B and not shown). To determine whether Ghfp is surface located, we performed flow cytometry analysis with anti-Ghfp serum to detect Ghfp on the surface of N. gonorrhoeae F62 and F62Δghfp, and fHbp on N. meningitidis M1239 and M1239Δfhbp (Figure 1C and 1D). Results demonstrate that anti-Ghfp serum recognises V3.28 fHbp on the surface of N. meningitidis, but there was no detectable Ghfp on the gonococcal surface.
To exclude the possibility that the lack of detection of Ghfp by flow cytometry was due to low expression levels, we also exposed viable bacteria to proteinase K, and monitored the degradation of Ghfp, a surface protein i.e. the α-2,3-sialyltransferase, Lst [18], and the cytoplasmic protein RecA by Western blot analysis. Ghfp and RecA were unaffected by exposing cells to proteinase K. The relative amounts of full length protein after incubation with reducing concentrations of proteinase K (serial three-fold dilutions from 3 ng/ml, Figure 1E) were: Ghfp, 94-90-89-100; RecA, 101-98-101-100. In contrast, digestion of Lst was observed : the amount of Lst peptides with reducing concentrations of proteinase K were 214-158-128-100 (Figure 1E). Recombinant Ghfp is cleaved by these concentrations of proteinase K (data not shown) demonstrating that Ghfp is susceptible to cleavage by this protease. In conclusion, our results demonstrate that Ghfp is expressed by N. gonorrhoeae but is not located on the bacterial surface, in keeping with previous predictions.
Due to its high sequence identity with V3 fHbp, which binds fH with a KD in the nM range [13], [14], fH binding to Ghfp was tested by far Western analysis. Surprisingly, there was no detectable fH binding to Ghfp using normal human serum as the source of fH (Figure 2A). Therefore, we compared the sequence of Ghfp with V2 and V3 fHbps that bind fH at high affinity, and identified five amino acids that are consistently different between Ghfp and V2/V3 fHbps i.e. R176, D199, D212, R288 and D318 of Ghfp (amino acid numbering according to fHbp V1.1 structure [12], Supplementary data Figure S2). Recent studies have shown that Ghfp is highly conserved [16] which we confirmed by sequencing Ghfp in a panel of 20 clinical isolates from the UK (not shown). All three Ghfp polymorphisms in our isolates had been identified previously [16] and are also present in V2 and/or V3 fHbp, so do not include residues (R176, D199, D212, R288 and D318) that are unique to Ghfp. The amino acids R176 and D199 are located in the predicted N-terminal barrel of Ghfp and, similar to C-terminal β barrel residue D212, are not located at the region of Ghfp corresponding to the interface of fHbp with fH [13]. In contrast, R288 is located in close proximity to the predicted fH:Ghfp interface, while D318 could be involved in interactions between the two predicted β barrels of Ghfp. To determine whether these five amino acid changes are responsible for the reduced fH binding to Ghfp, we modified these specific amino acids into the equivalent residues in the closely related V3.45 fHbp. Modification of all five residues in GhfpM1–5 (i.e. R176Q, D199G, D212S, R288H and D318G) was sufficient to enable Ghfp to bind fH by far Western analysis (Figure 2B). Analysis of GhfpM1 (R176Q), GhfpM1–2 (R176Q and D199G), GhfpM1–3 (R176Q, D199G and D212S) and GhfpM1–4 (R176Q, D199G, D212S and R288H) demonstrated that these modifications did not restore fH binding. However the substitutions R288H and D318G in GhfpM4–5 are sufficient to confer fH binding to Ghfp by far Western analysis (Figure 2C). To further analyse this interaction in more detail, the binding of Ghfp and V3.45 fHbp to fH6–7 was also investigated by Surface Plasmon Resonance (SPR, Figures 2D and 2E). The dissociation constant of V3.45 fHbp and complement control protein (CCP) domains 6 and 7 of fH (fH6–7) was 1±4 nM, similar to previous results for V3 fHbps [13], [14]. Consistent with our far Western analysis, no fH binding was detected to Ghfp by SPR under these conditions. Moreover, no fH binding was observed to GhfpM5 (D318G). There was fH binding detected to GhfpM4 (R288H) (KD of 16±0.3 nM) while the double substitution, GhfpM4–5 (R288H and D318G), resulted in fH binding that was equivalent to fHbp (KD i.e. of 2 nM).
To exclude the possibility that Ghfp interacts with fH via CCP domains other than fH6–7, we also examined fH binding by ELISA in which recombinant proteins were coated on the wells of plates and binding to purified full length fH was detected. Consistent with SPR, we observed fH binding to GhfpM4–5, partial fH binding to GhfpM4, and no fH binding to wild-type Ghfp or GhfpM5 (Figures 2E). In conclusion, despite its high amino acid identity with V3 fHbp, Ghfp does not bind fH to any significant degree, and there are only two amino acids responsible for the striking difference in affinity compared with fHbp. Due to its high sequence identity with V3 fHbp, we were able to map the Ghfp sequence on our V3 fHbp structure [13]. Figure 2G shows the location of these two important amino acids, R288 (altered in M4) and D318 (M5); while R288 lies on the face of Ghfp which interacts with fH in fHbps, D318 may influence the interaction between the two β barrels of the protein.
As the modifications R288H and D318G in Ghfp are sufficient to confer high affinity fH binding, we next investigated whether the corresponding residues in V3.45 fHbp are necessary for binding to fH. We generated V3.45 fHbp with H288R and G318D (fHbpM4–5); these modifications were not included in our recent analysis of fH:V3 fHbp interactions which involved alanine substitution of fHbp [13]. Initially binding of fH was examined by far Western analysis (Figure 3A) and showed loss of detectable fH binding to fHbpM4 (H288R), fHbpM5 (G318D) or fHbpM4–5 (H288R and G318D). To verify these results, binding of fH6–7 to fHbp wild type and modified proteins was analysed by SPR (Figure 3B). No detectable binding of fH was observed to fHbpM4, fHbpM5 and fHbpM4–5 under these conditions, demonstrating that both of these residues are necessary for high affinity interactions with fH. To exclude the possibility that these modified fHbp molecules interact with fH via CCP domains other than fH6–7, we examined binding to purified full length fH by ELISA. Consistent with SPR, we observed fH binding to fHbp V3.45 but no binding to fHbpM4, fHbpM5 and fHbpM4–5 (Figure 3C). Taken together, we conclude that the amino acids present at positions 288 and 318 are the basis for the profound difference in interactions with fH observed in the closely related proteins from the gonococcus and meningococcus.
Next we investigated the vaccine potential of Ghfp by examining the ability of sera raised against this protein to recognise fHbps expressed by a range of N. meningitidis isolates. Immune sera not only recognise closely related V3 fHbps expressed in whole cell extracts of N. meningitidis but also V1 and V2 proteins (Figure 4A). However, V2.23 fHbp expressed by N. meningitidis strain 5/99 was not detected by anti-Ghfp serum or by sera raised against V2 fHbp [19], suggesting that this strain expresses little or no fHbp. Therefore, we examined whether anti-Ghfp serum recognises equivalent amounts of different recombinant fHbps by ELISA. Surprisingly, immune sera raised against Ghfp detected all V1, V2 and V3 fHbps examined (Figure 4B and C). Serum bactericidal activity (SBA) is an established correlate of protective immunity against serogroup C meningococcal infection [8]. To determine whether immunisation with Ghfp elicits functional immune responses, we evaluated the SBA of anti-Ghfp serum against several N. meningitidis strains expressing V1, V2 and V3 fHbp (Figure 4D). We found SBA against strains expressing V1, V2 and V3 fHbps.
SBA can be influenced by expression levels of fHbp or inherent serum resistance of the bacteria due to capsule expression [20]. To study the observed cross-protection of anti-Ghfp serum independent of these factors, we constructed isogenic strains of N. meningitidis MC58 each expressing one of the seven most prevalent fHbps (i.e. V1.1, V1.4, V1.13, V2.16, V2.19, V3.45 and V3.47) from disease isolates in England and Wales, accounting for 70% of cases [19]. To this end, we inactivated the wild-type copy of fHbp and introduced a single copy of the gene encoding each of the selected variants at an ectopic site under the control of an IPTG inducible promoter. Expression of the different fHbps was confirmed by Western blot analysis of whole cell extracts (Figure 5A) and surface expression verified by flow cytometry (Figure 5B and C), showing higher expression levels compared to wild type fHbp expression. We determined the SBA of anti-Ghfp serum against the isogenic N. meningitidis strains and compared the findings with sera raised against V1.1 or V3.45 fHbp (Figure 5D). We found that anti-Ghfp serum exhibited SBA against N. meningitidis expressing V1.1, V1.4, V2.16, V2.19, V3.45 and V3.47. In contrast, anti-fHbp V1.1 serum only elicited SBA responses against V1.1 and V1.4 expressing strains, while anti-V3.45 fHbp serum had SBA against all V2 as well as the V3.45 expressing strains (i.e. strains expressing family A proteins). No detectable SBA was measured with any sera against the isogenic MC58Δfhbp strain or using sera from mice receiving adjuvant alone (data not shown). To examine this cross-protection in another genetic background, we constructed isogenic strains of N. meningitidis H44/76 in a similar way to express V1.1, V2.16 or V3.47, and observed similar cross-protective SBA responses (Figure 5E). In conclusion, Ghfp can elicit SBA against V1, V2 and V3 fHbp expressing N. meningitidis and is therefore a naturally occurring protein capable of providing cross-protection.
N. meningitidis and N. gonorrhoeae are two human specific, closely related pathogens that inhabit distinct niches in the body. N. gonorrhoeae causes sexually transmitted infections predominantly affecting the mucous membranes of the genito-urinary tract, while N. meningitidis colonises the nasopharynx [21]. Despite sharing many similarities of the genetic level, these bacteria employ entirely different mechanisms to evade immune responses, and in particular, to avoid complement activation on their surface [22]. For example, disease isolates of N. meningitidis express a polysaccharide capsule which is essential for high-level serum resistance [23], while N. gonorrhoeae is not encapsulated. Instead sialylation of lipopolysaccharide markedly promotes complement resistance in the gonococcus [24] but this has less impact on N. meningitidis [25].
Both organisms have evolved to bind fH to their surface to prevent complement activation (by down-regulating the alternative pathway) but use different strategies. The gonococcus recruits fH via an exposed surface loop of Por1A (loop 5), an outer membrane porin often expressed by isolates recovered from patients [15]. fH can also bind to gonococci expressing Por1B albeit to a lesser degree, with this interaction facilitated by lipopolysaccharide sialylation [26]. Although meningococci express class 3 and class 2 porins (which are related to Por1A and Por1B of N. gonorrhoeae, respectively), these are not involved in fH binding; loop 5 of the meningococcal porins lacks a region present in gonococcal Por1A, which probably accounts for its inability to bind fH [27]. Instead, the surface expressed lipoprotein fHbp mediates high affinity binding of fH by the meningococcus irrespective of variant group [13]. This interaction enhances bacterial survival in whole blood and prevents serum dependent killing [10], [11]. It is not clear why the organisms have adopted alternative approaches to exploit the same human molecule, but it is likely to be influenced by the affinity of the interaction, the local availability of fH and the density of the bacterial receptor, as well as other factors conferring complement resistance. Without capsules, gonococci are largely reliant on their capacity to recruit fH and C4bp to survive in the human host [28], [29]. Therefore the relatively low levels of fH in the genito-urinary tract may have favoured its recruitment by a highly abundant protein on the gonococcal surface, such as porin.
Here, we show that Ghfp, the gonococcal homologue of the meningococcal fH receptor, does not bind fH to any detectable extent despite its high sequence identity with fHbp. Remarkably, only two amino acids in Ghfp (R288 and D318) that differ from those in fHbp are responsible for this lack of interaction. Furthermore, the replacement of the equivalent amino acids in V3.45 fHbp (i.e. H288R and G318D) resulted in loss of fH binding. The H288R modification is located at the fH:fHbp interface; the side chain of fHbp H288 sits in a hydrophobic pocket in fH formed by H337, Y353 and the methylene groups of the R341. The extended side chain of Ghfp R288 is too long to fit into this pocket without remodelling the interface, and would also result in electrostatic repulsion with R341 of fH. The lack of fH binding to V3.45 fHbpM5 (i.e. G318D) is more difficult to explain as it is located away from the fH:fHbp interface, and is at the end of the final strand of the second β barrel. However, the register of this strand is such that the side chain of residue 318 points into the hydrophobic core of the barrel. Substitution of Gly with Asp is not possible without structural rearrangement due to steric clashes in the hydrophobic core as it is energetically unfavourable to place a negative charge in the hydrophobic environment. Given this final strand also makes crucial contacts with the first β barrel, this substitution could lead to structural rearrangements at interface between the two barrels and therefore alter the distal fH binding site (which comprises both barrels). Recently, we identified several residues in V1, V2 and V3 fHbps which are needed for high affinity interactions with fH through alanine scanning mutagenesis [13]. Here we found two further mutations that abolish fH:fHbp binding by analysing the binding characteristics of a natural protein.
While fHbp is located on the surface of the meningococcus, we demonstrate that Ghfp is not on the external surface of the gonococcus, as suggested previously [17]. Examination of Ghfp also reveals the absence of a signal sequence for export so the protein is likely to remain intracellular and not secreted into the extracellular milieu. We can however not exclude the possibility that the location of Ghfp changes during infection. The function of Ghfp remains unknown, although due to its high level of identity to fHbp, we cannot exclude that Ghfp promotes survival in the presence of antimicrobial peptide LL-37 [30] or has a role in siderophore binding [31]. However, these are unlikely functions for Ghfp given the location of the protein.
fHbp is a key component of protein sub-unit meningococcal vaccines under late phase clinical development [8], [9], [32]–[34]. Unfortunately, antibody responses against fHbp are thought to be largely variant/family specific. Therefore fHbp-based vaccines consisting of a single natural fHbp might be expected to have limited coverage. To overcome this issue, vaccines under development have included fHbp together with other antigens namely GNA2132, NadA, GNA1030, GNA2091 and a membrane vesicle [35], or multiple fHbp variants [9]. Here, we show that anti-sera raised against Ghfp has the potential to recognize representative V1, V2 and V3 fHbps, in contrast to sera raised against the widely used V1.1 fHbp and V3.45 fHbp [8]. More importantly, we showed that Ghfp has the potential to elicit SBA against wild-type N. meningitidis, and two different strains expressing the most common V1, V2 and V3 fHbps. SBA of murine immune sera was assayed in the presence of rabbit complement. Although a heterologous non-human source, rabbit complement has been used to validate the immunogenicity of conjugate vaccines in pre-clinical and clinical studies [36]. Furthermore, V1 fHbps differ in their recognition by sera raised against Ghfp (Figure 4A) but this cannot be explained at the level of overall sequence identity as the proteins we examined are all approximately 60% identical to Ghfp. The advantage of using isogenic strains over naturally occurring isolates is that this approach allows analysis of the effect of diversity in protein sequence on cross-protective responses, while excluding strain-specific, confounding factors such as levels of expression and other mechanisms of immune escape [37]. This breadth of activity was an unexpected finding which has been seen with synthetic fHbp molecules containing epitopes from different fHbp variants [38], [39] and is not the result of overexpression of fHbp in our isogenic strains; anti-sera raised against V1.1 only has SBA against V1.1 and V1.4 in line with previous results [14], while our work showed that sera raised against V3.45 was able to elicit SBA against strains expressing V2.16 and V2.19, again consistent with previous work demonstrating that raised anti-V3.45 fHbp serum does have SBA against some V2 fHbp expressing strains [14].
The mechanism underlying the broad protection of Ghfp is currently unknown. Based on the position of invariant amino acids in different fHbps, the protein has also been divided into five variable segments, designated as VA-VE. Using these five segments, fHbps can be categorized into six modular groups [40]. A closer inspection of the fHbps expressed in our isogenic strains reveals that all alleles and Ghfp harbour an identical variable segment D (VD). However, this common sequence cannot be the basis of the cross-protection offered by Ghfp as fHbp V3.45 and V1.1 also harbour this region yet do not provide the same breadth of SBA. Another possibility is that the immunogenic properties of Ghfp are not solely dependent on its primary amino acid sequence but instead a result from its conformation. For instance, it is possible that due to D318 or other residues, the folding of Ghfp is altered compared with V3.45, altering accessibility of certain regions of fHbp to B cell receptors and therefore inducing cross-protective responses. We also compared Ghfp and fHbp V3.45 with the rationally designed fHbp that showed broad cross-protection due to introduction of V2 and V3 epitopes into V1.1 fHbp [39]. However all the amino acids introduced into V1.1 to obtain the cross-protection are present in both Ghfp and fHbp V3.45 and so cannot explain the cross-protection we observed with Ghfp.
In summary, Ghfp is a promising vaccine candidate against N. meningitidis since the protein not only offers a broad range of protection, but is also a naturally occurring non-fH binding molecule. There are potential drawbacks for the use of functional fHbps as a vaccine antigen due to its high affinity binding with fH. The extensive binding of fH to fHbp could shield immunogenic epitopes on the antigen resulting in less effective antibody responses [12]. Moreover, binding of fHbp to fH might reduce the immunogenicity at the site where antibody responses are initiated [41] or it could lead to formation of anti fH responses in the human host [42]. Indeed, non-functional fHbps have demonstrated non-inferior or enhanced immunogenicity compared with wild-type proteins in transgenic mice [13], [43], although any benefit of these antigens and Ghfp will need to be assessed in clinical trials.
The bacterial strains used in this work are shown in Table 1 and Table 2. N. meningitidis was grown in the presence of 5% CO2 at 37°C on Brain Heart Infusion (BHI, Oxoid, Basingstoke, United Kingdom) plates with 5% (vol./vol.) horse serum (Oxoid) or in BHI broth at 37°C. N. gonorrhoeae was grown in the presence of 5% CO2 at 37°C on GC agar (Oxoid) plates with Vitox (Oxoid) or in GC broth (15 g Protease peptone (Oxoid), 4 g K2HPO4, 1 g KH2PO4, 5 g NaCl per litre (Sigma Aldrich) with 10 ml Kellogg's supplement (40 g glucose, 0.5 g glutamine, 50 mg Fe(NO3)9H2O, 1 ml 0.2% thiamine pyrophosphate per 100 ml, Sigma Aldrich). N. gonorrhoeae strains were obtained from across the UK in 2012, and provided by the Sexually Transmitted Bacterial Reference Unit, Public Health England (kind gift of Dr. Ison and Dr. Quaye). Escherichia coli was grown on LB agar plates or LB liquid at 37°C with appropriate antibiotics.
Strain MC58Δfhbp [44] and H44/76Δfhbp (constructed as MC58Δfhbp) were complemented with pGCC4 [45] containing fhbp V1.1, 1.4, 1.13, 2.16, 2.16, 3.45 and 3.47. PCR to amplify fhbp was performed using genomic DNA from strains listed in Table 1 and using primers in Table 3. PCR products were ligated into pGEMT (Promega) then pGCC4. Transformation of N. meningitidis strain MC58ΔfHbp was performed as described previously [46]. M1239Δfhbp was constructed as MC58Δfhbp and F62Δghfp was a kind gift of Dr. M Pizza (Novartis).
N. meningitidis was grown overnight and re-suspended in phosphate buffered saline (PBS). The concentration of bacteria was determined by measuring the O.D. at 260 nm of bacterial lysates in 1% SDS/0.1 M NaOH [46] and adjusted to 109 CFU per ml. Samples were mixed with an equal volume of 2× SDS-PAGE loading buffer and boiled for 10 minutes, then run on SDS-PAGE gels and transferred to Immobilon PVDF membranes (Millipore). Membranes were blocked with 3% skimmed milk in 0.01% Tween in PBS (PBS-T) then incubated with primary (immune sera at a 1∶10000 dilution) and subsequently with secondary antibodies (goat anti-mouse conjugated HRP IgG, Dako, 1∶20000 dilution) all in PBS-T with 3% skimmed milk. fH binding to fHbp expressed by N. meningitidis or recombinant proteins was analysed by far Western blotting. Blots were incubated with normal human serum (diluted 1∶100) for 45 minutes, then incubated with anti-fH (Quidel 1∶1000 dilution), followed by rabbit anti-goat-HRP conjugated IgG (Santa Cruz 1∶20000 dilution). Binding of secondary antibodies was detected using the ECL kit (Amersham).
Genes were amplified without their signal sequence by PCR with genomic DNA using primers described in Table 3. PCR products were ligated into pGEMT then into pET28a (Invitrogen, after digestion with BamHΙ and EcoRΙ) or pET21b (Invitrogen, using HindΙΙΙ and XhoΙ, or NdeΙ and XhoΙ). Proteins were expressed in E. coli and purified using Nickel affinity chromatography followed by a HiTrapQ HP column (GE Healthcare) [13]. Mutations were introduced into ghfp by overlapping PCR and into fHbp by QuikChange Site-Directed mutagenesis (Agilent Technologies) using primers described in Table 3.
SPR was performed using a Biacore 3000 (GE Healthcare). Ghfp (50 µg/ml) was first digested with 0.5 µg/ml trypsin for 2 hours at room temperature under constant shaking (300 rpm), then 0.1 mg/ml Pefabloc SC plus (Roche) was added and incubated for 10 minutes prior to dialysis against PBS. Recombinant proteins were immobilized on a CM5 sensor chip (approximately 600–1000 RU) (GE Healthcare) and increasing concentrations of fH6–7 (0.5 nM–32 nM) were injected over the flow channels (40 µl/min). Dissociation was allowed for 300 seconds. BIAevaluation software was used to calculate the KD.
Proteins (3 µg/ml, 50 µl per well) were coated on the surface of wells (F96 maxisorp, Nunc), and after blocking with 4% BSA in PBS-T, anti-Ghfp serum was added at different dilutions and detected with goat anti-mouse HRP antibody (1∶5000 diluted) followed by substrate (Becton Dickinson). To measure fH binding to Ghfp and fHbp, proteins were coated onto wells (3 µg/ml, 50 µl per well), then incubated with fH (1 µg/ml, Sigma) and fH binding was detected using anti-fH poly clonal antibody (Quidel, 1∶1000 dilution) followed by an HRP-conjugated rabbit anti-goat IgG (Dako, 1∶5000 dilution).
Six female BALB/C mice (6–8 week old, Charles Rivers, Margate) were immunised with antigens (20 µg) absorbed to aluminium hydroxide (final concentration 3 mg/ml), 10 mM Histidine-HCL, 2 M NaCl (final concentration 9 mg/ml) in ndistilled H2O and mixed overnight at 4°C. The antigens were given via the intraperitioneal route on days 0, 21 and 35. Sera were collected on day 49 by terminal anaesthesia and cardiac puncture. All procedures were conducted in accordance with Home Office guidelines.
N. meningitidis was grown on BHI plates supplemented with 1 mM IPTG overnight and suspended in PBS supplemented with 0.1% glucose (PBS-G) to a final concentration of 5×104 CFU/ml. Bacteria were mixed with an equal volume of baby rabbit complement (Cedarlane) diluted 1∶10 in PBS-G. Heat inactivated serum, pooled from at least six mice was added to the wells. Control wells contained either no serum or no complement. Following incubation for 1 hour at 37°C in the presence of 5% CO2, 10 µl from each well was plated onto BHI plates in duplicate and the number of surviving bacteria were determined. SBA was performed with two-fold dilutions of serum starting at 1∶32. The bactericidal activity was expressed as the dilution of serum needed to kill more than 50% of bacteria in three independent experiments. Killing was calculated by comparing the number of surviving bacteria with those recovered from wells containing complement only.
N. gonorrhoeae strain F62 was grown overnight in GC liquid at 37°C then diluted 1∶20 and grown for approximately six hours until an OD A600 of approximately 0.5. An aliquot (1 ml) of the bacterial culture was centrifuged at 13,000× g then re-suspended in 300 µl of 3 ng/ml Proteinase K (Qiagen) or 3 times dilutions from this. After incubation for 30 minutes at 37°C, Pefablock SC inhibitor (Roche, final concentration 1 mM) was added for 15 minutes at room temperature. Samples were then spun and suspended in 100 µl 1× sample buffer. Digestion was assessed by Western blot analysis with antibodies against Ghfp (1∶10000 diluted), RecA (Abcam, 1∶5000 diluted), and α-Lst [18] (1∶20000) followed by goat anti-rabbit (Santa Cruz Technology, 1∶20000) or goat anti-mouse HRP conjugated IgG (Dako, 1∶20000). The relative amounts of full length protein after incubation with reducing concentrations of proteinase K was measured using AIDA software.
Bacteria (1×109) were fixed in 1 ml of 3% formaldehyde for two hours then washed with PBS. To measure fHbp expression, 5×107 bacteria were incubated with 50 µl anti-Ghfp serum (diluted 1∶500) in PBS-T for 30 minutes at 4°C with shaking, washed in PBS-T then incubated with FITC conjugated goat anti-mouse antibody (DAKO, diluted 1∶50) for 30 minutes. After washing, fHbp expression was measured by flow cytometry using the FACS calibur, calculating the mean FL1 of 10000 bacteria.
|
10.1371/journal.ppat.1006025 | Chemokine Levels in the Penile Coronal Sulcus Correlate with HIV-1 Acquisition and Are Reduced by Male Circumcision in Rakai, Uganda | Individual susceptibility to HIV is heterogeneous, but the biological mechanisms explaining differences are incompletely understood. We hypothesized that penile inflammation may increase HIV susceptibility in men by recruiting permissive CD4 T cells, and that male circumcision may decrease HIV susceptibility in part by reducing genital inflammation. We used multi-array technology to measure levels of seven cytokines in coronal sulcus (penile) swabs collected longitudinally from initially uncircumcised men enrolled in a randomized trial of circumcision in Rakai, Uganda. Coronal sulcus cytokine levels were compared between men who acquired HIV and controls who remained seronegative. Cytokines were also compared within men before and after circumcision, and correlated with CD4 T cells subsets in foreskin tissue. HIV acquisition was associated with detectable coronal sulcus Interleukin-8 (IL-8 aOR 2.26, 95%CI 1.04–6.40) and Monokine Induced by γ-interferon (MIG aOR 2.72, 95%CI 1.15–8.06) at the visit prior to seroconversion, and the odds of seroconversion increased with detection of multiple cytokines. Coronal sulcus chemokine levels were not correlated with those in the vagina of a man’s female sex partner. The detection of IL-8 in swabs was significantly reduced 6 months after circumcision (PRR 0.59, 95%CI 0.44–0.87), and continued to decline for at least two years (PRR 0.29, 95%CI 0.16–0.54). Finally, prepuce IL-8 correlated with increased HIV target cell density in foreskin tissues, including highly susceptible CD4 T cells subsets, as well as with tissue neutrophil density. Together, these data suggest that penile inflammation increases HIV susceptibility and is reduced by circumcision.
| The per-contact risk of infection with HIV through sexual exposure is low and highly variable. Understanding the biological basis for this variability could help in the development of new methods to prevent infection. There is some evidence that penile inflammation, even in the absence of any clinical symptoms, may increase HIV-susceptibility by recruiting CD4 T cells, the immune cell type that is the principal target of HIV. We analyzed soluble inflammatory mediators in prepuce swabs collected longitudinally from initially HIV-negative men enrolled in a randomized controlled trial of adult circumcision. We found that these inflammatory mediators were elevated in men who went on to acquire HIV. We also found that higher levels of these mediators were associated with an increased density of HIV-susceptible target cells in the underlying foreskin tissue and that circumcision reduced their levels, which may help to explain why circumcision reduces HIV risk by 60% or more. Together, these data suggest that penile inflammation, in the absence of genital infections, increases HIV susceptibility and is reduced by adult male circumcision.
| Two million individuals acquired HIV-1 (HIV) in 2014, contributing to the nearly 37 million living with this still incurable infection [1]. While most individuals acquired the virus through heterosexual sex [2], the per act risk of female-to-male transmission is generally low (less than 1/250 per coital act in low income countries [3]). This risk is also highly variable, and is dependent on factors in both the infected and uninfected partner [4]. Susceptibility of an uninfected male partner has been epidemiologically linked to younger age [5, 6], race [7], genital co-infections [8], and lack of male circumcision [9, 10]. However, the biological mechanisms by which these parameters alter HIV susceptibility remain incompletely understood.
Mucosal inflammation and immune activation are hypothesized to enhance HIV susceptibility. In the genital mucosa, CD4 T cells expressing CCR5 are the primary targets of HIV [11–14], with potential transport and amplification by local dendritic cell subsets [15]. Thus, if inflammation leads to the recruitment of CCR5+ CD4 T cells, it will provide additional target cells for HIV. HIV also preferentially infects and replicates in activated CD4 T cells [16–21], and so augmented immune activation may also facilitate the establishment of productive mucosal infection. Contribution of mucosal inflammation to genital HIV susceptibility is consistent with data from female rhesus macaques, where pro-inflammatory cytokines promote the recruitment and activation of CD4 T cells in the vaginal mucosa [13], and the number of CCR5+ CD4 T cells at the site of mucosal challenge dictates the likelihood of subsequent productive SIV infection [22]. Furthermore, observational studies in South African women have linked pro-inflammatory genital cytokines to HIV acquisition [23] and increased CD4 T cells in the cervical mucosa [24]. While there are no similar data from men, asymptomatic herpes simplex virus type-2 (HSV-2) infection is associated with a 3-fold increased risk of HIV acquisition in heterosexual uncircumcised men [25], perhaps due to increased CCR5+ CD4 T cells in foreskin tissue [26, 27].
Randomized clinical trials have conclusively shown that male circumcision reduces HIV susceptibility in heterosexual men [28–30], but the biological mechanisms underlying this protection remain incompletely understood. One hypothesis is that circumcision reduces genital inflammation and immune activation, either through the prevention of viral STIs [31], the reduction of inflammatory anaerobic bacteria [32], or through other mechanisms yet to be defined, which in turn this reduces the density of potential target cells for HIV. This hypothesis is supported by ex vivo experiments demonstrating that the inner aspect of the foreskin has an increased density of HIV target cells [33–35] and more efficient virus transfer from Langerhans cells to local CD4 T cells [15] than the outer aspect, which is contiguous with the shaft skin that remains after circumcision. These observations suggest that the intact foreskin constitutes an immunologically activated tissue milieu that promotes target cell recruitment and dendritic cell maturation [36–38].
We hypothesized that elevated levels of pro-inflammatory penile cytokines would be associated with HIV acquisition in uncircumcised men and with an increased density of HIV target cells in foreskin tissue, and that cytokine levels would be reduced by circumcision. To test these hypotheses, we performed a case-control study of coronal sulcus cytokines and HIV acquisition among men who participated in a randomized controlled trial (RCT) of male circumcision in Rakai, Uganda [29]. We then examined whether these inflammatory cytokines declined after circumcision in a subset of men who were enrolled in the trial but who did not acquire HIV. Finally, we used samples from a cross-sectional study of men undergoing elective circumcision [39] to assess the correlation between prepuce cytokine levels and foreskin HIV target cell density.
To assess the relationship of coronal sulcus cytokines with seroconversion, we performed a nested case-control study comparing men who acquired HIV during the Rakai RCT of circumcision (n = 60, cases) to men who remained persistently seronegative (n = 120, controls). All men in this analysis were randomized to receive delayed circumcision and remained uncircumcised throughout the trial. Participant demographics are presented in Table 1. HIV seroconversion was associated with occupation, marital status, number of sex partners, condom use, alcohol consumption, and self-reported genital STI symptoms (genital ulcer, genital warts, urethral discharge), as previously reported [40].
All cytokines examined were detected in coronal sulcus swabs, although many were detected infrequently. IL-8 was most common, detected in 60% of coronal sulcus swabs (concentration range >1.5–7405.7pg/ml in swabs suspended in 1ml transport medium), followed by MIG (range >0.3–6.9pg/ml), which was detected in 25% of swabs. Other cytokines, (GM-CSF, MCP-1, MIP3α, IL-1a and RANTES) were detected infrequently (<10% of participants, Table 2). Cytokine detection was not associated with sexual behavior or demographic factors (S1 and S2 Tables), but was associated with self-reported STI symptoms (genital ulcer, genital warts, urethral discharge).
Men who acquired HIV were more likely to have detectable levels of the chemoattractant cytokines IL-8 (aOR 2.58, 95% CI: 1.40–6.40) and MIG (aOR 3.05, 95% CI: 1.15–8.06) at the visit prior to seroconversion (Table 2). The increased odds of HIV acquisition did not change after adjusting for covariates associated with either the detection of cytokines (S1 and S2 Tables) or seroconversion (Table 1), including self-reported STI symptoms.
HIV seroconversion was not associated with the detection of other cytokines (GM-CSF, MCP-1, MIP3α, IL-1a and RANTES), but power was limited due to the low prevalence of these cytokines. However, when the total number of detectable cytokines was considered as the primary exposure, the odds of seroconversion was found to increase significantly with the presence of two or more cytokines (aOR 3.88, 95% CI 1.21–12.50; Fig 1 and Table 2).
Since coronal sulcus cytokines were associated with increased HIV susceptibility, we examined circumcised and uncircumcised men who remained persistently seronegative to determine how circumcision impacts coronal sulcus cytokines levels. Enrolment demographics of men randomized to receive either immediate circumcision (“circumcised”, n = 80) or delayed circumcision (“uncircumcised”, n = 80) were similar (S3 Table). Detectable IL-8 declined significantly after circumcision (Fig 2), even though the prevalence of detectable coronal sulcus cytokines was similar between the two groups at enrollment (S4 Table). Among men who received circumcision, the prevalence of detectable coronal sulcus IL-8 declined significantly by month 6 post-circumcision (PRR month 6 compared to enrollment was 0.59, 95% CI: 0.44–0.87; Fig 2) and continued to decline throughout the 24 month follow-up period (PRR 0.29, 95% OR 0.16–0.54); the decline between months 6 and 24 was significant (PRR 0.49, 95% CI 0.25–0.96). There were no significant changes in IL-8 detection among uncircumcised men. Even though MIG was associated with seroconversion, it did not change significantly after circumcision. Likewise, the prevalence of other coronal sulcus cytokines (MIG, MCP-1, MIP3α, IL-1a and RANTES) showed no significant change after circumcision (S4 Table).
We found that coronal sulcus IL-8 and MIG were associated with increased HIV susceptibility, and that circumcision significantly reduced IL-8. Given that IL-8 and MIG are both chemoattractant cytokines associated with recruitment of immune cells to sites of inflammation [41], we therefore examined the link between levels of prepuce cytokines and the density of pro-inflammatory and HIV-susceptible immune cell populations in foreskin tissues. We measured IL-8 and MIG levels in coronal sulcus swabs collected from 89 men who underwent elective adult circumcision at the Rakai Health Sciences Program (RHSP) Circumcision Service Program, in whom we previously characterized foreskin T cell populations [39]. Participant demographics are provided in S5 Table; no behavioral characteristics recorded correlated with levels of IL-8 or MIG. IL-8 levels were above the LLOQ in 94.4% of participants (84/89), and MIG was detectable in 51.7% (46/89). We examined the correlation of each cytokine with the density of total CD4 and CD8 T cells, and also with the following HIV target cell populations: (1) CD4 T cells expressing the HIV co-receptor CCR5 (CD3+/CD4+/CCR5+); (2) Th17 cells (CD3+/CD4+/IL-17A+); (3) Th1 cells (CD3+/CD4+/ IFNγ+); and, (4) CD4 T cells producing TNFα (CD3+/CD4+/ TNFα+). IL-8 concentration correlated with the density of both CD4 and CD8 T cells (Fig 3A and 3B; p<0.05), and with the density of CD4 T cell subsets known to be preferential HIV target cells: CCR5+ CD4 T cells, Th17 cells, Th1 cells, and TNFα+ CD4 T cells (Fig 3C, 3D, 3E and 3F; all p≤0.02). Having detectable coronal sulcus MIG was only associated with a non-significant trend of increased total CD4 (44.0 vs. 33.5 cells/mm2, p = 0.08; Fig 3G), but with a significant trend to increased CD8 T cell density (35.5 vs. 22.7 cells/mm2, p = 0.04).
To investigate the relationship of other tissue immune cell populations with prepuce cytokine levels, we next assessed neutrophil (CD15+) and dendritic cell (CD207+ Langerhans and CD11c+ dermal dendritic cell) density in a subset of men with high (n = 5; median IL-8 = 3422.6pg/ml, all MIG detectable) and low (n = 5; median IL-8 = 1.8pg/ml, all MIG undetectable) coronal sulcus cytokine levels (Fig 4A–4C). Neutrophils were found in both the epidermis and dermis, often in concentrated foci; Langerhans cells were found almost exclusively in the epidermis, and CD11c+ cells were predominantly in the dermis, as previously reported [42]. Men with high prepuce cytokine levels had a 4-fold higher density of tissue neutrophils than men with low cytokine levels (22.6 vs. 5.6 cells/mm2, p = 0.016; Fig 4D). However, densities of dendritic cell populations were similar in men with high and low cytokine levels (CD11c: 10.0 vs. 8.6 cells/mm2 dermal tissue, ns; CD207: 80.6 vs. 49.8 cells/mm2 epidermal tissue, ns).
To rule out potential confounding of coronal sulcus cytokine levels by the vaginal secretions of a female sexual partner, cytokines were also assessed in female partner vaginal swabs collected on the day of male circumcision [43] for all 89 men in this analysis. Both IL-8 (median 699.5pg/ml, range 1.5–5693.9pg/ml) and MIG (median 5.9pg/ml, range 0.3–853.4pg/ml) were detectable in vaginal swabs, but we found no correlation between vaginal and penile cytokines within couples (Spearman’s rho: IL-8 = 0.17, MIG = 0.14, both not significant), suggesting that cytokines detected in coronal sulcus swabs did not originate from the female partner.
Our study demonstrates a significant link between pro-inflammatory coronal sulcus cytokines and HIV acquisition in heterosexual men. Specifically, the chemoattractant cytokine IL-8 was associated with both an increased odds of seroconversion and an increased density of highly-susceptible HIV target cells in the foreskin. In addition we found that male circumcision, a procedure that significantly reduces HIV acquisition, progressively reduced detection of coronal sulcus IL-8 during two years of follow-up (PRR of 0.29 at 24 months post-circumcision). Overall, these results suggest that the protective effect of male circumcision against HIV may be mediated in part through reductions in genital inflammation and the subsequent inflammation-mediated recruitment of HIV-susceptible cells to the foreskin. Although the mechanism(s) underpinning the relationship between cytokines and HIV susceptibility could not be fully elucidated by this observational study, the observation that coronal sulcus IL-8 and MIG were associated with HIV seroconversion is in keeping with a recent report that pro-inflammatory vaginal cytokines in women predict HIV acquisition [23], and with in vitro experiments demonstrating that IL-8 increases HIV susceptibility in cervical explants [20].
Both IL-8 and MIG belong to the chemokine family, a group of structurally similar small molecules that act as chemoattractants for immune cells expressing appropriate receptors. Since HIV predominantly infects CD4 T cells [11, 13, 44], both the availability and the HIV-permissivity of local CD4 T cells may dictate whether or not infection is established, with a limited number of highly susceptible cells driving initial mucosal infection [45, 46]. Therefore, recruitment or activation of specific subsets of CD4 T cells that are especially permissive to HIV may be important; Th17 and Th1 cells are highly HIV-permissive in vitro and are preferentially depleted in vivo during acute infection [47–50], and Th17 cells have recently been shown to be the primary targets of SIV, representing 64% of infected cells 48 hours after vaginal challenge [44]. Furthermore, men who are regularly HIV exposed but remain seronegative (HESN men) have a decreased relative abundance of both Th17 and TNFα+ CD4 T cells in their foreskin tissue [43]. Our finding that detectable coronal sulcus IL-8 was significantly associated with an increased overall number of CD4 T cells, including an increased density of highly susceptible Th17, Th1 and TNFα+ CD4 T cells, suggests that target cell availability may contribute to the association between coronal sulcus chemokines and HIV susceptibility.
While the association between foreskin IL-8 and HIV target cells does not prove this cytokine recruits or is produced by HIV target cells, a causal relationship is plausible. Epithelial cells (including keratinocytes) and macrophages responding to early immune stimuli are the main source of IL-8 during skin inflammation [51–54]. While many cell types express the IL-8 receptors CXCR1 and CXCR2 [55], IL-8 primarily recruits and activates neutrophils [51, 56, 57], which is consistent with our data showing a 4-fold higher density of neutrophils in foreskin tissues of men with high prepuce IL-8 levels. Neutrophils recruited by IL-8 are activated by bacterial antigens in the presence of inflammatory cytokines (IFNγ) to produce Th17-recruiting chemokines (MIP-3α and MCP-1) and Th1-recruiting chemokines (MCP-1 and IP-10) [58–60]. In turn, Th17 and Th1 cells produce IL-17 and IFNγ, respectively [61], both feeding back into the inflammatory cascade: IL-17 stimulates epithelial cells to produce more IL-8 [62], and IFNγ contributes to neutrophil chemokine production [60, 63]. Reciprocally, Th17 cells may directly contribute to IL-8 levels: Th17 cells have recently been shown to be the only subset of CD4 T cells to produce high levels of IL-8 [60, 64, 65]. In support of this, treatment with antibodies preventing the formation and action of Th17 cells (secukinumab, specific for IL-17A/IL-23) prevents neutrophil recruitment to the skin and keratinocyte immune activation, and reduces local levels of IL-8 [66]. Thus, IL-8 may be part of a positive-feedback loop, whereby local innate immune cells recruit neutrophils through IL-8, which in turn recruit HIV target cells through MIP-3α and MCP, and these HIV target cells then produce inflammatory cytokines, feeding back into local innate immune cell activation and IL-8 production. We hypothesize that this positive-feedback loop provides a causal basis for the association that we observed between prepuce IL-8 levels and HIV target cells, and that HIV target cells recruited by this mechanism drive the observed increased risk of seroconversion.
HIV seroconversion in our cohort was also associated with the detection of MIG and was increased if multiple cytokines were present. MIG is produced by macrophages in response to IFNγ and directly recruits activated T cells through CXCR3 [67], consistent with our observations that men with detectable MIG had increased numbers of foreskin CD8 T cells and a trend towards increased CD4 T cells (p = 0.08). The recruitment of IFNγ-producing T cells by MIG may also feed into to the relationship between IL-8 and HIV target cell recruitment, since IFNγ contributes to neutrophil activation and Th1/Th17 cell recruitment (as above). Of note, MCP-1 and MIP-3α were the next most abundant cytokines detected in this study (after IL-8 and MIG), and these two chemokines have been shown to specifically mediate the recruitment of Th17 cells by neutrophils [60]. We therefore hypothesize that increased risk of seroconversion with multiple cytokines is due to MIG, MCP-1 and MIP-3α feeding into the local inflammatory feedback loop described above. However, while this observational study provides important in vivo human data, further in vitro or animal studies will be necessary to completely define the causal nature of these associations.
Other cell types, such as Langerhans cells and dermal dendritic cells have been shown to facilitate HIV infection in the foreskin [68], and may contribute to increased HIV-susceptibility observed in this study. While tissue density of Langerhans cells and dermal dendritic cells was not associated with prepuce IL-8 or MIG, local tissue inflammation may contribute to dendritic cell maturation and function, thereby facilitating HIV transfer to susceptible T cell populations, as has been previously described [68, 69].
The reduction in coronal sulcus chemokines after male circumcision sheds light on a potential biological mechanism by which circumcision protects against HIV acquisition: reduced penile inflammation. We observed a significant decrease in coronal sulcus IL-8 at the first follow-up visit after circumcision (PRR = 0.59 at 6 months, 95% CI: 0.40–0.88), and IL-8 continued to decline up to study conclusion (PRR = 0.29 at 24 months, 95% CI: 0.16–0.54), which was significantly lower than at month 6. This may reflect a gradual decline in HIV target cells within remaining penile tissue, as effector T cells have been shown to be slow to clear a site of previous infection, and are enriched in tissue sites for months after antigen becomes undetectable in the skin [70]. In vitro studies also suggest that target cell availability contributes to the efficacy of male circumcision, as the inner aspect of the foreskin contains a comparatively high density of HIV target cells [15, 33–35]. Male circumcision may reduce HIV susceptibility by reducing penile inflammation and HIV target cell availability.
The factors causing penile inflammation could not be fully delineated in this study. Prepuce cytokine levels in uncircumcised men do not correlate with urethral cytokines (Kaul R and Galiwango R; unpublished), and we found no association with female partner vaginal cytokines, suggesting that prepuce cytokines do not derive from these sources, and may be produced in foreskin tissue. No significant associations were observed between IL-8 or MIG and age, number of sex partners, condom use, genital washing or seroprevalent STIs (HSV-2 and syphilis). While self-reported STI symptoms were associated with increased detection of both IL-8 and MIG, this did not fully account for the association between these cytokines and seroconversion. Additionally, even in symptom-free participants uninfected by HSV-2 and syphilis, we observed IL-8 concentrations ranging over 1000-fold (>1.5 to 2626.9pg/ml), suggesting that “normal” penile immune parameters are highly variable, and may be affected by factors other than classical STIs. This is in agreement with findings in South African women, where vaginal chemokine levels were associated with increased risk of HIV acquisition, but were incompletely explained by the presence of STIs [23]. Determining the factors contributing to heterogeneity in genital inflammation warrants further research as prior simulation studies have shown that variability in HIV susceptibility can affect HIV epidemic dynamics significantly and may explain differences in HIV epidemic trajectories between populations [71].
One possible contributor to penile inflammation is the resident microbiome. Alterations in the vaginal bacterial community in women, such as bacterial vaginosis, are associated with increased risk of HIV [72–75], possibly due to local inflammation [76, 77]. Th17 cells are essential in the defense against bacterial infections [78, 79], and colonization by pathogenic bacteria may increase HIV susceptibility by increasing Th17 cell density [50]. We have previously found that uncircumcised men are more likely to have BV-associated anaerobic bacteria in their sub-preputial space [80], and that circumcision gradually reduces both the total bacterial load and the abundance of these anaerobes [32]. Of note, while anaerobe abundance decreased rapidly within 6 months post-circumcision, it continued to decline for up to 24 months. This is similar to the gradual decline in IL-8 levels that we observed over the same period, and so the role of the penile microbiome as a driver of tissue inflammation and HIV susceptibility may be an interesting area for future study.
A limitation of the current work was the low concentration of cytokines in coronal sulcus swabs, especially swabs collected during the circumcision RCT. Swabs from the RCT were stored at -80°C for up to 10 years between collection and cytokine analysis. Cytokines, including IL-8, have been shown to degrade after 4 years, despite ideal storage conditions [81]. This likely explains the difference in detectability of IL-8 between swabs collected during the RCT and those collected from men attending the Circumcision Service Program, as swabs from the latter group were analyzed within one year of collection. However, it is unlikely that IL-8 degradation can account for differences observed between comparator groups in this study. In the case control study of HIV seroconversion, controls were matched to cases based on visit (time) and swab storage time did not vary between groups (median 4.7 years for both groups). Additionally, we found no correlations between IL-8 concentration and date of swab collection, suggesting that variability in swab storage time due to the relatively short duration of the trial (August 2003- November 2006) cannot account for the differences in IL-8 levels observed when examining the impact of circumcision on cytokine levels. It is possible that associations with other cytokines may have been missed due to low analyte concentration, explaining why cytokines observed to be released ex vivo from foreskin explants were not detectable in swabs in this study [33, 35, 69].
In conclusion, penile inflammation is an important risk factor for HIV acquisition in heterosexual men. HIV acquisition was associated with elevated levels of coronal sulcus IL-8 and MIG, which correlated with an increased density of T cells in the underlying foreskin tissue. In particular, prepuce concentrations of IL-8 were correlated with both an increased overall tissue density of CD4 T cells, as well as an increased density of specific highly HIV-susceptible CD4 T cell subsets. Finally, circumcision progressively reduced coronal sulcus IL-8 for up to 24 months after the procedure, which suggests a reduction in penile inflammation may be one mechanism by which circumcision is protective against HIV. Identifying causes of penile inflammation and immune activation in otherwise healthy men may lead to novel interventions to reduce the sexual transmission of HIV.
We examined samples and data collected from two study populations enrolled through the Rakai Health Sciences Program (RHSP) in Uganda: one enrolled in an RCT of male circumcision, conducted from 2003–2006 [29]; and the second enrolled in an observational cross-sectional study through the RHSP Circumcision Service Program between 2010–2011 [39]. Study design and sample population selection are described in detail in the Statistical Methods Section, below.
Dacron swabs moistened with sterile saline and rotated twice around the full circumference of the penis at the coronal sulcus were collected from all men at enrollment and each follow up visit during the RCT, and once, immediately prior to circumcision, from the Circumcision Service participants. The same clinical officers collected swabs throughout both studies and care was taken to collect each swab in a consistent manner. Female partners of Circumcision Service Participants were asked to insert a Dacron swab into the vagina, rotate once, and remove it. Swabs were immediately placed in 1ml undiluted AMPLICOR STD Specimen Transport Kit medium (Roche Diagnostics, Indianapolis, IN) at 4°C for less than 4 hours, and then suspended, aliquoted and stored at −80°C. Foreskin tissue removed during circumcision was also collected from Circumcision Service Participants. Tissue was processed immediately upon surgical removal: two sections from distal locations on the foreskin (one from the approximate center of the inner aspect, and one from the center of the outer aspect) were snap frozen into cryomolds in Optimal Cutting Temperature (OCT) compound (both Fisher Scientific, Toronto, Canada) for immunohistochemistry; and one large section containing equal area of the inner and outer aspects reserved for T cell isolation.
An electrochemiluminescent detection system using a custom Human Ultra-Sensitive kit from Meso Scale Discovery (Rockville, MD) was used to assay cytokines in coronal sulcus swabs from both RCT and Service Program participants. Cytokines assessed were: IL-1α (interleukin-1α), IL-8, MCP-1 (monocyte chemotactic protein-1), MIG (monokine induced by γ-interferon), MIP-3α, RANTES (Regulated on Activation, Normal T cell Expressed and Secreted), and GM-CSF (granulocyte macrophage colony-stimulating factor). Samples from each of the three analysis sets (Coronal sulcus cytokines and HIV acquisition, Impact of circumcision on coronal sulcus cytokines, and Prepuce cytokines and foreskin T cell density) were run on plates from the same manufacturer’s lot, with samples from participant groups in each analysis set (i.e. cases and controls, circumcised and uncircumcised) distributed randomly and evenly proportioned across plates. Samples were run in duplicate, and results with a coefficient of variation (CV) above 20% for the two wells were re-run. An internal biological control was run in at two concentrations in duplicate on every plate to monitor plate-to-plate variability: internal control was made from pooled mucosal samples from 5 donors, with any low level analytes augmented by adding the relevant recombinant human protein. This sample was aliquoted for single use, and run both neat and diluted 1/20 on each plate (for biological and low-concentration controls). Plates were re-run if the concentration of any analyte in the internal control was >3 standard deviations different from the average concentration from that analyte for the first 5 plates run. Plates were imaged using the Sector Imager 2400A platform (Meso Scale Discovery). The study lower limit of quantification (LLOQ) for each analyte were as follows: IL-1α = 0.6pg/ml; IL-8 = 1.5pg/ml; MCP-1 = 0.6pg/ml; MIG = 0.3pg/ml; MIP-3α = 3.0pg/ml; RANTES = 0.6pg/ml; and GM-CSF = 0.3pg/ml. Cytokine concentrations reported are not normalized, and are that of swab resuspended in 1ml transport buffer. Levels are therefore significantly lower than true concentration on the penis surface.
T cells were isolated from foreskin samples obtained from Service Program participants as previously described [39]. Mononuclear cell counts were determined by trypan blue exclusion and 10-20x106 cells (depending on yield) were stimulated with either 1ng/ml phorbol-12-myristate-13-acetate (PMA) and 1μg/ml ionomycin (both from Sigma; St. Louis, MO, USA) or vehicle (0.1% DMSO) in the presence of 5μg/ml Brefeldin A (GolgiPlug, BD Biosciences). After stimulation, samples were stained for CD3 (UCHT1), CD4 (RPA-T4), CD8 (SK1) and CCR5 (2D7/CCR5; all BD Biosciences). Samples for intracellular staining were permeabilized using BD Cytofix/Cytoperm solution and stained with TNFα (MAb11), IFNγ (B27; all BD Biosciences), or IL-17A (eBio64DEC17; eBiosciences). Samples were acquired using a FACSCalibur flow cytometer (BD Systems). Gating was performed as previously described [39] by investigators blinded to participant status and cytokine levels.
Proportions of T cell subsets were converted to absolute numbers per mm2 foreskin tissue using CD3 IHC as previously described [83]. Briefly, OCT-cryopreserved tissues were sectioned, fixed in 2% formaldehyde, and frozen for batch staining. Sections were stained with anti-CD3 antibody, followed by biotin-labeled secondary, Alkaline Phosphatase Streptavidin Labeling Reagent and Substrate Kit Vector Red (all Vector Labs, Burlingame, CA), and then counterstained with Mayer’s Hematoxylin (Fisher Scientific). The number of CD3+ T cells per mm2 of tissue for each patient was derived from the average of two biopsies taken from distal locations on the foreskin (median 6.10mm2 tissue/patient analyzed). Whole sections were scanned using the TissueScope 4000 (Huron Technologies, Waterloo, Canada) and image analysis software (Definiens, München, Germany) was used to delineate the apical edge of the epidermis to a depth of 300μm into the dermis (excluding artifacts or folds). CD3+ cells within this area were manually counted by a single investigator blinded to cytokine levels and participant status.
Neutrophil, Langerhans cell, and dermal dendritic cell density was assessed using immunofluorescence in a subset of men with high (n = 5) and low (n = 5) levels of coronal sulcus cytokines. Tissues cryopreserved in OCT were sectioned to 5μm using a Leica CM3050 cryostat (Leica Microsystems, Wetzlar, Germany), mounted on glass microscope slides, fixed for 7 minutes in ice-cold acetone, air-dried, and frozen at -80°C for batch staining. For staining, slides were thawed, permeabilized in PBS-Tween 20 for 20 minutes, and blocked using a streptavidin/biotin blocking kit (Vector Labs) and 10% normal rabbit serum. Neutrophils were visualized using biotin-labeled mouse anti-human CD15 antibody (eBiosciences) followed by DyLight 488 Streptavidin (Vector Labs) secondary. Dendritic cells were visualized with goat anti-human CD207 antibody (R&D Systems) and biotin-labeled mouse anti-human CD11c (eBiosciences). Slides were then washed and mounted using Vectashield HardSet Mounting Medium with DAPI Counterstain (Vector Labs), according to manufacturer’s instructions. Whole sections were scanned using the Zeiss Axioscan (Carl Zeiss Microscopy, Cambridge, UK) and image analysis software (Definiens) was used to delineate and quantify the epidermal and dermal tissue (excluding artifacts or folds). Definiens was then used to count cell populations using a threshold set by an investigator blinded to cytokine levels and participant status. CD15+ cells were counted within total foreskin area (median area analyzed = 4.06 mm2), CD207+ cells in the epidermal tissue (1.36 mm2), and CD11c+ cells within dermal tissue (3.39 mm2).
We used Stata 13.1 for Mac (College Station, TX, USA) to conduct statistical analysis and Prism 5.0 (GraphPad Software; La Jolla, CA, USA) to construct graphs. Flow cytometry data was analyzed in FlowJo 9.8.2 (Treestar; Ashland, OR, USA). All tests two-sided with α = 0.05.
|
10.1371/journal.pntd.0001684 | Local Increase of Arginase Activity in Lesions of Patients with Cutaneous Leishmaniasis in Ethiopia | Cutaneous leishmaniasis is a vector-borne disease that is in Ethiopia mainly caused by the parasite Leishmania aethiopica. This neglected tropical disease is common in rural areas and causes serious morbidity. Persistent nonhealing cutaneous leishmaniasis has been associated with poor T cell mediated responses; however, the underlying mechanisms are not well understood.
We have recently shown in an experimental model of cutaneous leishmaniasis that arginase-induced L-arginine metabolism suppresses antigen-specific T cell responses at the site of pathology, but not in the periphery. To test whether these results translate to human disease, we recruited patients presenting with localized lesions of cutaneous leishmaniasis and assessed the levels of arginase activity in cells isolated from peripheral blood and from skin biopsies. Arginase activity was similar in peripheral blood mononuclear cells (PBMCs) from patients and healthy controls. In sharp contrast, arginase activity was significantly increased in lesion biopsies of patients with localized cutaneous leishmaniasis as compared with controls. Furthermore, we found that the expression levels of CD3ζ, CD4 and CD8 molecules were considerably lower at the site of pathology as compared to those observed in paired PBMCs.
Our results suggest that increased arginase in lesions of patients with cutaneous leishmaniasis might play a role in the pathogenesis of the disease by impairing T cell effector functions.
| The leishmaniases are a complex of diseases caused by Leishmania parasites. Currently, the diseases affect an estimated 12 million people in 88 countries, and approximately 350 million more people are at risk. The leishmaniases belong to the most neglected tropical diseases, affecting the poorest populations, for whom access to diagnosis and effective treatment are often not available. Leishmania parasites infect cells of the immune system called macrophages, which have the capacity to eliminate the intracellular parasites when they receive the appropriate signals from other cells of the immune system. In nonhealing persistent leishmaniasis, lymphocytes are unable to transmit the signals to macrophages required to kill the intracellular parasites. The local upregulation of the enzyme arginase has been shown to impair lymphocyte effector functions at the site of pathology. In this study, we tested the activity of this enzyme in skin lesions of patients presenting with localized cutaneous leishmaniasis. Our results show that arginase is highly upregulated in these lesions. This increase in arginase activity coincides with lower expression of a signalling molecule in lymphocytes, which is essential for efficient activation of these cells. These results suggest that increased arginase expression in the localized cutaneous lesions might contribute to persistent disease in patients presenting with cutaneous leishmaniasis.
| The leishmaniases are a complex of vector-borne diseases caused by the parasite Leishmania. They are neglected tropical diseases, that affect the poorest population and cause major morbidity and mortality, estimated to 2.4 million disability-adjusted life-years [1]. Currently, these diseases affect an estimated 12 million people in 88 countries, and approximately 350 million people are at risk [2]. Leishmaniases can present with a wide range of clinical syndromes that may be cutaneous or visceral: cutaneous leishmaniasis (CL) is manifested as localized (LCL), mucocutaneous (MCL) or mucosal (ML) and diffuse (DCL) disease [3]. Visceral leishmaniasis (VL), the most severe form of leishmaniasis, is a systemic disease, in which the mortality rate can be as high as 100% if left untreated. Adequate treatment results in an overall cure rate of >90% [3].
Leishmaniasis is one of the most important vector-borne diseases in Ethiopia, where it is mainly prevalent in the highlands. However, there is still only very limited information from epidemiological studies about the number of VL and CL cases. According to the Ethiopian National Guidelines for the Diagnosis and Treatment of Leishmaniasis, Ethiopia has the second largest number of VL cases in sub-Saharan Africa with an estimated 4500 to 5000 new cases every year. VL is associated with high mortality and morbidity, and is worsened by poor nutrition, isolated location of VL endemic areas and co-infections with HIV [4]. Similarly, there is limited data about the frequency and distribution of CL in Ethiopia [5], [6], [7], [8], [9], [10], [11]. CL in Ethiopia is mainly caused by Leishmania (L.) aethiopica, and rarely by L. tropica or L. major [12] and can manifest as LCL, with localized cutaneous nodular lesions, that can ulcerate and heal, leaving depressive scars (LCL); DCL, which is characterized by disseminated nodular lesions; and MCL, with lesions spreading into the nasal and/or oral mucosa [13]. LCL usually heals spontaneously within 1 year [7], however, persistent LCL as well as MCL and DCL require treatment; relapses are frequent after treatment in DCL and MCL [3], [12].
One experimental model of cutaneous leishmaniasis caused by L. major has been extensively studied: in this model, infection of BALB/c mice induces progressive nonhealing lesions; this inability to control infection has been associated with a polarized T helper (Th) 2 response. In contrast, C57BL/6 or CBA mice can efficiently control parasite replication and become immune to secondary challenge, this has been ascribed to a Th1 response [14], [15], [16]. In sharp contrast, infections of different strains of mice and other rodents with L. aethiopica does not lead to obvious clinical symptoms, even though parasites can be isolated from L. aethiopica infected BALB/c mice [17], [18], [19]. The exceptions are the Syrian and the CBC hamsters, which can be successfully infected into the nose, and produce lesions similar to those observed in DCL patients [19]. There is also very limited information on the immune response in L. aethiopica infected patients. It has been shown that LCL, but not DCL patients will respond to leishmanin skin test [20]. Furthermore, whereas mononuclear cells from LCL patients can proliferate and express cytokines in vitro in response to antigenic restimulation, those from DCL patients have an impaired capacity to become activated [20], [21], [22]. While the mechanisms responsible for this hyporesponsiveness are not yet clarified, it has been suggested that lower levels of IFN-γ and increased expression of IL-10 might contribute to immunosuppression in DCL patients [21]. CD8+ T cells and NK cells may also play a protective role [23].
The catabolism of L-arginine by arginase is emerging as a critical mechanism of immune regulation [24], [25], [26]. Arginase, which is typically considered to be an enzyme of the urea cycle in the liver, hydrolyzes L-arginine to urea and ornithine, which is further metabolized into polyamines. Arginase can be upregulated by cytokines such as IL-4 and IL-13, which can synergize with IL-10 and IL-21, as well as by inflammatory stimuli (summarized in [26]).
Upregulation of arginase in myeloid cells results in increased uptake of extracellular L-arginine, thus reducing L-arginine levels in the microenvironment. Since T cells unconditionally require L-arginine for efficient activation, decrease in L-arginine results in impaired T cell responses [24], [25], [26]. The downregulation of T cell responses by arginase-induced L-arginine depletion has been studied in several cancer models [25], in corneal transplantation [27] and pregnancy [28] and increased arginase activities have been associated with a variety of infectious diseases such as schistosomiasis [29], trypanosomiasis [30], tuberculosis [31], leishmaniasis [32], [33], hepatitis B [34] and HIV [35].
We have recently shown in an experimental model of cutaneous leishmaniasis that high arginase activity is a hallmark of nonhealing disease [32] and that this increased arginase contributes to persistent nonhealing leishmaniasis by causing local suppression of T cell responses [33].
To determine whether our experimental data translate to human disease, we tested whether enhanced arginase activity is present in biopsies of LCL patients and whether this coincides with T cell suppression.
The study was approved by the Ethiopian National Research Ethics Review Committee (NRERC, reference 310/18/03), by Addis Ababa University Medical Faculty Institutional Review Board (IRB, reference 023/2009) and by the Joint UCL/UCLH Committees on the Ethics of Human Research (Committee Alpha, reference 09/H0715/93). For this study, a cohort of 15 patients with localized cutaneous leishmaniasis was recruited from the Leishmaniasis Research and Diagnostic Laboratory, Addis Ababa University, Ethiopia. Ten healthy controls were recruited among the staff of the hospital; they had careful physical examinations and showed no cutaneous lesions and had no prior history of cutaneous leishmaniasis. Informed written consent was obtained from each patient and control and all data analyzed were anonymized. 8–20 ml of blood in EDTA tubes and 1 or 2 biopsies (3 or 4 mm) were collected from each patient from the edge of the active lesion before the treatment started; or from intact skin on one forearm from the healthy controls. Patients positive for HIV were excluded from the study. Both the biopsies and blood were processed immediately after harvesting. Peripheral blood mononuclear cells (PBMCs) were isolated by density gradient centrifugation on Histopaque®-1077 (Sigma). Cells were washed in phosphate buffered saline (PBS) and were immediately used for flow cytometry; PBMCs used for arginase and protein determination were immediately resuspended in lysis buffer (0.1% Triton X-100, 25 mM Tris-HCl and 10 mM MnCl2, Sigma) and then frozen at −20°C until further use. Biopsies were collected in PBS and homogenized in PBS for flow cytometry or in lysis buffer and frozen until arginase and protein assays were performed.
The enzymatic activity of arginase was measured as previously described [35]. Briefly, cell lysates were activated by heating for 10 min at 56°C. L-arginine hydrolysis was conducted by incubating the activated lysates with 50 µL 0.5 M L-arginine (pH 9.7) at 37°C for 60 min. The reaction was stopped with 400 µL H2SO4 (96%)/H3PO4(85%)/H2O (1∶3∶7, v/v/v). Twenty µL α-isonitrosopropiophenone (ISPF, dissolved in 100% ethanol, Sigma) was added and incubated for 45 min at 100°C, followed by 30 min at 4°C. The optical density (OD) was measured at 550 nm. One unit of enzyme activity is defined as the amount of enzyme that catalyzes the formation of 1 µmol of urea per min.
To determine the protein concentration of each PBMC sample, serial dilutions of each PBMC sample were made in PBS (Sigma). BCA Protein Assay Reagent (Pierce) was added to each PBMC dilution following supplier's recommendations. A bovine serum albumin (BSA) standard (Pierce) was serially diluted using PBS. Following 30 min incubation at 37°C, the optical density (OD) was measured at 570 nm.
Antibodies used were as follows: anti-CD4 (clone 13B8.2, Beckman Coulter), anti-CD8 (clone RPA-T8, BD Biosciences), anti-CD3ζ (Santa Cruz: clone 6B10.2), anti-CD14 (BD Pharmingen: cloneM5E2), anti-CD15 (Clone H198, BD Pharmingen); anti-arginase I (HyCult Biotechnology: clone 6G3) and the isotype control (BD Pharmingen: clone MOPC21) were coupled with Alexa FluorR 488 (Molecular Probes). Cells were washed with PBS, the fixation step was performed with 2% formaldehyde in PBS and the permeabilisation step with 0.5% saponin in PBS.
The determination of intracellular arginase was performed as described in [35]. The percentages for the isotype controls were <1.5%. Acquisition was performed using a FACSCalibur (BD Biosciences) and data were analyzed using Summit v4.3 software.
Data were evaluated for statistical differences using a two-tailed Mann-Whitney test, Wilcoxon pair test or Spearman's rank test when appropriate (GraphPad Prism 5) and differences were considered statistically significant at p<0.05.
Fifteen patients with lesions of LCL that were typical in their history and appearance were recruited for this study. All patients lived or had travelled in regions of Ethiopia endemic for CL caused by L. aethiopica; however, the infecting parasites were not typed. The diagnosis was confirmed by demonstration of amastigotes in skin scraping and growth of promastigotes in NNN medium. Out of the 15 patients recruited in this study, 7 were female and 8 were male, with a median age of 19±2.4 (Table 1). The large majority of the patients presented with nodular lesions (13 patients), 1 patient with an ulcerated lesion and 1 patient with ulcerated and nodular mixed lesions. Ten patients had 1 lesion and 5 patients had 2 lesions. The majority of the lesions (12) were found on the face (forehead, ear, cheek, lip, nose), 3 on the forearm, 1 on the neck and 1 on a finger. The duration of their illness ranged from 4 to 48 months (median ± SEM: 12 months±3.6): 7 patients had lesions for <12 months and 8 patients had lesions for >12 months (Table 1).
We first assessed the levels of arginase activity in PBMCs of LCL patients and compared it with healthy controls. The arginase activity was not statistically increased in the PBMCs of LCL patients (54.5 vs 45.1 mU/mg protein, p = 0.2751, Figure 1). We then determined the phenotype of arginase expressing cells in the PBMCs of LCL patients and controls and the results showed that the cells expressing arginase are low-density granulocytes (LDGs = CD15+ CD14low, Figure 2A). In all 15 patients tested, >93% of CD15+ cells expressed arginase. Similarly, the large majority of arginase-expressing cells in the PBMCs obtained from the controls were CD15+ (data not illustrated). In contrast, the frequency of arginase-expressing monocytes (CD14+ CD15− arginase+) was below 1% (Figure 2B), except for 2 patients (1.1% and 1.2%, data not illustrated). Both cells types - LDGs and monocytes - were present in distinct regions of the forward and side scatter (FSC/SSC) dotplot: LDGs were found in region R2 (Figure 2C) and monocytes in region R3 (Figure 2C). The frequencies of LDGs (Figure 3A), monocytes (Figure 3B) and the ratio of LDGs/monocytes (Figure 3C) in PBMCs were similar between controls and LCL patients (median±sem: %CD15+: 6.0±0.66 vs 5.7±1.16; %CD14+ cells: 10.0±1.48 vs 10.2±1.21; ratio CD15+/CD14+: 1.90±0.43 vs 1.7±0.71, p>0.05).
The results presented in Figures 1, 2, and 3 show that the arginase activity and frequency of arginase-expressing cells in PBMCs of LCL patients are not significantly increased. These results also establish that the arginase-expressing cells in the blood of patients and controls are neutrophils.
As shown in Figure 4, high levels of arginase activity were measured in lesions of LCL patients; notably, they were significantly higher than those measured in intact control skin (279.2±41.1 vs 18.8±6.3 mU/mg protein, p = 0.0002). There was no significant correlation between arginase activities and lesions size or duration of lesions (p>0.05).
To identify the phenotype of arginase-expressing cells in the lesions, we used the same combination of cell surface and intracellular arginase labeling as for the PBMCs. We were able to isolate enough cells (>1000 CD15+ cells) to be analyzed by flow cytometry from homogenates of skin biopsy from 3 out of 10 patients, who had a 4 mm biopsy taken. A CD15+ population was detected in patients' biopsy (Figure 5), in a region that was similar to that of LDGs detected in the PBMCs of the same patient (FSS/SSC: LDGs = 95/85, PBMCs = 101/82). In contrast, the frequency of CD14+ cells in the biopsies was very low (<250 events) and it was therefore not possible to characterise these cells in detail.
These results show that arginase activity is considerably increased in biopsied skin lesions of LCL in Ethiopia and suggest that arginase-expressing CD15+ cells are also present in the lesions of patients with cutaneous leishmaniasis in Ethiopia.
Our results depicted in Figures 1 and 4 and summarized in Figure 6 show that cells isolated from cutaneous lesions express significantly higher levels of arginase activity per mg of protein as compared to cells isolated from peripheral blood of the same LCL patients (n = 10, 279.2±41.1 vs 53.4±6.4, p = 0.0020). We have previously shown that in a mouse model of CL, high arginase activity causes depletion of L-arginine, which impairs antigen-specific T cell responses [33]. Decreased expression of CD3ζ in T cells has been extensively used as marker of arginase-induced T cell suppression [24], [25]. Therefore, here we determined whether high arginase activity observed in the cutaneous lesions of patients coincides with lower expression levels of CD3ζ in CD4+ and CD8+ T cells as compared to those in the peripheral blood. First we compared the frequency and ratio of CD4+ and CD8+ T cells in the blood and the biopsies and show that the percentage of CD4+ T cells is similar in both compartments (Figure 7A). Interestingly, there was a higher frequency of CD8+ T cells in biopsies (p = 0.0071, Figure 7B). The ratio of CD4/CD8+ T cells was higher in the PBMCs, but it was not statistically significant (p = 0.075, Figure 7C). Of note, these frequencies and ratios were also comparable between PBMCs isolated from the blood from controls and patients (Table 2). Next, we measured the mean fluorescence intensity (MFI) of CD3ζ in T cells from homogenates of skin biopsy and compared it to those in cells isolated from the blood of the same patient. Results in Figures 8A–C show that the CD3ζ MFI in CD4+ T cells was lower in the biopsies of 10 out of 12 patients (p = 0.0024). Interestingly, the MFI of the CD4 molecule on T cells (CD4 MFI) was also lower in the biopsies as compared to the blood in 10 out of 12 patients (p = 0.0024, Figures 8D–F). Similar results were obtained with the expression levels of CD3ζ and CD8 molecules: it was decreased in 11 out of 12 patients tested (p = 0.0010, Figures 9A–C); moreover, CD8 MFI were lower in the biopsies of all patients tested (p = 0.0005, Figures 9D–F).
These results show that the expression levels of CD3ζ, as well as CD4 and CD8 molecules are reduced in the skin biopsies as compared to those present in the blood of the same patient.
Here we showed that our results obtained in a mouse model translate to human disease: the levels of arginase activity were similar in the cells isolated from peripheral blood of LCL patients and controls, showing that arginase is not increased in the periphery following infection with Leishmania parasites. In sharp contrast, arginase was clearly increased in skin lesions of LCL patients as compared to intact skin. Moreover, arginase activity was considerably higher in the cells isolated from the skin biopsies as compared to the PBMCs. These results are in agreement with those obtained in the mouse model, where we showed that arginase is upregulated at the site of pathology, but not in the periphery [33].
The cells expressing arginase were identified as neutrophils both in the blood and in the lesions. In the blood, these cells co-purify with PBMCs and have therefore been named low-density granulocytes (LDGs). We and others have already described these cells [36] [28], [35], [37] and have shown that they were likely to be activated granulocytes [38]. Whereas we have previously shown that bone marrow derived macrophages activated with IL-4 or IL-4 and IL-10 upregulate arginase [32], [39], [40], we have not yet identified the phenotype of arginase-expressing cells in lesions of BALB/c mice infected with L. major. We have previously shown that both macrophages and neutrophils are recruited into lesions of L. major infected mice [41], [42], therefore we cannot exclude that in these lesions, macrophage and/or neutrophils express arginase. We could identify only low numbers of macrophages in the lesions (<250 events), this was unexpected as Leishmania-infected macrophages can be identified in scrapings from cutaneous lesions and in fixed biopsies [43], [44]. It is possible that the technique used to homogenize the lesions damages the macrophages and it is therefore not possible to conclude from this study whether macrophages also contribute to the overall arginase measured in the lesions.
Increased arginase activity in macrophages has been shown to favour parasite growth [32], [45]. Leishmania parasites are taken up by neutrophils (reviewed in [46], [47] and it is possible that this gives them a survival advantage. However, since they have not been shown proliferate efficiently in neutrophils, here we favour the hypothesis that neutrophil arginase affects both wound healing (1) and T cell responses (2), two processes that are crucial in resolving cutaneous lesions:
Persistent leishmaniasis has been associated with immune suppression [20], [21], [22]. The biopsies analyzed in the present study were all collected from LCL patients. The observed natural history of LCL is that ∼70% heal within 12 months and 30% within 24 months [7]. However, it is not possible to predict whether a lesion will heal or become chronic and indeed, 8 patients had their lesions for 12 months or more. Therefore, we propose that high arginase results in impaired T cell responses and therefore contributes to the delay of healing that is characteristic in LCL in Ethiopia. Of note, the majority of LCL lesions analysed here were nodular and we cannot exclude that ulcerated lesions might express different levels of arginase and/or CD3ζ. The results of this study call for further work to analyse in detail different types of LCL lesions.
Cutaneous leishmaniasis causes serious morbidity in Ethiopia. The lesions, which are most commonly on the face, are chronic, disfiguring and sometimes disabling, and cause significant social stigma [62]. To date, the immunopathology of this disease has been poorly understood. Here we show for the first time that arginase is upregulated in lesions of patients with LCL and that this coincides with reduced levels of CD3ζ expression in T cells. Further study is needed to assess whether arginase-mediated L-arginine metabolism is a key element in the outcome of human leishmaniasis and this is currently ongoing in our laboratories. Therapeutic interventions that can regulate arginase and L-arginine metabolism might prove useful in the treatment of cutaneous leishmaniasis, and possibly in visceral leishmaniasis.
|
10.1371/journal.pcbi.1003131 | Predictors of Hepatitis B Cure Using Gene Therapy to Deliver DNA Cleavage Enzymes: A Mathematical Modeling Approach | Most chronic viral infections are managed with small molecule therapies that inhibit replication but are not curative because non-replicating viral forms can persist despite decades of suppressive treatment. There are therefore numerous strategies in development to eradicate all non-replicating viruses from the body. We are currently engineering DNA cleavage enzymes that specifically target hepatitis B virus covalently closed circular DNA (HBV cccDNA), the episomal form of the virus that persists despite potent antiviral therapies. DNA cleavage enzymes, including homing endonucleases or meganucleases, zinc-finger nucleases (ZFNs), TAL effector nucleases (TALENs), and CRISPR-associated system 9 (Cas9) proteins, can disrupt specific regions of viral DNA. Because DNA repair is error prone, the virus can be neutralized after repeated cleavage events when a target sequence becomes mutated. DNA cleavage enzymes will be delivered as genes within viral vectors that enter hepatocytes. Here we develop mathematical models that describe the delivery and intracellular activity of DNA cleavage enzymes. Model simulations predict that high vector to target cell ratio, limited removal of delivery vectors by humoral immunity, and avid binding between enzyme and its DNA target will promote the highest level of cccDNA disruption. Development of de novo resistance to cleavage enzymes may occur if DNA cleavage and error prone repair does not render the viral episome replication incompetent: our model predicts that concurrent delivery of multiple enzymes which target different vital cccDNA regions, or sequential delivery of different enzymes, are both potentially useful strategies for avoiding multi-enzyme resistance. The underlying dynamics of cccDNA persistence are unlikely to impact the probability of cure provided that antiviral therapy is given concurrently during eradication trials. We conclude by describing experiments that can be used to validate the model, which will in turn provide vital information for dose selection for potential curative trials in animals and ultimately humans.
| Innovative new approaches are being developed to eradicate viral infections that until recently were considered incurable. We are interested in engineering DNA cleavage enzymes that can cut and incapacitate persistent viruses. One hurdle is that these enzymes must be delivered to infected cells as genes within viral vectors that are not harmful to humans. In this paper, we developed a series of equations that describe the delivery of these enzymes to their intended targets, as well the activity of DNA cutting within the cell. While our mathematical model is catered towards hepatitis B virus infection, it is widely applicable to other infections such as HIV, as well as oncologic and metabolic diseases characterized by aberrant gene expression. Certain enzymes may bind DNA more avidly than others, while different enzymes may also bind cooperatively if targeted to different regions of viral DNA. We predict that such enzymes, if delivered efficiently to a high proportion of infected cells, will be critical to increase the probability of cure. We also demonstrate that our equations will serve as a useful tool for identifying the most important features of a curative regimen, and ultimately for guiding clinical trial dosing schedules to ensure hepatitis B eradication with the smallest number of possible doses.
| To date, cure of most chronic viral infections has remained an impossible goal. Replicating forms of hepatitis B virus (HBV), Herpes Simplex Virus (HSV) and Human Immunodeficiency Virus (HIV) can be targeted with potent small molecule therapies, thereby decreasing the burden of disease associated with these pathogens [1]–[4]. However, latent, non-replicating viral genomes persist within reservoirs for each of these infections, and high levels of viral replication typically resume soon after cessation of antiviral therapy, even after years of treatment [5]–[8]. Lifelong therapy is therefore often required, resulting in enormous costs to the healthcare system [9]. In addition, therapy can be complicated by lack of compliance, drug toxicity and resistance.
Curative approaches to these infections will need to target persistent, non-replicating viral genomes. DNA cleavage enzymes, including homing endonucleases (HE) or meganucleases, zinc-finger nucleases (ZFN), transcription activator-like (TALEN) effector nucleases and CRISPR-associated system 9 (cas9) proteins represent a promising new therapeutic approach for targeting these viral forms [10]. These enzymes can be designed to target specific segments of either episomal DNA for HBV and HSV, or integrated viral DNA for HIV, which are vital for replication [11], [12]. When viral DNA is cleaved, it is quickly repaired, allowing for repeated binding of the cleavage enzyme. DNA repair occurs by non-homologous end joining (NHEJ), an error prone process. The enzyme binds to the target site if no mutation occurs during repair. Eventually, target DNA incurs a deletion or insertion that prevents subsequent enzyme binding as well as translation of essential viral proteins. The remaining viral DNA is thereby rendered replication incompetent.
Zinc-finger nucleases are currently being used successfully ex vivo as a tool to modify the HIV entry receptors CCR5 and CXCR4 on CD4+ T-cells; altered cells which are resistant to HIV entry or replication have been transplanted back to infected animals as a form of adaptive immunotherapy with resultant decreases in viral load [13], [14], and this method is being tested in human clinical trials [15]. Similar modification of CCR5 in hematopoietic stem cells may allow reconstitution of the full immune system with exclusively HIV-resistant cells [16]. In contrast, our approach here is to use DNA cleavage enzymes that directly target latent viral genomes, rather than host cell viral entry receptors, to be delivered to infected cells as transgenes within viral vectors [10], [17]. However, numerous fundamental questions remain regarding such a gene therapy approach: which vectors are most appropriate for gene delivery? How many doses will be necessary for viral eradication? Can vector delivery be limited to decrease the probability of toxicity? If multiple doses are needed, how will immunity to the delivery vector impact the likelihood of cure? Does reconstitution of latently infected cells occur rapidly enough to necessitate a narrow interval between successive gene therapy doses? Is there benefit to delivering multiple transgenes per vector and should enzymes be engineered to target several regions of viral DNA [10]?
Mathematical models are crucial tools for identifying dynamics of active infection, and for designing antiviral regimens that maximize potency and avoid drug resistance [18], [19]. However, despite twenty years of experience with antiretroviral therapy, a unifying mathematical theory of pharmacodynamics, that contrasts different antiviral agents according to potency, likelihood of resistance, and therapeutic synergy has only recently been developed [20]–[25], and exists only for HIV-1 targeting agents. To maximize the probability that viral inactivation can be achieved, we believe that key quantitative components of gene therapy should be established early during development of DNA cleavage enzyme technology. With these fundamentals in place, dosing regimens can be designed rationally rather than blindly.
To this end, we developed theoretical models that capture different critical components of viral cure approaches with DNA cleavage enzymes. Our initial models and analyses focus on HBV infection. HBV infects hepatocytes, which are highly accessible to gene therapy delivery vectors and which can be assessed serially for clearance of non-replicating virus. For this reason, HBV may be the most promising initial target for cure. However, the model is easily expanded to account for parameters that govern HIV-1 and HSV infections.
We describe the mathematics of viral vector delivery to hepatocytes, and enzyme - substrate kinetics in the setting of heterogeneous density of episomal infection per hepatocyte. The theoretical problem of de novo resistance to DNA cleavage enzyme is also addressed. To this end, we consider concurrent vectorization of multiple enzymes that target separate DNA regions within the HBV episome. Finally, we incorporate simple differential equation models that capture dynamics of HBV persistence between gene therapy doses, and estimate how these dynamics may impact dosing strategies.
Our simulations suggest that therapeutic outcome is likely to hinge on four key factors: percent vector delivery to target cells per dose (which in turn depends on what proportion of vectors are removed by humoral immune mechanisms), enzyme-DNA target binding affinity and cleavage efficiency, degree of binding cooperativity between cleavage enzymes and target DNA, and number of transgenes delivered per vector. We predict that re-accumulation of the latent pool of HBV is unlikely to occur rapidly enough to overcome weekly dosing of delivery vectors, provided that viral replication is concurrently suppressed with available antiviral therapy. If cleavage enzymes that target single regions within the viral genome are used, de novo enzyme resistance could develop rapidly such that nearly all remaining episomes are therapy resistant following only a few doses of effective therapy. However, resistance to cleavage enzymes can be effectively mitigated if different DNA cleavage enzymes that cleave different regions of HBV episomes are dosed sequentially, or if single vectors can concurrently deliver several of these transgenes. As this model has yet to be confronted with empirical data, we also discuss potential cell culture and animal model experiments to help identify values for key model parameters, and to better inform future iterations of the model.
The HBV genome can exist in various states within a cell according to stage of replication. The persistent viral form is covalently closed circular (cccDNA), which is maintained with a half-life of months to years in cells [26], but is also a fundamental intermediate in the HBV replication cycle [27]. HBV is notable for an extraordinarily high burden of infection with most of the ∼2×1011 human hepatocytes harboring multiple HBV cccDNA episomes [28], as well as other replication intermediates including HBV that may be integrated into host chromosomal DNA [29], [30]. If fully suppressive antiviral therapy is given, the balance of remaining viral molecules is shifted in favor of cccDNA which remains in >95% of cells even a year after HBV DNA becomes undetectable in serum [31]. In model simulations, unless otherwise stated we assume that 99.67% of 2*1011 hepatocytes are infected based on a median of five infectious genomes per cell [32], [33]. We replicate the wide distribution of viral burden between cells for HBV with a Poisson distribution. The goal of gene therapy-delivered DNA viral cleavage enzymes will be to functionally disrupt all, or the vast majority of cccDNA, such that viral reactivation is impossible.
While we believe that parameters of gene therapy vector delivery and intracellular pharmacodynamics described in our models can ultimately be precisely identified, the parameter values that govern dynamics of HBV cccDNA persistence are likely to remain undetermined when our therapies are tested in animal and human trials. With these uncertainties in mind, the goal of our models is not to prove or disprove competing dynamical hypotheses of HBV cccDNA persistence [34], but rather to incorporate any possible features of reservoir maintenance that may challenge the effectiveness of gene therapy. As a practical matter, our assumptions are weighted to favor HBV persistence during treatment. We favor “pessimistic” models of latency to ensure that selected gene therapy regimens exceed thresholds for viral cure by a comfortable margin. Because it is generally agreed that cccDNA probably decays slowly even without eradicative therapies, we make the simplest pessimistic assumption, that cccDNA levels remain stable between doses, unless otherwise noted.
Gene therapy vectors can be delivered intravenously, allowing random dispersion to target hepatocytes. Entry into target cells can be achieved by utilizing vectors that are engineered to preferentially bind chosen cell surface receptors, such as sodium taurocholate cotransporting polypeptide, which are specific to HBV target cells [35]. Alternatively, viral vectors, which naturally target cell surface receptors that are ubiquitously expressed on the cell surface of target hepatocytes, such as the laminin receptor, heparan sulfate proteoglycans (HSPGs), sialic acids and other glycans, can be used [36]–[41]. During a single dose, all hepatocytes are equally susceptible but delivery of multiple vectors to one cell may occur allowing for multiple transductions. We make the assumption that entry of multiple vectors is not impaired following prior entry of a single vector.
In addition, the model is structured such that gene product, or DNA cleavage enzyme concentration in the cell nucleus is assumed to be directly proportional to the number of delivery vectors with successful entry and gene expression. Recombinant adeno-associated virus (AAV) can be produced in high titers in plasmid transfected cells [42] although non-infectious capsids exceed infectious DNA containing particles 10–30 fold [43]. A total of 60 copies of the capsid protein VP3 are needed to produce a single infectious particle during wild type AAV infections [44], suggesting that generation of a single double-stranded replicating form of AAV vector DNA can correspond with amplified transgene expression.
The heterogeneous distribution of gene therapy vectors to cells can be captured with an adjusted multiplicity of infection formula Pv = [(σ*m)v * e−(σ*m)]/v!, where m is the ratio of delivery vectors to target cells in the human body (including uninfected hepatocytes which presumably also take a high number of viral vectors), v is the number of vectors delivered per target cell, and Pv is probability of v transduced vectors per cell (Fig. 1a). Viral vectors such as AAV have been developed for delivery to the human liver and have successfully deployed at high does (2*1012 particles per kilogram) with success in human clinical trials for metabolic disorders [45]. This dose equates roughly to m = 1200, or 1200 particles per hepatocyte in a 60 kg adult.
Parameter σ is included to account for the fact that most vectors will not transduce their intended target cells. As a function of the development process, >90% of vector capsids lack viral DNA [43]. Other vectors may be removed by humoral immune mechanisms, enter cells which are not targets for HBV infection, or degrade due to shear forces or chemical stress in the blood. Finally, vector entry into a target cell's cytosol does not guarantee successful transduction, as viral nuclear localization sequences are required to bind nuclear transport receptors for nuclear entry [46]. Therefore the ratio of transduced vectors per target cells (σ*m), which we refer to as the functional MOI (fMOI), is likely to be far lower than the value for m which is ∼1200.
σ will take on a value of one if transduction of all dosed vectors occurs, and zero if no gene expression is achieved. This parameter value may be lower for infections such as HIV where latently infected cells potentially exist in anatomic sanctuaries such as the nervous system, as compared to HBV where vectors encounter the liver during first pass metabolism. Certain delivery vectors such as adenovirus (ADV) are immunogenic and delivery of identical serotypes will decrease with successive doses [47]. Enhanced neutralizing antibody response can prevent efficient delivery, thereby decreasing σ with successive doses, even when using less immunogenic vectors such AAV [48]–[54].
The delivery equation reveals a wide distribution of vector delivery and transduction when σ*m >1. If HBV infection is modeled with 2*1011 hepatocytes, even if 1012 vectors are delivered successfully (m = 1200, σ = 0.004, σ*m = 5), there is no transduction within a small percentage (Fig. 1b), but relatively large absolute number (∼109), of infected cells. If σ = 0.167 (σ*m = 20) is assumed, then a majority of hepatocytes will have multiple vector delivery (Fig. 1b). When σ*m<1, σ*m approximates proportion of cells with delivery and the majority of targeted cells contain only a single delivery vector (Fig. 1c).
The latter condition is unlikely to promote complete eradication of HBV cccDNA: if we make the simplifying and overly optimistic assumptions that delivery of one or more vectors automatically leads to lethal mutation of all viral genomes within a target cell, that no immunity to the viral vector or enzyme develops with successive doses of delivery vectors, and that there is no replenishment of infected hepatocytes or HBV cccDNA between doses, then the number of doses prior to eradication can be estimated with the formula Nn = N0 * (1−P(v>0))n where N0 is initial number of infected cells, Nn is the remaining number of infected cells following n doses, and cure occurs when Nn<1. The number of necessary doses increases dramatically if 50% delivery is not achieved while delivery greater than 99% dramatically decreases number of doses needed for cure (Table 1). In this analysis, we include HIV and HSV, which have lower numbers of total body latently infected cells (high estimates are 107 and 106, respectively) [55], [56], to highlight that large infectious burden necessitates considerably more doses for elimination of HBV than HIV or HSV (Table 1).
This analysis highlights the importance of high vector to target cell ratio, even under favorable assumptions regarding intracellular pharmacodynamics. Because the value of parameter m will be known as a function of dose, the key unknown parameter of delivery is σ, the proportion of vectors that enter target cells and are transduced.
Two factors will drive outcome of an infected cell following delivery of transgene-carrying vectors: the number of viral vectors transduced in the cell and the strength of the enzyme-substrate interaction. The critical biophysical interactions are the binding affinity between enzyme and substrate, the efficiency of enzyme cleaving following binding and the efficiency of precise DNA repair. These processes are captured indirectly with constant d in the formula λo = 1/(1+(v/d)) where v is number of vectors transduced in the cell and λo is probability that the genome will remain uncleaved. In this formula, d is scaled according to vector gene expression value per cell under the assumption that intracellular enzyme concentration is directly proportional to v [44]. The value of d determines whether one or multiple vectors will need to be delivered to the nucleus to ensure terminal mutation of most viral episomes. If d<<1, then transduction of one vector is likely to predict episomal cleavage. Alternatively, if d>1, then multiple vectors per nucleus will be necessary to disrupt all viral DNA.
A possible hurdle to disruption of latent genomes is resistance to the cleavage enzyme in question. cccDNA molecules may contain pre-existing mutations. The HBV mutation rate is relatively low [57], and pre-existing mutations to cleavage enzymes are likely to be relatively rare. DNA cleavage enzymes may also induce de novo mutations that render the site resistant to subsequent enzyme binding but do not incapacitate the virus. For example, if an enzyme repair event results in a 3 base pair mutation within the open reading frame, the ensuing loss of a single amino acid may theoretically not impair activity of the viral protein. However, if the DNA cleavage site is no longer recognized by the cleavage enzyme then this site has effectively become “enzyme resistant”. Presumably, this process will occur at a relatively low rate. For each cleavage and mutation event, the maximum probability of resistance is 33% as a deletion or insertion with a multiple of 3 is a pre-requisite for this event. However, addition or removal of one or several amino acids from the viral gene product will prove fatal to the virus on most occasions. Based on preliminary data using a target site in the N-terminus of a green fluorescent protein in a non-functional region, an absolute upper possible estimate is that ∼5% of cleavage/mutations events may result in de novo resistance [11], though we expect the actual rate to be considerably lower. If probability of cleavage is Pc = (1−λo), then the probability of resistance is Pr = Pc * Ψ where Ψ is the frequency of induced mutations that prevent further enzyme binding despite being non-lethal to the viral episome. Therefore, in our model, development of resistance is assumed to increase proportionally with amount of DNA cleavage. To isolate the more important effects of induced de novo mutations, we include no pre-existing mutations in model simulations.
Most persistent HBV exists as multiple non-replicating episomes within infected cells. For this reason, outcomes for a cell with intra-nuclear cleavage enzyme expression include only partial inactivation of genomes, as well as development of de novo resistance to cleavage enzymes in some but not all remaining viral molecules. The number of possible transition states of an infected cell following delivery of a vector is a function of the number of genomes within the cell (Fig. 2a): all or a portion of episomes can be disrupted by DNA cleavage, while all or a portion of disrupted episomes can develop de novo resistance. Each transition state has a certain probability following delivery of a certain number of delivery vectors, including the probability that the infected hepatocyte will undergo no change in its state. In general, development of resistance is less common than successful disruption and elimination of viruses (Fig. 2a). The total number of cells undergoing each transition is estimated by multiplying individual transition probabilities, by the number of cells with a certain number of cccDNA molecules, and amount of vector delivered.
Enhanced cooperative binding between HIV directed antiviral agents and their multivalent viral enzyme targets has been demonstrated as a key determinant in antiviral agent potency. For example at equivalent drug concentrations, HIV protease inhibitors can be 100,000 times more potent than HIV nucleoside reverse transcriptase inhibitors [20]. Similarly, enzyme binding to a single viral episome may enhance or impair binding of subsequent enzymes to neighboring episomes in the nucleus (Fig. 2b). Moreover, if multiple enzymes that target distinct genomic regions within a single cccDNA episome are dosed simultaneously, then there may be enhanced or impaired binding to these multiple episomal sites (Fig. 2c).
The mechanism to determine whether cooperative binding is present is generation of log-converted dose response curves with a particular emphasis on the slope of the curve, which translates to Hill coefficient (h*z) in the formula λo = 1/(1+(v/d)h*z). Parameter h represents enhanced binding of one enzyme product to multiple intranuclear episomes (Fig. 2b). A value of parameter h greater than one implies positive cooperative binding and will favor cleavage of multiple episomes (Fig. 2b), while a value less than one implies binding competition and will favor cleavage of only a single episome per transduction event. Under extreme conditions of negative binding cooperativity, the number of gene therapy doses will need to be equivalent to the maximum number of genomes per cell.
Parameter z represents the possibility of enhanced or impaired binding of multiple enzymes products to one viral genome at separate binding sites (Fig. 2c). If only one episomal DNA sequence is targeted, then z = 1 and the Hill coefficient is reduced to parameter h alone. The presence of multiple cleavage enzyme targets may be necessary to avoid resistance: while only one successful cleavage event will usually be required to neutralize replication activity of the episome, if cleavage at a certain site induces de novo resistance, then a different enzyme will need to bind a separate site to terminally disrupt the episome (Fig. 3). Under this set of rules, enhanced binding to secondary sites may prove advantageous. An episome that becomes resistant to all available enzyme products and maintains replicative capacity is termed fully resistant (Fig. 3). To reflect that parameters h and z may have opposing or complementary effects, they are included as a product in the equation. Also of note, parameter z may take on different values for different enzymes that are concurrently dosed, though for the purpose of theoretical simulations, we assume a single value.
Assuming that potency of a single DNA cleavage enzyme (z = 1) on an individual cccDNA episome level is captured with the equation λo = 1/(1+(v/d)h) where λo is probability of the episome remaining uncleaved, total cleavage enzyme activity within a single cell is represented by Pc(i) = (Si) * (1−λo)i * (λo)(S−i) where a cell has S enzyme susceptible cccDNA genomes and Pc(i) represents the probability of cleaving i episomes. At high levels of v/d, the probability of cleaving all episomes within a cell, or (1−λo)S, increases. Resistance to cleavage enzymes occurs as a function of cleavage events. Given i cleaved episomes within a cell, k episomes will become resistant according to formula: Pr(k) = = (ik) (Ψ)k(1−Ψ)(i−k).
To synthesize these concepts for HBV infection, we created a three-dimensional matrix. This model tracks total number of cells occupying different states over time. Between cleavage enzyme doses, the numbers of cells with every possible combination of replication competent enzyme susceptible (S) and enzyme resistant (R) genomes are measured. A third dimension is incorporated following each infusion of therapy, and accounts for different doses of vector transduction: each item within the matrix represents the total number of infected cells with a certain value for S, R and v. Transition probabilities are calculated for each cell according to Pc(i) and Pr(k). The matrix is updated accordingly following each dose (Fig. 2a).
Initial data suggest that enzyme activity and DNA mutations accrue over a week following vector delivery [11]. In practice, delayed enzyme activity following vector entry into target cells would prove problematic only if cccDNA levels reconstitute in a meaningful way during the time period between doses. Otherwise, dosing interval can simply be prolonged to wait for enzymes to exert their full effect, and this would not impair the probability of therapeutic efficacy. In the simulation model, for simplification purposes, DNA cleavage is assumed to occur instantly following delivery of vectors with a dosing interval of one week. In later model realizations, the possible effects of cccDNA reconstitution and slower enzyme onset are explored.
Strategies to bypass enzyme resistance will be analogous to those employed for antiviral therapy, namely design of cleavage enzymes that target separate regions within episomal HBV cccDNA (Fig. 2c, 3). Several possible dosing schemes exist (Table 2). Smaller vectors such as AAV can probably only carry 1–2 open reading frames, though different serotypes can theoretically be given with each successive dose with the goal of avoiding a strong humoral immune response. If AAV is employed, then multiple enzymes that target separate sites must be divided between separate vectors. These vectors can be dosed concurrently (thereby decreasing the delivery dose of each vector/enzyme combination). While this strategy will theoretically increase the proportion of genomes targeted with two enzymes, the overall number of eradicated genomes may decrease due to overlapping targeting within the genome leading to a lower overall number of targeted episomes: we term this hypothetical problem “antagonistic potency” (Fig. 4).
While a very high effective fMOI (>10 in Table 1) may overcome antagonistic potency, another approach would be to dose separate cleavage enzymes within AAV successively rather than concurrently. With sequential delivery of enzymes targeting different regions, the vector delivery equation remains unchanged. The equation λ0 = 1/(1+(v/d)h) again describes the probability of a genome remaining uncleaved. If q enzymes are available, then all remaining replication competent episomes are assumed to remain sensitive to subsequent doses through the first q doses (assuming a different enzyme is used with each dose). In other words, resistance to the first delivered enzyme will not impact activity of the second enzyme and so on. Transitions are mediated by Pc(i) = (Stoti)(1−λ0)i(λ0)(Stot−i), where Stot is the number of total episomes in a cell (either susceptible or resistant to prior delivered enzymes). Enzyme resistance is again captured with Pr(k) = (ik) (Ψ)k(1−Ψ)(i−k) and generation of single and multiple mutants is tracked following each dose.
Each cell within the liver may harbor different numbers of episomes with zero, single and multiple resistant sites (Fig. 5). We add a new dimension to the matrix with each delivery of a new cleavage enzyme such that the matrix contains q+2 dimensions given q total enzymes. For instance, a simulation with q = 3 (3 sequentially dosed enzymes with different DNA target sequences) will include the following dimensions: S (non-resistant episomes), R1 (single resistant episomes), R2 (double resistant episomes), R3 (triple of fully resistant episomes) and v (vectors). Transitions to a newly resistant state are mediated by prior resistant state of the episome: with development of de novo resistance, S transitions to R1, R1 transitions to R2, and R2 transitions to R3. Stot, defined above, is the sum of S, R1 and R2.
The model output is constructed in one of two ways: either the number of episomes with any resistance (Stot−S) are plotted against number of fully susceptible episomes (S); or the number of episomes with total resistance to all episomes (SRtot) are plotted against number of remaining episomes without total resistance (Stot−SRtot). Only after q doses are given is it possible to have SRtot>0 due to totally resistant episomes to each of the q available enzymes. Following q doses, our model assumes repeated dosing of the finally dosed enzyme, as the least number of resistant episomes will exist against this enzyme.
If vectors such as ADV with higher gene payload capacities are utilized, then two or more separate enzymes can be delivered and transduced within the same vector. This approach has the theoretical advantage of increasing per cell dose of cleavage enzyme, and has the potential to increase the proportion of targets that receive multiple cleavage enzymes, a process we term “synergistic potency” (Fig. 4). Moreover, if z>1 due to enhanced binding cooperativity between enzymes (Fig. 2c), then cccDNA cleavage will be augmented in this fashion as well. Unfortunately, ADV is highly immunogenic and may only achieve high delivery following the first dose, and would need to be followed with AAV or other smaller delivery vectors, or ADV of different serotypes.
Our model allows for analysis of potential benefits gained from delivery of multiple transgenes within a single ADV vector. The delivery equation is unchanged from prior simulations, as the number of vectors and therefore proportion of cells with no vector transduction (Pv = 0) remain the same. If a vector carries q enzymes, then intracellular concentration of cleavage enzyme increases by a factor q (Fig. 4). We isolate the compounded effects of multiple enzymes, as well as the possible accrual of multiple enzyme resistant mutants by sequentially evaluating the activity of individual enzymes within a cell using Pc(i) = (Stoti)(1−λ0)i(λ0)(Stot−i), where Stot is again equal to the number of total episomes in a cell (either susceptible or resistant to prior evaluated enzymes). Resistance is captured with Pr(k) = (ik) (Ψ)k(1−Ψ)(i−k) and generation of single and multiple mutants is tracked following each dose. The matrix again contains q+2 dimensions. The critical difference between sequential dosing and multiple enzyme delivery simulations is that for the latter, delivery is not updated between successive evaluation of enzyme activity. Only after all of the q enzymes are evaluated, do we sum the number of totally resistant (SRtot), partially resistant (Stot−SRtot−S) and susceptible (S) episomes in liver cells to update infectious burden within the entire liver.
To demonstrate characteristics of the model, we conducted simulations under different assumptions of vector delivery (fMOI), enzyme-substrate binding avidity/cleavage efficiency (binding dissociation constant or d), and cooperative binding of enzymes to multiple episomes (Hill coefficient or h). Initial simulations assumed a single transgene per vector and ignored de novo resistance. Pre-therapy conditions assumed fully suppressive antiviral therapy, a median of 5 episomes per cell, no inherent decay of infected cells or HBV cccDNA over time, and a total of 10 weekly doses. We defined infected cells as any cell with at least one remaining replication competent HBV cccDNA molecule. In initial simulations, we also assumed that the effect of each dose occurred instantaneously.
First, we performed a multi-parameter sensitivity analysis with parameter values drawn randomly from a pre-determined wide range (fMOI 0.5–5, binding dissociation constant 0.008–5, and Hill coefficient 0.2–5) using Monte Carlo selection methods. We generated 200 parameter sets and simulated the model to identify parameter effects on therapeutic outcome. Increasing fMOI (R2 = 0.50), and decreasing binding dissociation constant (R2 = 0.24) predicted lower remaining numbers of infected cells to a greater extent than increasing the Hill coefficient (R2 = 0.03).
To obtain a more mechanistic understanding of how model parameters interact to impact the extent of episome disruption, we created 80 parameter sets derived from 4 possible values for fMOI, 4 possible values for the Hill coefficient, and 5 possible values for binding dissociation constant. Model simulations were stochastic but produced equivalent results for repeat experiments with each parameter set. At low fMOI (m*σ = 0.5), decreasing the dissociation constant and/or increasing the Hill coefficient only allowed for a slight relative decrease in number of infected cells following 10 doses; at higher levels of vector delivery, each 5-fold decrease in the dissociation constant (change in color in Fig. 6a) resulted in a substantial decrease in infected cells following 10 doses. Increasing the Hill coefficient from 1 to 5 had a similar effect (change in shape in Fig. 6a), though this effect was absent at the highest simulated dissociation constants (all red lines in Fig. 6a), because a threshold of intracellular enzyme density was not surpassed to allow enhanced cooperative binding. At high fMOI and very low dissociation constants, episome binding saturated with or without the presence of enhanced cooperative binding (blue line under fMOI = 5 in Fig. 6a). Residual replication competent genomes during simulations with low dissociation constant and high binding cooperativity resulted from lack of vector delivery to a subset of cells (fMOI = 0.5 or 1.0) rather than lack of enzyme activity within infected cells.
If we assumed that humoral immunity removed an increasing proportion of vectors prior to delivery with each dose (successive decreases in parameter σ), then a greater number of cells retained replication competent episomes following 10 doses even with a potent regimen (Fig. 6b).
However, pre-treatment burden of infection as measured by median number of cccDNA episomes per cell prior to initiation of gene therapy, had only a small impact on remaining number of infected cells (Fig. 6c) and total replication competent episomes (Fig. 6d) following 10 equivalently potent doses of therapy.
If de novo enzyme resistance developed at a fixed rate per cleavage event and single enzyme therapy was assumed, then resistant genomes rapidly predominated following dosing with parameter combinations that would constitute potent regimens. If we assumed high delivery, avid enzyme – DNA substrate binding and positive binding cooperativity, and that the resistance rate was 5% or 1% per cleavage event, then only 2 or 3 doses were needed respectively prior to infected cells containing resistant genomes becoming the predominate infected cells. In addition, the set point of number of cells with resistant genomes was >0.5 log higher with an assumed resistance rate of 5% versus 1% (Fig. 7a). More potent regimens lead to more rapid predominance of cells with resistant HBV cccDNA but if enough doses were given, the set point of number of infected cells with resistant HBV was equivalent between more and less potent regimens with lower fMOI and higher dissociation constant, assuming equal probability of resistance per cleavage event (Fig. 7b). With potent regimens and a resistance rate of 5%, cells with multiple HBV episomes harbored a combination of susceptible and resistant forms, though many cells developed multiple resistant episomes, even after a single dose (Movie S1). Therefore, a key parameter to deduce experimentally will be rate of resistant mutants generated per cleavage event.
To avoid cleavage enzyme resistance, we next considered sequential delivery of 1, 2, 3, 4, or 5 enzymes in separate, weekly doses. A new enzyme was given each week until no new enzymes remained (at the sixth dose for the 5 enzyme condition, for example). At this point, the final enzyme was repeatedly redosed. Simulations assumed favorable potency parameters and a resistance rate of 1%. The addition of extra enzymes increased the time until enzyme resistant forms predominated, and lowered the steady state of cells retaining replication competent HBV cccDNA by ∼0.5 log with addition of each enzyme (Fig. 7c). Simulations with multiple successive enzymes resulted in lower numbers of infected cells than simulations with a single enzyme (dotted lines Fig. 7a, red line Fig. 7c). Yet, high numbers of enzyme resistant episomes still remained even following sequential dosing of five different enzymes (blue line, Fig. 7c).
We next simulated trials with a single dose of a multi-payload vector such as ADV carrying 1, 2, or 3 transgenes concurrently under different assumptions of fMOI and cooperative binding of enzymes to multiple episomes. A favorable enzyme-substrate binding avidity/cleavage efficiency was assumed for each simulation. Results from simulations with 36 pre-selected parameter sets (all following a single dose with assumed resistance rate = 1%) show that total remaining cccDNA genomes decreased with increasing fMOI, and that maximizing the transgene payload (blue line, Fig. 8a) increased effectiveness under high delivery conditions, especially in the presence of positive cooperative binding (circles, Fig. 8a), or lower dissociation constant (not shown). Even under lower delivery conditions (fMOI = 2), increasing the number of enzymes per vector dramatically decreased the total burden of infected cells containing viral genomes with at least one de novo enzyme resistance mutation (Fig. 8b) as well as the total number of infected cells containing HBV cccDNA molecules that were fully resistant to all of the q available delivery enzymes (Fig. 8c). Concurrently delivered DNA cleavage enzymes therefore are predicted to exhibit synergistic potency and decrease both the overall burden of infection and de novo resistant genomes (Fig. 4). Delivery remained a critical parameter for HBV cccDNA disruption and at low fMOI, most remaining cccDNA episomes were susceptible to the cleavage enzymes (Fig. 8d). Alternatively, delivery of multiple enzymes generally decreased percent of remaining episomes that were resistant. Positive binding cooperativity between enzymes generally increased the proportion of enzyme resistant episomes by virtue of its overall positive impact on cleavage: a similar effect occurred with lowering the binding dissociation constant (data not shown).
If cccDNA levels reconstitute at a meaningful rate and several day intervals are required between doses to allow effects of DNA cleavage enzymes to accrue, then this may imply the need for more prolonged therapeutic courses. We therefore examined the effects of underlying dynamics of HBV cccDNA survival, as well as the possible delayed effects of DNA cleavage enzymes following target cell entry. Several factors may drive changes in levels of HBV cccDNA during suppressive antiviral therapy. Hepatocytes with HBV cccDNA molecules periodically die at a rate equivalent to that of an uninfected hepatocyte (Fig. 9a). Decay of individual episomes at a slow rate is possible (Fig. 9b) but has not been explicitly documented and may be counterbalanced by low-level replication despite antiviral therapy, which may also allow spread to uninfected cells (Fig. 9c). Indeed, many patients do not achieve full virologic suppression [6]. Finally, most evidence supports division of nuclear cccDNA between daughter cells during homeostatic proliferation (Fig. 9d) [33]: as a result, patients on antiviral therapy appear to have a slow decay in levels of cccDNA over time though this decline is not rapid enough for viral eradication [58]. A less optimistic assumption for the standpoint of achieving cure would be that episomes divide along with human chromosomal DNA during cell division, limiting cccDNA decay (Fig. 9e).
In all simulations in Fig. 10, enzyme dosing occurred every two weeks. However, enzyme activity was assumed to accrue continuously over a week rather than instantaneously. We first assumed baseline conditions with high potency (Fig. 10) with no change in cccDNA levels between doses. A simulation with homeostatic proliferation of cccDNA (Fig. 9d), revealed a marginally lower level of remaining viral episomes after 10 doses, while episomal death concurrent with hepatocyte death (Fig. 9a) augmented episomal decay more substantially. If poor control of cccDNA replication was assumed due to incomplete suppression by antiviral drugs (Fig. 9c), then therapy was less potent.
We describe mathematical models that aim to capture critical features of DNA cleavage enzyme therapy for eradication of HBV. Our results identify potentially critical parameters that will determine whether cure will be feasible with available vector cleavage enzyme constructs. In particular, successful vector delivery to the majority of target cells with each infusion, and favorable intracellular binding kinetics between enzymes and DNA target sites appear to be pre-requisites for successful regimens. Cooperative binding of enzymes between multiple episomal targets could also potentially limit the number of doses needed prior to cure, particularly if enzyme concentration in cells only marginally exceed binding coefficient values.
While multiple doses of gene therapy will likely be required for cure, the first dose appears to be particularly critical. In order to enhance potency and limit resistance, this dose should have a high vector to target cell ratio, and if possible, multiple enzymes should be packaged within each delivery vector. Sequential use of different enzymes appears to be another useful strategy to avoid de novo resistance if only low-payload delivery vectors such as AAV are available.
While our integrated therapeutic model is relatively complex, its individual components (vector delivery, intracellular pharmacodynamics, resistance) are quite manageable. In total, the model contains only five unknown parameter values including 1) proportion of vectors removed prior to entry into target cells, 2) enzyme-DNA binding coefficient, 3) vector-DNA cleavage dose response slope (Hill coefficient), 4) resistance rate per DNA cleavage event and 5) dose response slope within a single episome if multiple enzymes are present in the cell nucleus. Each of these parameter values can be identified via specific experimental approaches for all vectors and cleavage enzymes of interest, which will allow for testing and refining of the model.
Vector delivery to target cells is best estimated initially in animal model studies. Humanized mouse models of HBV hold promise for this indication [59], [60]. Flow cytometry of liver biopsy tissue can be employed to quantify proportion of target cells without vector delivery following different doses of vector; the effective multiplicity of infection (σ*m) can be back calculated using Pv(0) = [(σ*m)v * e−(σ*m)]/v!. This effective delivery dose will represent a fraction of the pre-determined vector to target cell ratio (m), which in turn will allow for an estimate of proportion of vectors lost prior to target cell entry (1−σ). Ultimately, these experiments will need to be conducted in humans, as the human immune response to delivery vectors cannot be predicted from animal models. However, animal model parameters will serve as useful initial estimates that may be used within a Bayesian framework to assist in human clinical trial design.
A critical caveat of the functional MOI (fMOI) is that the vector to target cell ratio assumed in parameter m is inclusive of all cells that may serve as targets for vector entry, rather than only HBV infected cells. If a particular delivery vector also efficiently enters other intrahepatic cells such as Kupffer cells, endothelial cells or cells in other organs, then the fMOI will decrease accordingly. In effect, these cells will serve as vector sponges and will decrease the probability of high vector delivery to infected cells containing HBV cccDNA. Therefore, vector receptor specificity is critical not only to avoid untoward toxicity, but also to ensure that precious vector is not wasted.
A key experimental goal should be to determine which enzymes achieve avid binding and DNA cleavage activity (low values for d) and positive cooperative binding (h>1 or z>1) to their DNA targets. Dose response curve slope and enzyme-substrate binding coefficients can be obtained from cell culture models of HBV cccDNA infection in which infected cell lines are exposed to delivery vectors dosed at different multiplicities of infections. Using high throughput sequencing of the DNA target site, it will be possible to measure the proportion of target genomes with terminally disrupted DNA for each vector dose. Experimental dose response curves can be tested against our models describing enzyme DNA binding kinetics. If multiple enzymes are delivered concurrently in a single vector, then similar curves can be used to assess cooperative binding between several sites within a single episome.
To obtain a conservative upper limit for resistance rate per cleavage event, it will first be necessary to identify cells with confirmed vector delivery and HBV cccDNA cleavage. One possibility is to sort for vector transduced cells that are HBV e antigen positive and then look for mutation events within the cleaved open reading frame. For the purposes of informing clinical trial dose design, this estimate will be useful to ensure that doses exceed predicted thresholds for viral persistence.
When all unknown parameter values are estimated and a model structure is selected that best represents available data regarding vector delivery, enzyme/DNA substrate kinetics and resistance rate, then it will be possible to design regimens that maximize probability of cure while limiting excess dosing and possible toxicity. While it will be necessary to characterize all available delivery vectors and cleavage enzymes prior to predicting likelihood of therapeutic success, certain strategies are promising based on in silico simulations. For instance, if multiple transgenes targeting different viral DNA regions can be packaged within the same delivery vector, at least during the first dose, this may augment potency and decrease resistance when compared to multiple transgenes split among vectors. Ensuring high delivery during the first dose will maximize this effect.
A key challenge will be measuring therapeutic outcome. For HBV, it is difficult to take serial quantitative measures of episomal reservoirs of infection. While active viral replication can be tracked with quantitative PCR, burden of quiescent viral episomes can only be assessed with liver biopsy and tissue quantitation of uncleaved HBV cccDNA using sequencing. Even a tiny number of latently infected cells may theoretically be enough to reactivate and repopulate the reservoir. Because serial biopsies are likely to be feasible only in animal models of infection, therapeutic efficacy will ultimately need to be evaluated with close clinical follow up after cessation of antiviral therapy. For this reason, we make conservative assumptions in our model, so that the dosing schedule exceeds the presumed threshold for cure.
While we have focused on eradication of HBV, our model is easily adjusted to account for potential cure of other chronic viral infections such as HIV or HSV-2. The burden and properties of non-replicating viral stores differ dramatically between HBV, HIV and HSV [10]. As such, each infection presents a unique set of challenges for eradicative approaches. While latent HIV integrates as viral DNA into the human genome, HIV-1 DNA is present in only ∼107 cells during chronic infection, typically as a single genome per cell [61], [62]. However, the HIV-1 reservoir may be anatomically difficult to target with delivery vectors; while memory CD4+ T-cells are the central population of cells within the latent reservoir, the possibility that other immune cells form important reservoirs has not been completely excluded and if target receptors on these cells differ, then they may serve as sanctuaries from therapeutic cleavage enzymes [63]. Finally, due to rapid HIV-1 intra-host evolution in the context of ongoing immunological pressure, the HIV-1 reservoir is populated with diverse quasispecies, which may lead to pre-existing resistance to certain cleavage enzymes [64]. Therefore, phylogenetic techniques may be necessary to explore for bottleneck effects if a majority, but not all viral strains, are eliminated following repeated dosing of DNA cleavage enzymes.
HSV latency exists within a relatively low number of neuronal cell bodies in either the trigeminal or dorsal root ganglia [56], which may represent a therapeutic sanctuary where delivery of vectors is poor. For HSV-2, sampling of the dorsal root ganglia, the site of latency, is not feasible. Close clinical follow up following gene therapy will be necessary to evaluate for cure. As with HIV-1, the possibility of re-infection will need to be considered using phylogenetic sampling of pre and post-treatment positive PCR samples, as inactivation may not ensure protective immunity from re-exposure.
In summary, we present a model to capture the effects of gene therapy with DNA cleavage enzymes for chronic HBV infection. The model helps identify key therapeutic parameters that will be necessary for cure, and outlines appropriate experimental steps to identify dosing regimens that are most likely to disrupt all latent viral DNA following a minimal number of gene therapy doses.
Simulations were performed on C++ and using Microsoft Excel.
|
10.1371/journal.pcbi.1006839 | The gradient of the reinforcement landscape influences sensorimotor learning | Consideration of previous successes and failures is essential to mastering a motor skill. Much of what we know about how humans and animals learn from such reinforcement feedback comes from experiments that involve sampling from a small number of discrete actions. Yet, it is less understood how we learn through reinforcement feedback when sampling from a continuous set of possible actions. Navigating a continuous set of possible actions likely requires using gradient information to maximize success. Here we addressed how humans adapt the aim of their hand when experiencing reinforcement feedback that was associated with a continuous set of possible actions. Specifically, we manipulated the change in the probability of reward given a change in motor action—the reinforcement gradient—to study its influence on learning. We found that participants learned faster when exposed to a steep gradient compared to a shallow gradient. Further, when initially positioned between a steep and a shallow gradient that rose in opposite directions, participants were more likely to ascend the steep gradient. We introduce a model that captures our results and several features of motor learning. Taken together, our work suggests that the sensorimotor system relies on temporally recent and spatially local gradient information to drive learning.
| In recent years it has been shown that reinforcement feedback may also subserve our ability to acquire new motor skills. Here we address how the reinforcement gradient influences motor learning. We found that a steeper gradient increased both the rate and likelihood of learning. Moreover, while many mainstream theories posit that we build a full representation of the reinforcement landscape, both our data and model suggest that the sensorimotor system relies primarily on temporally recent and spatially local gradient information to drive learning. Our work provides new insights into how we sample from a continuous action-reward landscape to maximize success.
| Whether a previous action is successful or unsuccessful is an important contributor to sensorimotor learning. Indeed, binary reinforcement feedback (e.g., reward) is sufficient to cause adaptation of hand aim during a reaching task, independent from error feedback [1, 2, 3, 4, 5, 6, 7]. It has been proposed that updating aim of the hand based on reinforcement feedback is model-free and occurs by sampling a continuous set of possible motor actions until one or more actions are found that improve task success [8, 9]. Sampling motor actions presumably allows the sensorimotor system to use information from the reinforcement landscape to drive adaptation.
Here we broadly define the reinforcement landscape as the mapping between all possible motor actions and the expected reward of those actions. In this context, the sensorimotor system can maximize expected reward by ascending the reinforcement landscape [10]. However, for a meaningful change in behaviour to occur there has to be an underlying process that either evaluates or accounts for whether one action is better than another. More specifically for learning to occur the sensorimotor system must account for the gradient of the reinforcement landscape, which defines the rate of change in the expected reward with respect to a change in motor action. Intuitively, the evaluation of different actions may be easier with a steeper gradient, as there would be a more salient change in the expected reward for a change in action.
The form of the reinforcement feedback influences the shape of the reinforcement landscape. Reinforcement feedback can be binary or graded, and can be provided deterministically [1, 11] or probabilistically [2, 5]. Binary reinforcement feedback signifies only whether the action was successful or unsuccessful [1, 2, 5]. Graded feedback varies the magnitude of positive feedback (reward) or negative feedback (punishment) as a function of motor action [11, 12]. Thus, the reinforcement landscape gradient can be influenced by the magnitude and or the probability of feedback. Another consideration when using graded reinforcement feedback is that humans form a nonlinear relationship between different reward (or punishment) magnitudes and their perceived value [13]. This nonlinear relationship could potentially influence how the sensorimotor system evaluates perceived changes in expected reward.
Movement variability is also thought to influence the gradient of the reinforcement landscape by creating uncertainty between intended actions and actual actions. That is, the expected reward can change depending on whether it is a function of the intended action or the actual action [10]. Further, greater movement variability has been linked to faster learning in reinforcement-based tasks as it promotes exploration of the reinforcement landscape [14, 15].
Here we designed two experiments to examine how humans adapt the aim of their hand when receiving binary reinforcement feedback. Specifically, we tested the hypothesis that the gradient of the reinforcement landscape influences sensorimotor adaptation. We manipulated the reinforcement landscape gradient by altering the expected reward (the probability of receiving reward) given the angular distance between the hand location and target. To maximize reward, participants had to update the aim of their unseen hand to a location that was not aligned with the visually displayed target. Importantly, we normalized the reinforcement landscapes to baseline movement variability on an individual basis. This normalization allowed us to assess the influence of the reinforcement landscape gradient on learning while accounting for individual differences in movement variability. We used binary reinforcement feedback to eliminate the potentially confounding nonlinear relationship between different magnitudes of reward and their perceived value.
We tested the prediction that a steep reinforcement landscape would lead to faster learning than a shallow landscape (Experiment 1). Building on these results, in Experiment 2 we used a complex reinforcement landscape where each participant’s initial action was positioned in the ‘valley’ between two slopes that had different gradients (steep and shallow) and rose in the opposite direction (clockwise or counterclockwise). We predicted that participants would ascend the steeper portion of the complex reinforcement landscape. Finally, we introduce a model that relies on binary reinforcement feedback to update the aim of the hand during a reaching task.
In Experiments 1 and 2, 120 participants performed 450 forward reaching movements (Fig 1A). For each trial they began at a starting position and attempted to pass their hand (unseen) through a virtually displayed target. We recorded reach angle, which was calculated relative to the line that intersected the visually displayed target and starting position, the moment their hand was 20 cm away from the starting position.
Participants began by completing 50 baseline trials, where no feedback was received on whether reaches were successful or unsuccessful. During the next 350 experimental trials participants received binary reinforcement feedback according to their randomly assigned reinforcement landscape (see Experiment 1 and Experiment 2). Like baseline, the final 50 washout trials were also performed without feedback.
We instructed participants to “hit the target”. We informed participants that no feedback would be received if they missed the target, and for each target hit 1) the target would expand, 2) they would hear a pleasant noise, and 3) they would receive monetary reward, such that they could earn up to $5.00 CAD.
To test the idea that the gradient of the reinforcement landscape influences sensorimotor learning, we manipulated the probability of receiving positive reinforcement feedback (i.e., reward) as a function of reach angle. In Experiment 1 we tested the idea that the gradient of the reinforcement landscape would influence the rate of learning. In Experiment 2 we tested the notion that the sensorimotor system would use gradient information from a complex reinforcement landscape to find the best of multiple solutions that improved performance.
We tested the idea that the gradient of the reinforcement landscape influences the rate of learning. We predicted that a steeper reinforcement landscape would lead to a faster learning rate.
Participants either experienced a steep reinforcement landscape (n = 40) or a shallow reinforcement landscape (n = 40). To control for direction, the probability of positive reinforcement (reward) rose either in the clockwise (Fig 1B; Eq 2) or counterclockwise direction (Eq 3). We created these landscapes by manipulating the probability of reward as a function of reach angle. The width of each reinforcement landscape, that is the probability of reward given reach angle, was normalized to baseline movement variability on an individual basis. This normalization ensured that participants in an experimental group (steep or shallow) experienced the same gradient for a particular landscape, irrespective of movement variability. This also allowed us to calculate the change in reward probability for a change in intended aim (Fig 1C, Eqs 7–9) across participants, as well as the optimal intended reach aim (θ o p t a i m) that maximized success (Eq 10).
Reach angles were normalized by baseline movement variability on an individual basis and expressed as a z-score. Further, to allow for visual and statistical comparison irrespective of the direction that the reinforcement landscape rose (clockwise or counterclockwise), we multiplied the normalized reach angles by −1.0 for all participants that experienced a reinforcement landscape that increased in the counterclockwise direction [5, 16].
Similar to others [17, 18], we found two subpopulations of participants in Experiment 1: learner and non-learners. When examining the histogram of final reach position (average normalized reach angle of the last 100 experimental trials), we found a bimodal distribution (S1 Data, S1 Fig). Based on this analysis, we found that a cutoff z-score of 1.0 did well to partition the bimodal distribution and separate the learners from the non-learners.
Fig 2A and 2B shows individual data from two participants. The participant experiencing a steep reinforcement landscape quickly changed their behaviour towards a reach angle that maximized reward (z-score between 3 and 6). The participant experiencing a shallow reinforcement landscape took comparatively longer to change their reaching behaviour. The difference in learning rates between these two participants is most evident during the first 50 experimental trials.
Fig 2C shows the average reach angle over trials for participants (learners) that experienced either a steep or shallow reinforcement landscape. To compare the rate of learning between these two groups of participants, we fit an exponential function (Eq 6) over the experimental trials via bootstrapping (see Methods for details). We were interested in the time constant of the exponential function, λ, which defines the rate of learning. The exponential bootstrap fit analysis was performed separately first with the data from the learners alone, and then again with all participants (learners and non-learners together). As hypothesized, we found that the participants experiencing the steep landscape had faster learning (i.e., a lower exponential function time constant, λ) than those experiencing a shallow reinforcement landscape (p = 0.012 learners only, p = 0.021 for combined learners and non-learners, one-tailed). Fig 2D shows the posterior probability distribution and cumulative distribution of the time constant λ given the reach angles of participants experiencing either a steep or shallow reinforcement landscape. The inset of Fig 2D shows the posterior probability distribution of the time constant difference between the two experimental groups, from which we calculated the p-values reported directly above. The direction of the reinforcement landscape, clockwise or counterclockwise, did not influence the rate of learning (p = 0.540, two-tailed).
We also found that participants who experienced a steep landscape were more likely to be classified as learners than those who experienced a shallow reinforcement landscape (p = 0.036, two-tailed; Table 1).
In this experiment we tested the notion that the sensorimotor system uses gradient information from a complex reinforcement landscape to find the solution that maximizes reward. The probability of reward was at a minimum for reaches toward mean baseline behaviour but increased at different gradients (steep or shallow) for reaches in either direction (clockwise or counterclockwise) away from the target. We predicted that a significantly greater number of participants would adapt their reach aim in the direction of the steeper gradient.
Two different reinforcement landscapes were used in this experiment: one landscape had a steep slope that rose in the clockwise direction and a shallow slope that rose in the counterclockwise direction (steep clockwise; n = 20; Fig 3A; Eq 4), and the other landscape had a steep slope that rose in the counterclockwise direction and a shallow slope that rose in the counterclockwise direction (steep counterclockwise; n = 20; Fig 3C; Eq 5). As in Experiment 1, for both reinforcement landscapes we calculated the probability of reward given intended aim (Fig 3B and 3D; Eqs 7–9), as well as the optimal intended reach aim (θ o p t a i m) to maximize reward (Eq 10).
Here we were interested in the frequency of participants that changed their reach behaviour in the clockwise or counterclockwise direction, depending on whether they experienced the steep clockwise or steep counterclockwise reinforcement landscape. We used the average of the last 100 experimental trials to classify the direction of their final reach behaviour. Final reach direction was classified to be counterclockwise (z-score ≤ −1.0), center (−1.0 < z-score < +1.0) or clockwise (z-score ≥ +1.0). This classification was done separately for those experiencing a steep clockwise or steep counterclockwise reinforcement landscape.
Fig 4A and 4B show the average reach angle of steep learners, shallow learners and non-learners for participants experiencing the steep clockwise or steep counterclockwise reinforcement landscapes, respectively. The steep and shallower learners in Fig 4A respectively look qualitatively similar to the steep and shallow learners in Fig 4B when reflecting either of these figures about its x-axis. The behaviour of the non-learners was less consistent based on whether they experienced the clockwise or counterclockwise landscapes, but there was a limited frequency of non-learners (n = 2 and n = 3, respectively).
As an additional classification, participants that had a final reach position corresponding to the direction of the steep slope, shallow slope or a central location, were deemed steep learners, shallow learners and non-learners, respectively. This was done separately for participants that experienced either the steep clockwise or steep counterclockwise reinforcement landscape.
For this experiment we predicted that participants would ascend the steeper gradient of their assigned reinforcement landscape. Specifically, we expected more participants who experienced the steep clockwise reinforcement landscape to have their final average reach angle to be classified as clockwise. Similarly, we expected participants who experienced the steep counterclockwise reinforcement landscape to have their final average reach angle to be classified as counterclockwise. Using z-score cutoffs of −1.0 and +1.0, we found that there were significant differences in the final average reach classification between participants who experienced a steep clockwise or steep counterclockwise reinforcement landscape (p = 0.010, two-tailed, Fig 4C). These results were robust to whether we used z-score cutoffs of ±0.5 (p = 0.016, two-tailed) or ±1.5 (p = 0.020, two-tailed) to classify final reach direction. Further, we found that the direction (clockwise or counterclockwise) did not influence behaviour in terms of whether a participant was classified as a steep learner, shallow learner or non-learner (p = 0.810, two-tailed). Thus, the direction of the reinforcement landscape had an effect on their final reach direction, but it did not impact the frequency of steep learners, shallow learners, and non-learners.
Here we introduce a learning model that predicts reach angle (θn) on a trial-by-trial basis (Eq 1). This model takes the form
θn=N(θ¯naim,σn2)(1a),θ¯n+1aim,σn+12={ θ¯naim+α(θn−θ¯naim),σm2r=1θ¯naim,σm2+σe2r=0 (1b),(1c),
where n and n + 1 represent the current and next trial, respectively.
The model considers whether the current reach angle was successful (r = 1) or unsuccessful (r = 0). The model explores small regions of the workspace in a natural way via movement variability. Here, the variance of movement variability on the current trial (σ n 2) is a function of motor (execution) variance (σ m 2) after a successful reach, and the addition of both motor variance and exploratory variance (σ e 2) after an unsuccessful reach [2]. It was assumed that the variance of movement variability follows a Normal distribution N ( θ ¯ n a i m , σ n 2 ) [19, 20, 21], where θ ¯ n a i m represents the intended reach aim on the nth trial.
Inspired by Haith and Krakauer (2014) [22], the only action cached in memory is related to the location of the last successful reach. That is, an update in the intended reach aim (θ ¯ n a i m) occurs only after a successful reach. Specifically, this update is some proportion (α) of the difference between the current intended aim (θ ¯ n a i m) and the location of the last successful reach (θn). After an unsuccessful reach, the intended aim remains the same (i.e., θ ¯ n a i m is still stored based on the last successful reach) but the subsequent movement has greater variance (σm + σe). This results in a similar formulation to the equation just recently published by Therrien and colleagues (2018) [23]. There are some slight differences between the present model and the Therrien et al. (2015, 2018) model in terms of how they update the intended aim following a successful reach [23, 24] (see Discussion). Nevertheless, in the following we show the utility of this class of model in terms of replicating several features of sensorimotor adaptation. As previously suggested by van Beers (2009) [25] and Zhang et al. (2015) [26], our model assumes that the nervous system has some knowledge of movement variability when updating intended reach aim. This allows for an estimated difference between intended aim and actual reach angle, despite the participants have no vision of their hand during trials.
Our model has three free parameters: α = 0.40 (unitless), σm = 0.81 (z-score), and σe = 0.90 (z-score). The initial guesses of σm and σe for the fitting procedure were made with a trial-by-trial difference analysis (S2 Data, S2 Fig) that we modified from Pekny et al. (2015). It is expected that σm is slightly lower than a z-score of 1, or baseline movement variability, since here we were interested in the movement variability on a single-trial and not the additive variance that results from repeatedly subtracting two successive trials (see S2 Data, S2 Fig for further details). We found the best-fit parameters using a bootstrap optimization fitting procedure using only the data from Experiment 1 (S3 Data).
With our learning model, we simulated 40 individuals experiencing the steep reinforcement landscape of Experiment 1, and then simulated another 40 individuals experiencing the shallow landscape.
We found that simulated individuals displayed similar trial-by-trial variance and rates of learning compared to the behavioural data (compare Fig 5A and 5B to Fig 3A and 3B). We averaged across the 40 simulated individuals in each condition (steep or shallow reinforcement landscape). The model did well to capture between-subject variance. Similar to the behavioural data, we also found the emergence of exponential learning curves (Fig 5C).
We then simulated 100, 000 individuals experiencing the steep landscape and 100, 000 individuals experiencing the shallow landscapes. Simulating a large number of individuals allowed us to numerically converge on the theoretical exponential learning curves produced by the model. We then averaged across simulated individuals in each group and fit an exponential function. The best-fit time constant, λ, of the exponential function for the steep and shallow reinforcement landscapes were 28.0 and 49.6, respectively. Both values fall within the 95th percentile confidence intervals of the corresponding behavioural data. (steep [10.7, 36.2], shallow [27.4, 102.1]; Fig 2D).
In S2 Data, S2 Fig we present a trial-by-trial analysis, as a function of reinforcement history, of both the model simulations and behavioural data. We show in S4 Data with model simulations that changing the initial reward probability of the shallow landscape has a marginal influence on learning rates.
Here we simulated Experiment 2 using our learning model (n = 100, 000 simulated individuals) by using the best-fit parameters obtained from the behavioural data in Experiment 1. To compare the model to the behavioural results, we combined the data from all participants in Experiment 2. This was accomplished by multiplying the normalized reach angles by −1.0 for participants that experienced the steep counterclockwise reinforcement landscape.
Fig 6A shows a histogram of the final reach angle of both the behavioural data and model simulations. We then used the same final reach direction classification for the model simulations that we used for the behavioural data. Based on these classifications, we found that the model produced a similar frequency of steep learners, shallow learners and, to some extent, non-learners as the behavioural data (Fig 6A and 6B). Further, we found that the model did well to explain reach angle over trials for these three different groups (R2 = 0.85; Fig 6B).
We also performed an analysis to explore the influence of reinforcement feedback during the initial periods of experimental trials. To this end, we calculated how a participant’s Nth success predicted their final reach classification. This was done separately for successful reaches made on the shallow (Fig 6C) and steep (Fig 6D) slopes of the complex reinforcement landscape. We found that if a participant had their 1st success on the steep slope that they would likely be classified as a steep learner (Fig 6D). Conversely, a 1st success on the shallow slope was not a good predictor of final reach classification (Fig 6D). However, a participant was likely to be classified as a shallow learner if their 15th success and beyond was on the shallow slope. As shown, the model and data were highly correlated with each other (R2 = 0.933 and R2 = 0.995, respectively). This analysis shows that the participants and model simulations were both heavily influenced by early exploration and gradient information when they experienced a complex reinforcement landscape.
Using the same set of best-fit parameters found from the data of Experiment 1, we replicated the results of Izawa and Shadmehr (2011) and our previous work [5] (see Fig 7A and 7B, respectively). In the study by Izawa and Shadmehr (2011), participants were only provided binary feedback if they hit a target region that was gradually rotated from a visual displayed target. In our previous work [5], cursor position was laterally shifted according to a skewed probability distribution and participants received binary feedback on whether the laterally shifted cursor hit the visually displayed target. In both these studies, participants had no vision of their hand or arm. We had our model experience the same reported conditions from both these studies. Our model did very well to capture average reach behaviour, between-subject variance, trial-by-trial movement variability as a function of reinforcement history (see [2]; S2 Data, S2 Fig), and suboptimality.
Here, we define suboptimality as approaching but not quite reaching the optimal behaviour that maximizes reward (i.e., x o p t m a x ( h i t s ) in Fig 7B). Suboptimality is often a feature of ‘greedy’ algorithms that place greater emphasis on locally optimal information rather than globally optimal information [27]. Our learning model would be considered a greedy algorithm since it samples from spatially local motor actions and updates its aim based on the last recent success. A greedy algorithm can lead to suboptimal performance in non-symmetrical landscapes (e.g., [5], Fig 1B and 1C) and complex landscapes with local maximums (e.g., Fig 2). Behaviourally, this was particularly evident in Experiment 2 where a relatively high proportion of participants (22.5%) performed suboptimally by ascending the shallow slope and having a final reach direction aligned with a local maximum.
Further motivated by the model of Haith and Krakauer (2014) [22], we ran simulations to examine how movement variability influences the rate of learning and whether our model could capture random-walk behaviour. There is some debate to whether movement variability is beneficial [14, 22] or detrimental [15, 28, 29, 30] when learning from error feedback, which to some extent may be explained by the consistency (entropy) of the environment [31]. Recent work has suggested that movement variability is important when learning from reinforcement feedback and can influence the rate of learning [14]. Here we manipulated both motor (σm) and exploratory (σe) contributions to movement variability when simulating the experimental conditions of Experiment 1. We found that increasing the variance of movement variability, either σm or σe, led to increased rates of learning for both the steep (Fig 8A) and shallow (Fig 8B) reinforcement landscapes. However, it should be noted that with different amounts of movement variability there may exist a trade-off between the rate of learning and the probability of reward.
In previous literature, random-walk behaviour along task-irrelevant dimensions has been attributed solely to error-based learning [32, 33, 34, 35]. In the study by van Beers and colleagues (2013), participants received error (visual) feedback when reaching to large targets (Fig 9D). They displayed random-walk behaviour (i.e., trial-by-trial correlations) along the task-irrelevant dimensions that had no bearing on task success. Here we tested whether reinforcement feedback can also lead to random-walk behaviour. To test this idea, we used our model to simulate the experiment van Beers et al. (2013). Critically however, we did not use error feedback as in the original study—instead we only provided binary reinforcement feedback to the model based on whether it had hit or missed the target. Interestingly, we found that random-walk behaviour along task-irrelevant dimensions also emerged from our model (Fig 9A, 9B and 9C). Thus, our simulations suggest that random-walk behaviour, at least in part, may be attributed to reinforcement-based processes.
Our model relies on updating intended reach aim by using only the recent success (temporally current information) based on sampling the reinforcement landscape via movement variability (spatially local information). Given the strong relationship between our model and the behavioural data throughout the simulations above, our results suggest that the sensorimotor system largely depends on temporally recent and spatially local information to update where to aim the hand during our reinforcement-based sensorimotor learning task.
We found that manipulating the gradient of the reinforcement landscape influenced sensorimotor learning. First, we found that a steep reinforcement landscape led to faster learning. Second, participants were more likely to adjust their aim in the direction of the steepest portion of a complex reinforcement landscape. Our learning model that relies on reinforcement feedback to update aim of the hand was able to replicate the results in Experiment 1 and predict the results found in Experiment 2. Taken together, our data and model suggest that the sensorimotor behaviour observed in our experiments does not necessitate a full representation of the entire reinforcement landscape (storing the expected reward for all possible actions). Rather, the majority of learning behaviour can be captured using temporally recent and spatially local information about actions and rewards.
Participants learned faster when they experienced a steep reinforcement landscape, compared to those experiencing a shallow landscape. To our knowledge this is the first work showing that the gradient of the reinforcement landscape influences the rate of learning. The present study may be distinguished from previous work showing that a graded reinforcement landscape can augment error-based learning [11, 12]. Here we show that the gradient of a binary, positive reinforcement landscape influences learning in the absence of error feedback.
Using a visuomotor rotation task, Nikooyan and Ahmed (2014) used both graded reinforcement feedback and error feedback to study their effects on learning. Participants moved a cursor which was rotated from the unseen hand as it moved away from a start position towards a virtual target. Participants performed the task either with or without error (cursor) feedback. They experienced a graded reinforcement landscape, such that the magnitude of reward changed with the angular distance of the hand from the target, according to either a linear or cubic function. The maximum reward magnitude occurred when the rotated cursor hit the target. Relative to learning using only error feedback, linearly and cubically graded reinforcement landscapes combined with error feedback accelerated learning. They also found differences in the amount of adaptation between participants who experienced only graded reinforcement feedback (without any visual error feedback) based on either a linear or cubic reinforcement landscape. However, these differences reversed in direction during the course of the experiment and, in some instances, opposed theoretical predictions from a temporal-difference (TD) reinforcement algorithm [11, 36]. These inconsistent findings may have been caused by not controlling for individual differences in movement variability [14] or the nonlinear relationship between different reward magnitudes and their perceived value [13].
In our experiments, we used binary feedback that always had the same magnitude of reward. This eliminated the nonlinear relationship between different reward magnitudes and their perceived value [13]. Further, we controlled for individual differences in movement variability, which can influence exploration and the rate of learning in reinforcement-based tasks [14, 15, 37]. Thus, our work is the first to our knowledge that has isolated how the gradient of the reinforcement landscape influences the rate of sensorimotor learning.
In our second experiment, each participant’s initial action was positioned in the ‘valley’ between two slopes that had different gradients (steep or shallow) and rose in opposite directions. As predicted, we found participants were more likely to ascend the steepest portion of a complex reinforcement landscape. While the majority of participants ascended the steep slope, several participants ascended the shallow slope. The probability of whether they would be classified as a steep learner or shallow learner seemed related to initial success on either the steep or shallow portion of the landscape. In particular, participants were very likely to be classified as a steep learner if their first successful reach was on the steep slope of the complex landscape.
Our learning model did well to capture trial-by-trial behaviour, between subject variability and exponential learning curves in Experiment 1. Using the same set of best-fit parameters found using Experiment 1 data, we then simulated Experiment 2. The model produced similar distributions of steep-learners, shallow-learners and, to some extent, non-learners. The model was also able to capture several aspects of learning reported in previous work [1, 2, 5].
As mentioned, the behavioural findings of Experiment 2 were well predicted by our learning model. Critically, our model does not build up a full representation of the reinforcement landscape. Rather, it relies on using movement variability for spatially local exploration and temporally recent reinforcement feedback to update intended reach aim. Considering that the model does not build up a representation of the reinforcement landscape and that it was highly correlated with the behavioural results, suggests that whether participants ascended up the shallow portion or the steep portion of the complex reinforcement landscape was largely due to movement variability and the probability of reward. As an example, a participant’s initial reach angle had an equal probability of being aligned with either the steep or shallow slope due to movement variability. However, a participant’s initial reach was more likely to be rewarded on the steep slope because of its higher rate of reward. Moreover, the further a participant ascended either the steep or shallow slope it became increasingly unlikely that future successes would promote them from descending a slope. In particular, the steep slope had a stronger effect of promoting participants to ascend since its reward rate was double that of the shallow slope. This is evident in Fig 6D, where both the participants and model simulations were very likely to be classified as a steep learner when they had their 1st success on the steep slope. Conversely, final reach classification for both the participants and model simulations only became reliable after approximately the 15th success on the shallow slope (Fig 6C). Thus, participants and the model were more likely to be initially rewarded on the steep slope and also more likely to ascend the steep slope. Taken together, our behavioural results and model simulations support the idea that the nervous system does not build up a representation of the reinforcement landscape. Rather, the nervous system seems to rely on spatially local movement variability for exploration and temporally recent reinforcement feedback to update hand aim. Importantly, our findings also suggest that early exploration is highly influential when attempting to avoid local maximums and discover a global maximum.
Several hallmarks of motor learning simply emerged from our phenomenological learning model. Specifically, we found that the model produces exponential learning curves, between-and within-subject movement variability, suboptimal performance, increased learning rates with greater movement variability, trial-by-trial variance given a successful or unsuccessful reach (S2 Data, S2 Fig), reduced variability when hand aim approaches the optimal solution to maximize success, and random-walk behaviour in task-irrelevant dimensions. To our knowledge, random-walk behaviour has only been previously associated with error-based learning [32, 33, 34, 35]. Future work should examine whether random-walk behaviour can be replicated with experiments involving only reinforcement feedback.
The model of Haith and Krakauer (2014) [22] and the recently published model of Therrien and colleagues (2018) [23] would also be able to reproduce the rich set of behavioural phenomena mentioned in the above paragraph. These two models also rely on movement variability for exploration and caching a single aim direction that can be updated based on recent feedback. The Haith and Krakauer model stems from a Markov chain Monte Carlo (MCMC) algorithm and relies on sampling different motor actions. Actions are drawn from a probability distribution with a previously cached action acting as the distribution mean. If a recently experienced action is deemed less costly and or more rewarding than the previously cached action, this recent action becomes the newly cached action. Although this model was demonstrated with error-based tasks (i.e., visuomotor rotation and force-field adaptation), it could be extended to update hand aim using reinforcement feedback. As mentioned above, the work of Haith and Krakauer (2014) [22] and Pekny et al. (2015) [2] provided the motivation for our model. This resulted in a similar set of equations as recently proposed by Therrien and colleagues (2018) [23], albeit with some slight differences in terms of how the model updates hand aim. In their model, updating hand aim relies on the assumption that the sensorimotor system has perfect knowledge of additional exploratory movement variability following an unsuccessful reach and partial knowledge of the motor (execution) variability following a successful reach. Conversely, our model assumes that the same proportion of motor and exploratory movement variability are known by the sensorimotor system when updating hand aim. While some studies have explored the idea that the sensorimotor system has some awareness of movement variability [25, 26], to our knowledge no study has explored what proportion of movement variability is known by the sensorimotor system following a successful or unsuccessful reach. Nevertheless, our present work highlights the utility of this class of models, which rely on movement variability for exploration and caching a single action, to predict sensorimotor adaptation.
Emergent behaviour and simplicity are perhaps the most attractive features of our learning model. The model uses movement variability to sample the reinforcement landscape locally, and temporally recent information to update where to aim the hand. These features distinguish our model from several mainstream reinforcement algorithms in the motor literature that rely on building a full representation of the reinforcement landscape [1, 11, 37, 38]. The explicit goal of these algorithms is to maximize reward. For many of these reinforcement learning models, exploration and maximizing reward is accomplished by selecting actions using a soft-max function that considers the expected value of all possible actions. In general, such models rely on a large number free parameters and assumptions. Depending on the task and the discretization of considered actions and states, storing a representation of the reinforcement landscape in real-world situations could require vast amounts of memory and may be implausible. In comparison, our model (similarly, [22, 23]) has a small number of free parameters, makes few assumptions, implicitly maximizes reward, and uses minimal memory.
Our learning model does well to capture several aspects of behaviour during learning. For the model to adapt however, there has to be a non-zero gradient within the range of naturally occurring movement variability. Thus, the model is limited to small areas of the workspace. It has been shown in previous studies that participants are unaware of a change in aim when operating over small areas of the workspace [1, 39]. In our task, the average change in behaviour was ∼ 7.0 degrees, suggesting that the participants in our experiments were also likely unaware of the small shifts in reach angle [40]. Learning beyond these small areas of the workspace would likely also require active (cognitive) exploration strategies [41] and explicit awareness of the reinforcement landscape [17]. Nonetheless, our model did well to capture many features of sensorimotor adaptation over small areas of the workspace.
Behaviourally, we found that a steeper reinforcement landscape leads to faster learning. We also found that humans are more likely to ascend the steepest portion of a complex landscape. Our model was able to replicate our findings without the need to build up a representation of the reinforcement landscape. Further, several hallmarks of human learning simply emerged from this model. Taken together, our data and our model suggest that the sensorimotor system may not rely on building a representation of the reinforcement landscape. Rather, over small areas of the workspace, sensorimotor adaptation in reinforcement tasks may occur by using movement variability to locally explore the reinforcement landscape and recent successes to update where to aim the hand.
80 individuals participated in Experiment 1 (20.1 years ± 2.8 SD) and 40 individuals participated in Experiment 2 (20.5 years ± 2.8 SD). Participants reported they were healthy, right-handed and provided informed consent to procedures approved by Western University’s Ethics Board.
In both experiments, participants held the handle of a robotic arm (InMotion2, Interactive Motion Technologies, Cambridge, MA; Fig 1A) and made right-handed reaching movements in a horizontal plane. An air-sled supported each participant’s right arm while providing minimal friction with the desk surface during the reaching movements. A semi-silvered mirror blocked vision of both the participant’s upper-limb and the robotic arm, and projected images from an LCD screen onto a horizontal plane passing through the participant’s shoulder. An algorithm controlled the robot’s torque motors and compensated for the dynamical properties of the robotic arm. The position of the robotic handle was recorded at 600 Hz and the data were stored for offline analysis.
During both experiments, participants were exposed to one of several different reinforcement landscapes. We manipulated the gradient of the reinforcement landscapes by controlling the probability of positive reinforcement (reward) as a function of reach angle. These landscapes were constructed such that participants had to learn to change their reach angle, relative to baseline performance, to maximize the probability of reward.
The width of the reinforcement landscape experienced by a participant was normalized to the variability of their baseline reach angles. Reach angle was measured at the position where the robot handle first became 20 cm away from the center of the starting position, and was calculated relative to the line that intersected the starting position and the displayed target. The last 25 baseline trials were used to calculate their average baseline reach angle and the standard deviation of their angular movement variability. All reach angles were converted into z-scores. Specifically, reach angles were expressed relative to the average baseline reach angle and then normalized by the participant’s average standard deviation recorded during baseline. Thus, a z-score of 0.0 corresponded with their average baseline reach angle. A z-score of 1.0 or −1.0 indicated that a reach angle was ± 1 SD away from their average baseline reach angle in the clockwise or counterclockwise direction, respectively.
Defining the reinforcement landscape in terms of a z-score served two purposes. First, we controlled for slight differences in individual aiming bias by positioning all participants on the same location of the reinforcement landscape during the start of the experimental trials. Second, we normalized the width of the reinforcement landscape for each participant based on baseline movement variability, allowing us to isolate how the reinforcement landscape gradient influenced learning.
We performed data analysis using custom Python 2.7.11 scripts. For all participants in both Experiments, we recorded their endpoint reach angle for each of the 450 trials. Reach angles were normalized based on baseline reach behaviour, as described above, and expressed as a z-score.
Tests between means were performed using bootstrapped hypothesis tests with 1, 000, 000 resamples (Python 2.7.11) [5, 45, 46, 47]. Fisher’s exact test was used to test frequency tables (R 3.2.4). Coefficient of Determination (R2) was used to compare model simulations to behavioural data (Python 2.7.11). One-sided tests were used for planned comparisons based on theory-driven predictions. For all other comparisons we used two-tailed tests. Multiple comparisons were corrected for Type-I error using the Holm-Bonferroni procedure [48]. Statistical tests were considered significant at p < 0.05.
|
10.1371/journal.pcbi.1002315 | Two Birds with One Stone? Possible Dual-Targeting H1N1 Inhibitors from Traditional Chinese Medicine | The H1N1 influenza pandemic of 2009 has claimed over 18,000 lives. During this pandemic, development of drug resistance further complicated efforts to control and treat the widespread illness. This research utilizes traditional Chinese medicine Database@Taiwan (TCM Database@Taiwan) to screen for compounds that simultaneously target H1 and N1 to overcome current difficulties with virus mutations. The top three candidates were de novo derivatives of xylopine and rosmaricine. Bioactivity of the de novo derivatives against N1 were validated by multiple machine learning prediction models. Ability of the de novo compounds to maintain CoMFA/CoMSIA contour and form key interactions implied bioactivity within H1 as well. Addition of a pyridinium fragment was critical to form stable interactions in H1 and N1 as supported by molecular dynamics (MD) simulation. Results from MD, hydrophobic interactions, and torsion angles are consistent and support the findings of docking. Multiple anchors and lack of binding to residues prone to mutation suggest that the TCM de novo derivatives may be resistant to drug resistance and are advantageous over conventional H1N1 treatments such as oseltamivir. These results suggest that the TCM de novo derivatives may be suitable candidates of dual-targeting drugs for influenza.
| The influenza A subtype H1N1 (H1N1/09) pandemic raised public concerns due to drug resistance strains. Drug resistance occurs from conformational changes causing the original drug to lose binding ability and exhibit biological effects. The world's largest TCM Database@Taiwan was employed to screen for potential leads that simultaneously bind to H1 and N1. Three de novo compounds derived from Rosemarinus officinalis and Guatteria amplifolia were identified as having dual binding properties to H1 and N1. Structural analysis indicated that the candidates bind to multiple residues in both H1 and N1. In addition, the de novo derivatives were predicted as bioactive using four different computational models. The compounds are validated as potent dual targeting influenza drug candidates through multiple validations. Key advantages of the candidates include (1) binding to H1 and N1 through multiple amino acids, and (2) not binding to known mutation residues in H1 or N1. Such advantages can reduce drug resistance caused by single point mutations. On a broader context, features important for successful H1N1 drug development are discussed in hopes of providing starting templates for drug development and improvements.
| The first global pandemic of the 21st century was announced by the World Health Organization (WHO) in 2009 due to the worldwide spread of influenza A subtype H1N1 (H1N1/09) [1]. More than 214 countries have reported laboratory confirmed cases, and more than 18,449 deaths have been recorded [2]. Currently, the neuraminidase inhibitor Tamiflu® (oseltamivir) remains the primary drug prescribed to patients infected with H1N1/09 [3]. However, the emergence of drug resistant viral strains [4] and limited drug administration window [5] exemplifies the need for additional therapies.
Important constituents of influenza surface membrane proteins include hemagglutinin, neuraminidase, and the matrix protein 2 (M2) proton channel [6], [7]. Hemagglutinin mediates the binding of viral particles to host cell surface sialic acid and the invasion of viruses into host cell [8]–[10]. Neuraminidase is responsible for the cleavage of sialic acid residues to promote the release of progeny viruses [11], [12]. M2 proton channels are critical for viral mRNA incorporation into the virion and virus budding [13]. Over one hundred serological subtypes [14] have been identified through different combinations of the 16 hemagglutinin (H1–H16) and nine neuraminidase groups (N1–N9) currently known. The 3D-structure of M2 proton channels have recently been solved in both influenza A and B [15], [16], allowing more in depth studies regarding its biological function and action mechanism [17]–[19]. These proteins have been used as targets for rational attempts to design drugs for influenza [20]–[27].
The H1N1/09 virus strain is a triple reassortant that contains gene segments from avian, swine and human influenza viruses [28]. In addition to antigenic shift that can lead to fundamental changes in influenza surface antigens, antigenic drift could reduce binding affinity of host antibodies to antigens [29], [30]. A major challenge in influenza vaccine development is the rapid evolution of influenza viruses, causing vaccines to be easily outdated and reformulation necessary each year [31]–[33]. Although the H1N1/09 virus is susceptible to neuraminidase inhibitors, cases regarding oseltamivir-resistant viruses with neuraminidase mutation (such as H275Y) have been reported [34], [35]. Given that influenza viruses have RNA genomes that are prone to changes, it is imperative to devise new therapies. Much effort has been made to investigate the mechanism and devise alternative drugs against the drug-resistance issue of H1N1 [36]–[40]. Developing inhibitors that target both H1 and N1 antigens can reduce resistance issues resulting from the mutation of a single target antigen.
Computational approaches have been widely applied to molecular biology and medicine [41]–[50]. Structure-based methods, including docking and MD simulation, are invaluable tools in drug discovery and design. Computational docking is important for investigating ligand-protein interactions and elucidating binding mechanisms [51]–[57]. Since publication of the pioneer paper in 1977 [58], it has been established that low-frequency motions existing in proteins and DNA can help reveal dynamic mechanisms underlying fundamental biological functions [59]–[63]. NMR observation later confirmed such inferences and the findings were applied to medical treatments [64]–[67]. In recent years, application of molecular dynamics to investigate internal motions and biological functions of biomacromolecules has opened new frontiers. Vast amounts of information on molecular recognition and binding [68]–[71], conformations or conformational changes [72]–[75], molecular mechanisms of bioactivity and stability [76]–[79], and drug discovery [80]–[84] have been found. To understand interaction of drugs with proteins or DNA, consideration should be given not only to the static structures but dynamical information obtained by simulation through a dynamic process. In this regard, both docking and MD simulation were utilized in this study to provide comprehensive analysis protein-ligand interactions under static and dynamic conditions.
Much effort has been placed on developing new, effective influenza treatments, but most have focused on neuraminidase or M2 as the target protein [37], [38], [85]–[87]. To date, no hemagglutinin inhibitor is available. Traditional Chinese medicine (TCM) has been used extensively for finding effective drugs [88], and we have successfully designed novel medicinal compounds and identified potential drug leads through traditional Chinese Medicine Database@Taiwan (TCM@Taiwan) [89]. Preliminary studies conducted in this lab show potential for TCM compounds to serve as neuramindase and hemagglutinin inhibitors individually [90]–[95]. In view of the current needs for drugs effective against native and mutant H1N1/09 and our promising preliminary results, the present study integrates the concept of “dual targeting” with the aforementioned computational tools and TCM in the attempt to identify dual-targeting inhibitors of H1N1 that may be useful for drug development.
The experimental procedures and screening results after each filtering step are summarized in Figure 1. Among the 829 native TCM compounds, 81 docked into both H1 and N1 and were used for de novo evolution (Table S1). De novo compounds with dual binding capacities to H1 and N1 were ranked by combined DockScore and the top ten derivatives are listed in Table 1. Nine of the ten top ranking de novo compounds were derived from Rosmaricine, a natural compound isolated from Rosemarinus officinalis [96]. The remaining de novo compound was based on Xylopine, which is naturally found in Guatteria amplifolia [97].
The top three derivatives, Xylopine_2, Rosmaricine_14 and Rosmaricine_15, have in common a pyridinium addition to their native structure (Figure 2). The pyridinium addition could be the main explanation for higher DockScores of these three derivatives compared to their native compounds and the other derivatives. Rosmaricine_14 and Rosmaricine_15 differed by the number of fused rings, but the slight difference in DockScore suggests that addition of an acyclic ring has little influence on binding affinity.
Docking of the de novo compounds back to the receptor provides insights to modifications that can be made to modulate or enhance molecular properties and also highlights important protein-ligand interactions. When docked into the N1 protein binding site, Xylopine_2 interacts with Asp151 via a protonated amino group and has pi and hydrogen bond (H-bond) interactions with Trp179 and Glu228, respectively (Figure 3A). Rosmaricine_14 (Figure 3B) and Rosmaricine_15 (Figure 3C), have interactions with Asp151 and Arg293 via the carbonyl group and Glu228 through the 2-aminopyridinium group. Tamiflu® forms H-bond interactions with Arg156, Arg293 and Arg368, but not with Asp151 or Glu228 (Figure 3D). Both Asp151 and Glu228 have been reported as one of the major residues in the N1 ligand binding site [98], [99]. The ability of the de novo derivatives to form interactions with both Asp151 and Glu228 may account for the higher DockScores.
Binding of the top three de novo derivatives to H1 site is detailed elsewhere [92]. The ability to bind with important H1 residues Asp103 and Arg238 [100] indicates the dual targeting possibility of the candidates.
The top ranking model generated by genetic function approximation (GFA) includes the following descriptors: ES_Sum_dssC, CHI_3_C, Kappa_1, Jurs_PNSA_1, and Jurs_RPCS. Utilizing these five descriptors, the MLR model established for the neuramindase inhibitors is:
Correlation between the observed and predicted activities of the 27 ligands are shown in Figure 4A. All values were within the 95% prediction bands and the r2 value = 0.8043. The SVM model was constructed using identical molecular descriptors and ligands as the MLR model. The r2 value of the SVM model was 0.8605 and the correlation between observed and predicted activities of 27 ligands are illustrated in Figure 4B.
Table 2 summarizes the pIC50 values of Tamiflu® and the top three candidates as predicted by the generated MLR and SVM models. The predicted activity of Tamiflu® using the generated MLR model (pIC50 = 7.613) is similar to observed bioactivity values reported in the literature (pIC50 = 7.823) [101]. This indicates that the generated MLR model is a good prediction model. Predicted activity values using the SVM model indicate a lower pIC50 with regard to Tamiflu®. Nonetheless, both models indicate that all TCM de novo derivatives are good candidates with neuraminidase inhibitory activity. MLR and SVM models for predicting hemagglutinin inhibitory activity were not established due to the lack of available hemagglutinin inhibitor structures in the literature.
To further investigate docking features, CoMFA and CoMSIA models were built and validated using 27 neuraminidase inhibitors listed in Table S2. The PLS analyses results for CoMFA and CoMSIA models are shown in Table 4. The CoMFA model was generated using both steric and electrostatic fields and yielded a non-cross validated r2 value of 0.924 and a cross validated q2 value of 0.524 with an optimal number of components as 5. The optimal CoMSIA model (r2 = 0.937, q2 = 0.673, ONC = 5) consisted of steric and hydrophobic fields, H-bond acceptors and donors. When compared against actual observed activities [102], both CoMFA and CoMSIA models had good predictability, predicting pIC50 values that differed only marginally from the actual pIC50 values of 24 compounds (Table 5).
The validated CoMFA and CoMSIA maps were used to assess ligand bioactivity. Contour of the de novo compounds at 20 ns MD simulation to the relative spatial positions of CoMFA and CoMSIA feature maps are shown in Figure 11. In Xylopine_2, Rosmaricine_14 and Rosmaricine_15, the H-bond between the 2-aminopyridinium group and Glu228 matched the electropositive group feature of the CoMFA model (Figure 11A,11C,11E) and the H-bond donor feature in CoMSIA model (Figure 11B,11D,11F). The hydrophobic benzene structures of Xylopine_2 matched the steric favoring region of the CoMFA map and the hydrophobic feature of the CoMSIA map. The carbonyl groups in Rosmaricine_14 and Rosmaricine_15 which formed H-bonds with Tyr402 satisfied the H-bond acceptor feature in the CoMSIA model. Tamiflu® also contours to both CoMFA and CoMSIA models. The 3-methoxypentane group close to Arg293 and Asn344 matched the steric favoring region of CoMFA (Figure 11G) and the hydrophobic feature of CoMSIA (Figure 11H). This residue has similar characteristics to the 2-aminopyridinium group in the de novo derivatives. In addition, the N-methylacetamide group in Tamiflu®, which forms H-bond with Tyr402, is located near the H-bond donor feature in CoMSIA.
Though all compounds contoured to the N1 inhibitor features identified by CoMFA and CoMSIA, a critical difference was observed between Tamiflu® and the TCM de novo derivatives. All compounds except Tamiflu® formed H-bonds at Glu228. As Glu228 is a primary binding site of N1 [99], ability of the TCM de novo derivatives to maintain stable binding with Glu228 during MD simulation supports the potential of these compounds as drug alternatives to Tamiflu®.
Due to the lack of reported H1 ligand bioactivities in the literature, direct assessment of bioactivity through construction of CoMFA and CoMSIA models was not possible. Alternatively, indirect support was provided by assessing the ability of de novo derivatives to maintain contour to the N1 CoMFA/CoMSIA maps while forming interactions at key residues in H1, Glu83 and Asp103 [92]. As illustrated in Figure 12 the TCM de novo derivatives docked into the H1 binding site and formed critical interactions at Glu83 and Asp103 without losing contour to the CoMFA and CoMSIA maps. These results suggest that not only were the TCM de novo derivatives capable of docking into both H1 and N1, but that biological activity was also predicted in both binding sites, thus it is possible to develop dual-targeting drugs from the selected de novo derivatives.
Important features for potential H1 and N1 inhibitors are summarized in Figure 13. For H1, a salt bridge with Glu83 and H-bond donor and/or electrostatic interactions with Asp103 are important characteristics that should be met. Potential inhibitors for N1 should have salt bridge and/or H-bond formation at Glu228 and interactions with Asp293. These features can be used to identify or design novel drugs for H1 and/or N1. In the case of the TCM de novo derivatives from this study, each compound could structurally fulfill the requirements of both H1 (Figure 13A,13B,13C) and N1 (Figure 13D,13E,13F) binding sites, thus supporting their potential as dual-targeting compounds.
In this research, we identified Xylopine_2, Rosmaricine_14, and Rosmaricine_15 as the top three de novo derivatives exhibiting binding affinity to H1 and N1. Addition of a pyridinum residue to the native structures of xylopine and rosmaricine contributes to bond formation at key residues in both H1 (Glu83, Asp103) and N1 (Glu228, Arg292). The de novo derivatives were predicted as active by the SVM and MLR models, and contoured well to the 3D-QSAR models. The TCM de novo derivatives were able to maintain contour while forming key binding interactions in H1, thus providing indirect support for bioactivity in H1. The results of this study indicate that the TCM de novo derivatives not only can bind to, but can also exhibit biological activities in both H1 and N1.
Key binding locations of the de novo derivatives include Glu83 and Asp103 for H1, and Glu228 and Arg292 for N1. Mutations currently attributed to oseltamivir resistance are located at H275 and N295S of the NA [103]. Since the key binding locations of the TCM derivatives do not overlap with those causing oseltamivir resistance, derivatives will be able to bind to viruses that are currently resistant to Tamiflu®. In addition, the de novo derivatives do not bind to amino acids in H1 or N1 that are prone to mutation (Table 6, Table 7) [40], [104], thus would likely be able to exert activity across a range of mutant H1N1 viruses. Last but not the least, multiple bond formations observed in MD provide additional insurance against possible mutations at key binding residues. In the case of a single point mutation, the de novo compounds will remain bound to the H1 and N1 sites through another key residue, therefore resisting the development of drug resistance in the virus. Based on the results and observations of this study, the TCM de novo derivatives may be attractive compounds for designing novel dual-target inhibitors for H1 and N1.
Virtual screening, de novo derivative generation, and molecular dynamics (MD) simulation were performed using Discovery Studio Client v2.5.0.9164 (DS2.5; Accelrys Inc., San Diego, CA). The two-dimensional and three-dimensional structures of TCM compounds were generated using ChemBioOffice 2008 (PerkinElmer Inc., Cambridge, MA). Comparative molecular field analysis (CoMFA) and comparative molecular similarities indices analysis (CoMSIA) models were constructed using SYBYL© 8.3 package (Tripos Inc., St. Louis, MO).
Compounds from the TCM Database@Taiwan were docked to H1 and N1 protein active sites reported in our previous study [91]. All procedures were completed under the forcefield of Chemistry at HARvard Molecular Mechanics (CHARMm) [105]. The virtual screening process was performed using LigandFit. The conformational search method was based on the Monte Carlo algorithm. Rigid body minimization following initial ligand placement was completed using Smart Minimizer. Scoring functions used by LigandFit were DockScore.
TCM compounds that docked into both H1 and N1 proteins were selected and then ranked by the sum of their H1 and N1 DockScore. Tamiflu® was used as the control for N1, and its N1 docking score was set as the minimum requirement. The top TCM compounds that passed the filtering were selected for de novo evolution.
In de novo evolution, TCM compounds were placed into the H1 and N1 protein binding sites described previously, and Ludi-fragments were attached to the native structure. The new derivatives were generated in full evolution mode. Derivatives from de novo evolution were subjected to additional screening through Lipinski's rule [106] to rule out orally unstable or pharmacologically inapplicable compounds. As de novo products generated for H1 and N1 proteins differed, all de novo products were re-docked to H1 and N1 proteins to assess binding affinity. De novo products that docked into both H1 and N1 proteins were selected and ranked by the sum of their respective H1 and N1 DockScore. The top ten compounds with the highest DockScore were selected for further structure-based analysis.
The 27 neuraminidase inhibitors used, including 24 training set compounds and 3 test set compounds, were adapted from Zhang's study [102]. Compounds were drawn using ChemBioOffice 2008 (PerkinElmer Inc., Cambridge, MA) and modified to physiological ionization using the Prepare Ligand function in DS 2.5. Bioactivity values (IC50) were also obtained from Zhang's study though the original sources were not clarified, and converted to pIC50 (log(1/IC50)). Molecular descriptors of the compounds were calculated using Calculate Molecular Properties in DS 2.5 and the GFA was used to select the best representative molecular descriptors [107]. Utilizing the best representative molecular descriptors identified through GFA, MLR and SVM models were constructed using MATLAB (The Mathworks Inc., Natick, MA) and LibSVM [108], respectively, and used to predict the bioactivity of TCM de novo compounds.
The MD simulation was performed using the Molecular Dynamics package of DS 2.5. The complexes were created with a 10 Å solvation shell of TIP3 water around the protein. Sodium cations were added to each system for neutralization. Minimization using Steepest Descent and Conjugate Gradient were performed at 500 cycles each. Each protein-ligand complex was gradually heated from 0K to 310K over 50 ps, followed by a 200 ps equilibration phase. The production stage was performed for 20 ns using NVT canonical ensemble and trajectory frames were saved every 20 ps. SHAKE algorithm was applied to immobilize all bonds involving hydrogen atoms throughout the MD simulation. Long-range electrostatics were treated with PME method. Time step was set to 2 fs for all MD stages. The temperature coupling decay time for the Berendsen thermal coupling method was 0.4 ps. Post processing of the trajectory was performed using Analyze Trajectory module. Torsion angles of each bond were also monitored through DS 2.5. LIGPLOT [109] was used to generate schematic diagrams of protein-ligand interactions for each candidate and control in H1 and N1.
CoMFA and CoMSIA models were constructed through the partial least square (PLS) analysis using previously described neuraminidase inhibitors [102]. The optimal number of components was obtained from leave-one-out method to yield the highest r2 and q2 values in non-cross validation and cross-validation, respectively. Biological activities of the TCM de novo compounds were evaluated based on contour to the generated 3D-QSAR map.
|
10.1371/journal.pgen.1008352 | Increased vulnerability of nigral dopamine neurons after expansion of their axonal arborization size through D2 dopamine receptor conditional knockout | Parkinson’s disease (PD) is a neurodegenerative disorder characterized by the loss of dopamine (DA) neurons in the substantia nigra pars compacta (SNc). Rare genetic mutations in genes such as Parkin, Pink1, DJ-1, α-synuclein, LRRK2 and GBA are found to be responsible for the disease in about 15% of the cases. A key unanswered question in PD pathophysiology is why would these mutations, impacting basic cellular processes such as mitochondrial function and neurotransmission, lead to selective degeneration of SNc DA neurons? We previously showed in vitro that SNc DA neurons have an extremely high rate of mitochondrial oxidative phosphorylation and ATP production, characteristics that appear to be the result of their highly complex axonal arborization. To test the hypothesis in vivo that axon arborization size is a key determinant of vulnerability, we selectively labeled SNc or VTA DA neurons using floxed YFP viral injections in DAT-cre mice and showed that SNc DA neurons have a much more arborized axon than those of the VTA. To further enhance this difference, which may represent a limiting factor in the basal vulnerability of these neurons, we selectively deleted in mice the DA D2 receptor (D2-cKO), a key negative regulator of the axonal arbour of DA neurons. In these mice, SNc DA neurons have a 2-fold larger axonal arborization, release less DA and are more vulnerable to a 6-OHDA lesion, but not to α-synuclein overexpression when compared to control SNc DA neurons. This work adds to the accumulating evidence that the axonal arborization size of SNc DA neurons plays a key role in their vulnerability in the context of PD.
| Parkinson’s disease motor symptoms have been linked to age-dependent degeneration of a class of neurons in the brain that release the chemical messenger dopamine. The reason for the selective loss of these neurons represents a key unsolved mystery. One hypothesis is that the neurons most at risk in this disease are those with the most extensive and complex connectivity in the brain, which would make these cells most dependent on high rates of mitochondrial energy production and expose them to higher rates of oxidative stress. Here we selectively deleted in dopamine neurons a key gene providing negative feedback control of the axonal arbor size of these neurons, in the objective of producing mice in which dopamine neurons have more extensive connectivity. We found that deletion of the dopamine D2 receptor gene in dopamine neurons leads to dopamine neurons with a longer and more complex axonal domain. We also found that in these mice, dopamine neurons in a region of the brain called the substantia nigra show increased vulnerability to a neurotoxin often used to model Parkinson’s disease in rodents. Our findings provide support for the hypothesis that the scale of a neuron’s connectivity directly influences its vulnerability to cellular stressors that trigger Parkinson’s disease.
| PD is a neurodegenerative disorder primarily characterized by a massive loss of DA neurons in the SNc that is also thought to be accompanied by the loss of other types of neurons in a select subset of brain regions including the locus coeruleus and the pedunculopontine nucleus [1]. Canonical symptoms include a range of motor deficits, but PD patients also often suffer from non-motor symptoms including olfactory deficits and constipation. Inherited mutations in gene products such as Parkin, Pink1, DJ-1, α-synuclein, LRRK2 or GBA are found in approximately 15% of cases. These gene products are involved in basic cellular processes including mitophagy, oxidative stress handling, mitochondrial antigen presentation, vesicular trafficking and lysosomal function. One of the key unanswered questions in PD research is why alterations in such ubiquitous processes lead to selective degeneration of a select subset of neuronal populations in the brain including SNc DA neurons. A striking example of this selectivity is the much higher resilience of the neighboring DA neurons of the ventral tegmental area (VTA), which are far less affected than SNc DA neurons in PD [1]. In the last few decades, many hypotheses have been raised about the core characteristics of SNc neurons that are responsible for their large bioenergetic requirements and that could explain their selective vulnerability. These include, but are not limited to, pacemaking activity [2], high DA- and iron-related toxicity [3,4] and possessing a highly elaborate, long-range axonal arborization [5–8]. All these characteristics are thought to exert an important pressure on the capacity of these cells to efficiently produce energy and cope with the associated oxidative stress. In this context, any other subsequent cellular stresses associated with some of the genetic alterations mentioned above, as well as aging and exposure to environmental toxins could trigger the disease. We have previously showed in vitro that SNc DA neurons have a higher basal rate of mitochondrial oxidative phosphorylation and ATP production and a smaller reserve capacity compared with the less vulnerable DA neurons of the VTA, characteristics that appear to be the result of the highly complex axonal arborization of these neurons [8]. We therefore postulated that the size of this axonal arborization might be a significant contributor to the differential vulnerability of SNc and VTA DA neurons in PD.
Based on our previous work and on modelling of the impact of the axonal arborization size on energy requirements [9,8], it is possible that the relatively small size of the axonal arborization of mouse DA neurons compared to humans (10 fold smaller) could explain the apparently high resilience of mouse DA neurons and the associated difficulty to produce optimal animal models in this species. Indeed, mouse models with genetic deletions of the key genes found in familial forms of the disease generally do not present age-dependent neuronal loss [10–14]. If the smaller axonal arborization size of mouse DA neurons is a key limiting factor for their vulnerability, it might be possible to increase this vulnerability by increasing their axonal arborization size in vivo. In order to reach this objective and test our hypothesis, we generated mice with a conditional deletion of the DA D2 receptor in DA neurons (D2-cKO). Increased DA terminal density has been suggested to occur under chronic D2 antagonist administration [15,16] and in the constitutive knockout model of this receptor [17] and D2 agonists have been shown to reduce the density of axon terminals established by DA neurons [15,18]. Here we surmised that a cell-specific knockout of this receptor in DA neurons should lead to an increased size and complexity of the axonal arborization of these neurons and increase their intrinsic vulnerability.
We find that in the intact mouse brain, the axonal arborization size of SNc DA neurons is 3-fold larger than that of less vulnerable VTA DA neurons. We further demonstrate that in D2-cKO mice, the axonal arborization size of SNc DA neurons is 2-fold larger relative to control mice, a phenotype associated with impaired evoked DA release and increased vulnerability to 6-OHDA, but not to α-synuclein overexpression. This work provides strong evidence in favor of the hypothesis that the axonal arborization size of SNc DA neurons plays a key role in regulating their basal vulnerability in the context of PD.
If axonal arborization size is a critical determinant of the selective vulnerability of SNc DA neurons, a prediction is that the axonal domain of these neurons should be more arborized than that of the more resilient VTA DA neurons in vivo. Since there is no specific axonal marker to distinguish between SNc and VTA DA neurons, we injected a small amount of floxed AAV2-eYFP in the SNc or the VTA of adult DATCre/+ mice to label a few hundred (~300–1000) DA neurons from one or the other population (Fig 1A). For more examples of SNc targeted infections, see. S1 Fig. After immunolabeling and manual counting of the infected neurons, we quantified the extent of the related axonal arborization within the striatum using confocal imaging to systematically sample images from slices throughout the rostro caudal axis (Fig 1B). As expected, the majority of SNc DA neurons projections were found in the dorsal striatum and the majority of VTA DA neurons projections were found in the ventral striatum. We next extrapolated the arborization density obtained from each slice to the size of the striatum on that slice and normalized it by the number of infected neurons. Finally, we plotted the arborization area obtained as a function of bregma coordinates for VTA (Fig 1C) and SNc (Fig 1D) targeted infections. Comparing the extent of the total arborization revealed a 3-fold larger axonal arbour for SNc compared to VTA DA neurons (Fig 1E and 1F).
Because an increase in axonal arbour size could increase the vulnerability of DA neurons, we aimed at increasing the axonal arborization of DA neurons by selective genetic deletion of the DA D2 receptor. To do so, we crossed DATIRES-Cre mice with Drd2loxP mice and generated DATIRES-Cre/+; Drd2loxP/loxP mice as previously described [19]. Control mice were heterozygotes for Cre expression (DATIRES-Cre/+; Drd2+/+). In these D2-cKO adult mice, we examined the axonal varicosities of DA neurons by measuring TH and DAT immunolabeled structures using confocal imaging in the ventral and dorsal striatum (Fig 2A and 2B). We observed no change in the area covered by TH signal (Fig 2C), the TH mean signal intensity (Fig 2D) or total TH signal (Fig 2E) in any part of the striatum. However, we observed an increased area covered by the DAT signal in the dorsal striatum (Fig 2F) with an increased DAT signal intensity (Fig 2G), which resulted in a more than 2-fold increase in total DAT signal (Fig 2H). No changes were observed in the ventral striatum. This increased DAT signal in the dorsal striatum was not the result of changes in the number (Fig 2I) or size (Fig 2J) of striosomes and was not a result of an increased number of DA neurons in the SNc, VTA or retrorubral field (RRF), as determined by unbiased stereological counting (Fig 2K).
To confirm that this increased dorsal striatal DAT signal was the result of an increase in the axonal arborization size of SNc DA neurons, we again used conditional viral labelling to visualize the axonal domain of SNc and VTA DA neurons in D2-cKO mice. We observed a 2-fold increase in the axonal arborization size of SNc DA neurons in D2-cKO mice (Fig 3A), with no change for VTA DA neurons (Fig 3B). Comparing axonal arborization size of SNc and VTA DA neurons from control mice again showed a 3-fold difference between the two populations (Fig 3A vs 3B). To better characterize this expanded axonal arbour originating from SNc D2-cKO DA neurons, we next measured the level of colocalization of virally-expressed YFP with DAT or TH (Fig 3C). There was an increased colocalization of TH or DAT with the YFP-labelled axonal varicosities of D2-cKO mice and a general increased colocalization of TH and DAT inside these processes (Fig 3D). To evaluate if these new processes were likely to be functional and able to release DA, we measured the colocalization of VMAT2 with YFP (Fig 3E) and found it to be unchanged (Fig 3F). We also found an increased colocalization of VMAT2 and DAT inside these processes.
An increased density of dopaminergic axonal fibers in the striatum, as well as the genetic removal of the pre-synaptic D2R known to control DA synthesis and release, could lead to increased DA release. Alternately, the enhanced bioenergetic requirements associated with a broader axonal arbour could lead to impaired DA neurotransmission. To distinguish between these possibilities, we quantified DA release evoked by single electrical pulses in acutely prepared striatal brain sections from D2-cKO and control mice using fast-scan cyclic voltammetry (Fig 4A). We found that DA release was significantly reduced in the dorsal and ventral striatum (Fig 4B). However, this difference was greatly diminished following incubation with the DAT antagonist nomifensine (Fig 4C). This observation of a partial rescue with nomifensine, coupled with our observation of increased striatal DAT immunoreactivity (Fig 2C–2E) could imply that increased DAT function in D2-cKO mice was the cause of the reduced activity-dependent DA overflow. Alternately, as DAT blockers including nomifensine and cocaine have been reported to also promote DA release though other mechanisms [20–22], the apparent rescue could result from an enhancement of DA release and not reuptake. To distinguish between these two possibilities, we examined the kinetics of DA release. Comparing D2-cKO and control mice, we found no change in kinetics of DA reuptake (tau) or in the maximal rate of reuptake (Vmax) in the dorsal (Fig 4D), or ventral (Fig 4E) striatum, suggesting no robust change in DAT function in D2-cKO mice. To further address this issue, we also performed a surface biotinylation assay from the striatum of a separate cohort of control and D2-cKO mice and confirmed that there were no significant changes in surface DAT levels in the striatum in the absence of D2 autoreceptors (Fig 4F).
As an increase in axonal arbour size in D2-cKO SNc DA neurons is predicted to induce a larger bioenergetic burden on these neurons, we next examined their vulnerability in two different mouse models of PD: the α-synuclein viral overexpression model and the intra-striatal 6-OHDA model. AAV-mediated wild-type α-synuclein overexpression was achieved by stereotaxic injection into the mesencephalon (Fig 5A). Three months after virus injection, stereological counting revealed a loss of 25–35% of DA neurons in the SNc and RRF (Fig 5B and 5C), with no change in the number of non-DA neurons (Fig 5D) and no significant change in the VTA (Fig 5E). This cell loss in the SNc and RRF was not significantly different in D2-cKO mice compared to control mice. We also observed the presence of phosphorylated α-synuclein positive cell bodies (Fig 5A), a good indicator of the toxicity induced by the overexpression. In the dorsal striatum only (Fig 5F), we observed a small 20% decrease in TH signal area (Fig 5G) and total signal (Fig 5H) with no change in signal intensity (Fig 5I).
At the behavioral level, mice overexpressing α-synuclein only showed a modest increased preference for ipsilateral paw use (S2A Fig), with no change in the total number of steps and no difference between D2-cKO and CTL mice (S2B Fig). In the rotation test, neither basal nor amphetamine-induced rotational preferences were altered (S2C and S2D Fig), with amphetamine inducing an expected increase in total number of rotations (S2E Fig). These finding are in keeping with the modest level of cell loss and striatal denervation in this model.
We next examined the vulnerability of DA neurons using a second, different model of PD using the DA neuron-specific toxin 6-OHDA. Unilateral injection in the dorsal striatum at a low dose (1.5μg in 0.5 μL) (Fig 6A) was performed in order to produce a partial loss of dopaminergic cell bodies (Fig 6B). In control mice, one month after the 6-OHDA lesion, we measured an approximate 40% loss of DA neurons in the SNc (Fig 6C) and the RRF (Fig 6D), with no significant loss in the VTA (Fig 6E) or for non-DA of the SNc neurons (Fig 6F). Interestingly, in the D2-cKO mice, approximately 60% of SNc DA neurons were lost, representing almost 50% more neurodegeneration than for control mice (60% loss vs 42% loss for CTL) (Fig 6C). As for axon terminals, TH signal area (Fig 6G) and total TH signal (Fig 6H) were both reduced by approximately 50% in the dorsal striatum, with no change detected in the ventral striatum, confirming the specificity of the lesion. In addition, DAT signal area (Fig 6J) and total signal (Fig 6K) were reduced by approximately 75% in the dorsal striatum. There were no significant changes in TH and DAT signal intensity (Fig 6I and 6L), suggesting loss of axonal terminals rather than simply reduced TH and DAT levels. Even if more neurons were lost in the SNc in D2-cKO mice compared to control mice, no significant difference was observed between the two genotypes (Fig 6G–6L) at the terminal level, compatible with compensatory axonal sprouting.
In line with the modest decrease in TH signal within the striatum of these mice and the absence of genotype effect in striatal denervation, we failed to detect a difference between D2-cKO mice and controls in motor behaviors (Fig 7). However, the 6-OHDA lesion caused an increased preference for the ipsilateral paw in the stepping test (Fig 7A) with no change in total number of steps (Fig 7B). In the rotation test, we observed no changes in rotational preference at basal levels (Fig 7C), but an increased preference for ipsilateral rotations under amphetamine was detected (Fig 7D). Finally, the total number of rotations was significantly increased following amphetamine administration (Fig 7E).
One of the key unanswered questions in PD research is why DA neurons of the SNc are particularly vulnerable. In the last few decades, a number of hypotheses have been raised regarding the core characteristic responsible for this vulnerability, including DA and iron toxicity, pacemaking activity and the establishment of a large and complex axonal arborization [7,3,4,23]. One commonality between these features is that they all lead to increased oxidative stress and bioenergetic demands, that are easily destabilized in pathology. Compatible with this model, we previously showed in vitro that SNc DA neurons have a higher rate of mitochondrial oxidative phosphorylation and basal oxidative stress compared with less vulnerable DA neurons of the VTA, characteristics that appear to be the result of their highly complex axonal arborization [8]. These results suggest that the size of this axonal arborization might be a critical determinant differentiating between surviving and degenerating neurons in PD.
The size of the axonal arbour of SNc DA neurons was measured previously in the intact rat brain [24,25], but no direct comparison of this parameter with less vulnerable VTA DA neurons was available prior to the present work. However, by dividing the estimated number of terminals in the rat ventral and dorsal striatum with the corresponding number of DA neurons in the VTA and SNc, it had been previously estimated that SNc DA neurons have an 8-fold broader striatal axonal arborization compared to VTA DA neurons [7]. In the present work, we directly measured axonal arborization size of both neuronal populations in the entire striatum and similarly found a much larger axonal (3-fold) arborization for SNc compared to VTA DA neurons in mice. The smaller difference between our finding (3-fold) and the previous estimate (8-fold) could be due to the use of different species (rats vs mice), but we additionally took into account that VTA DA neurons also project to the dorsal striatum; projections which were not considered in the previous estimation [7]. The projections of VTA neurons to the dorsal striatum were much more diffuse, but because of the much larger size of the dorsal striatum compared to the ventral striatum, they accounted for a significant amount of the total number of projections from the VTA. These projections were also previously examined in a single neuron tracing study in mice [26], but in this work, the authors did not compare VTA to SNc neurons. They nonetheless confirmed that part of VTA DA neurons projections were outside of the ventral striatum, compatible with previous classical work describing mesocortical and mesolimbic pathways [27]. It is also possible that we underestimated the axonal arborization size of SNc DA neurons, since we were not able to selectively label neurons from the most ventral part of the SNc, who are known to be particularly vulnerable in PD [28], since they were too close to the VTA. It is possible that these highly vulnerable neurons could have an even broader axonal arborization.
Another potential caveat of this study is that we did not quantify axonal processes outside of the striatum, which could have led to an overestimated difference between SNc DA neurons and VTA DA neurons, since VTA neurons are known to also project to other brain regions such as the cortex, amygdala and septum. However, in initial experiments, a global evaluation of these regions revealed only a very low relative density of dopaminergic processes compared to the striatum. We thus limited our quantification to the striatum in the present study, which is the main projection site for both populations of DA neurons. Although this represents a limitation, we consider it unlikely that our estimates were significantly affected by this focus on striatal projections. In keeping with this possibility, the relative difference between the size of the axonal arborization of VTA and SNc DA neurons found in the present study is quite similar (2–3 fold larger for SNc compared to VTA) to what we previously observed in vitro [8]. Another limitation of the present work is that due to the quantity and volume of injected virus, we were not able to separately quantify the axonal arborization size of different subtypes of SNc or VTA DA neurons. Considering that the ventral tier of the SNc is much more vulnerable in PD compared to the dorsal tier [28] and that projections from the different subpopulations of DA neurons reach different subregions of the striatum [29], it would be of major interest in future work to examine axonal arborization size of different subpopulations of the SNc in relation to their differential vulnerability in PD. The use of intersectional genetic tools might be better suited than trying to reduce viral injection volume to tackle this question.
In the present study, we used D2-cKO mice to examine the vulnerability of DA neurons under conditions where these neurons develop an even larger axonal arborization. Increased DA terminal density in the dorsal striatum had been previously described in a constitutive [17] knockout model of this receptor. In order to focus on cell-autonomous mechanisms of vulnerability, we deleted the D2R gene selectively in DA neurons by crossing Drd2loxP mice with DATIREScre mice. Using these DATIREScre/Drd2loxPmice, we surprisingly did not find any changes in TH signal in the striatum, an observation that could reflect the highly plastic and homeostatic nature of TH expression in response to perturbations such as neurotoxins, which might make it somewhat unreliable to assess the extent of loss of axonal processes [30–36]. On the other hand, we did observe a significant increase in DAT immunoreactive processes in the dorsal striatum, as shown previously in the constitutive KO model [17], with no increase in the number of DA neurons. This is also similar to what has been observed previously in the hippocampus of this cKO model [19]. However, DAT expression and localisation can be altered by many mechanisms [37,38]. For example, this transporter is known to form protein-protein interactions with the D2 DA receptor, which is thought to promote DAT localisation to the plasma membrane [39–42]. For this reason, the lack of D2 receptors in our D2-cKO models could have altered the expression of DAT by compensatory mechanisms and not directly as a result of an increase in axonal arborization size.
To evaluate if the increase in DAT immunoreactive processes reflected an increase in axonal arborization size and was originating from SNc DA neurons and not from DA neurons from other regions such as the VTA, we took advantage of a viral labelling strategy to conditionally express a fluorescent reporter protein in SNc or VTA DA neurons. Doing so, we confirmed that SNc but not VTA DA neurons have an increased number of axonal processes in the striatum of the D2-cKO mice. To validate whether expanded axonal domains contained terminals that were likely to release DA, we also quantified the presence of TH, DAT and VMAT2 in these virally-labelled axonal processes. We found that there was an increase in colocalization with TH and DAT and no change in VMAT2 density in axonal processes, arguing that the increase in axonal size did not come at the expense of a loss in neurochemical identity.
We next used fast scan cyclic voltammetry to measure DA release in the striatum and to gain further insight into the functionality of dopaminergic axons in this model. We found a general decrease in DA release that was partially rescued in the presence of a DAT antagonist. Our finding of a decrease in evoked DA overflow, although somewhat counter-intuitive when considering the autoreceptor function of the D2 receptor, is in line with previous observations of constitutive or conditional D2R KO mice [43–47] (but see [48]). While we found here that this reduced DA release could be rescued by nomifensine, in a previous study, the use of a DAT antagonist was not sufficient to return DA levels to normal in the engrailed1-based D2-cKO [44]. It should be noted however that in this later work, while the control condition had both alleles of englailed1, the D2-cKO mice had only one allele of this transcription factor, which is otherwise critical for the development of DA neurons. Since knockout of even only one allele of engrailed1 has been shown to affect the number of DA neurons and the density of their terminals [49,50], it is possible that DA release in this model was affected by both the KO of the D2 receptor and the reduced engrailed1 expression as well as by the possible removal of the D2 receptor in engrailed expressing non-DA neurons of the VTA and SNc.
It is also important to note again that there have been reports that activation of D2 receptors in dopaminergic terminals regulates positively the localization of the DAT to the plasma membrane [39–42]. In our D2-cKO mice, although we detected an increase in DAT levels by immunofluorescence, we did not observe any significant change in reuptake kinetics as assessed from cyclic voltammetry recordings. We also did not detect a significant change in DAT surface levels using a DAT surface biotinylation assay. However, a reduction in Vmax has been observed in a previous study in which the D2 receptor was knocked down acutely using siRNA [47], although reuptake kinetics (tau) were not reported. The difference with our data could be explained by the acute nature of the deletion in this previous study. In the context of the absence of a change in reuptake kinetics, our finding of an apparent rescue of DA release in the presence of the DAT blocker nomifensine is puzzling. One possibility is that nomifensine was able to rescue a deficit in axon terminal function at a step which is independent from DAT activity. Previous work has indeed shown that DAT blockers including cocaine and nomifensine are able to enhance the exocytotic release of DA through a mechanism that is not yet clearly defined but that has been suggested to involve synapsin [20–22].
The goal of this work was to provide a first in vivo test of the importance of axonal arborization size on the vulnerability of SNc DA neurons. We confirm here that D2-cKO mice represent a model in which an expansion of the axonal arborization of SNc DA neurons can be detected. Based on our previous work performed with primary DA neurons [8], we predicted that this should lead to increased vulnerability of SNc DA neurons. In keeping with this hypothesis, we found that D2-cKO SNc DA neurons were more vulnerable to a 6-OHDA lesion initiated at the axon terminal level. An alternate interpretation of this increased neuronal loss in D2-cKO mice could be that the basal increase in DAT-positive varicosities observed in these mice led to an increased uptake of 6-OHDA. Although this possibility cannot be formally excluded, its likelihood is limited because our cyclic voltammetry reuptake kinetic measurements argue for an absence of change in DAT functionality at the plasma membrane, a finding that is in line with our observation of a lack of change in DAT surface levels in the striatum of D2 cKO mice. In the 6-OHDA model, we also observed a stronger loss of cell bodies than striatal terminals, with similar levels of striatal TH and DAT fiber density in D2-cKO mice compared to control mice. This finding argues for robust axonal sprouting from surviving neurons in the D2-cKO mice. This is in line with work showing presence of compensatory reinnervation in this lesion model [51,52] and is also supported by the absence of exacerbated 6-OHDA induced behavioral impairements in the D2-cKO mice. In future work, it would be of interest to look at the vulnerability of VTA DA neurons to 6-OHDA in the D2-cKO model using toxin injection targeted to the ventral striatum, as these neurons do not show any changes in their axonal arborization size, but are thought to participate in intense axonal spouting in this lesion model [51,53].
Because the D2 receptor regulates many cellular processes, we cannot completely exclude the possibility that lack of D2 receptors could have increased the vulnerability of SNc DA neurons through mechanisms other than the increased axonal arborization size. Future work will be required to determine the origins of this enhanced neuronal loss, but an increased level of basal oxidative stress in SNc DA neurons could be implicated and synergistically lead to sufficient oxidant stress to initiate apoptotic death of DA neurons [54–56]. ROS production induced by 6-OHDA has also been reported to impair axonal transport in dopaminergic neurons [57] and to deplete ATP content and antioxidant reserve [58], which could affect to a greater extent D2-cKO SNc DA neurons since they have a larger axonal compartment to maintain. Additionally, increased phosphorylation of α-synuclein to its pSyn-129 toxic form has been reported in the 6-OHDA model [59], which could play a role in the observed toxicity. However, it is unlikely that this effect on α-synuclein is the main mechanism leading to cell death in the present study because we failed to detect any change in vulnerability when we overexpressed α-synuclein, even if we observed the presence of pSyn-129 in surviving cell bodies.
This lack of an increased vulnerability to α-synuclein overexpression in the present model is presently unresolved, but it might be explained by the fact that pathology is initially induced in the cell bodies in this model, as opposed to its initiation in the terminals in the 6-OHDA model and that the time course of neurodegeneration is much longer in the overexpression model (months vs days for 6-OHDA). Additionally, the α-synuclein model is thought to trigger degeneration by causing pathological protein aggregation and impaired proteasome/lysosome function [60–62], unlike the 6-OHDA model, which directly impairs mitochondrial function by inhibiting mitochondrial complex I and IV and by inducing oxidative stress [63,64].
However, it has been suggested that α-synuclein overexpression can also influence mitochondrial function, but through different mechanisms. It has been proposed that once oligomerized, α-synuclein influences mitochondrial fusion/fission, transport, clearance and protein import mechanisms [65,66], as well as complex I and ATP-synthase function [67] and therefore increases oxidative stress [68]. Since α-synuclein oligomerization seems to be a necessary step for all these alterations, overexpression of WT α-synuclein should take much more time than 6-OHDA injections to elevate oxidative stress to critical levels. It should therefore also leave much more time for neurons to attempt to compensate for these changes compared to the 6-OHDA model where ATP and antioxidant depletion and oxidative stress are rapidly induced. In combination with the much more modest loss of striatal TH immunoreactive processes in the α-synuclein overexpression model, this could in part explain why behavioral alterations were almost absent in this model.
Additionally, it is also possible that the potentially enhanced level of oxidative stress in the nigro-striatal system of D2-cKO mice was not sufficiently elevated to promote enhanced vulnerability in response to all triggers of PD pathology. In line with this possibility, a global assessment of superoxide anion production and NADPH oxidase activity in the striatum and mesencephalon of the D2 cKO mice failed to reveal an increased stress level (S3 Fig). Further experiments would be needed to examine selective markers of oxidative stress in the axonal and somatodendritic compartment of DA neurons, without the confounding presence of signal originating in striatal neurons and glial cells.
Interestingly, even in the absence of exogenous triggers such as 6-OHDA or α-synuclein overexpression, features of PD pathophysiology such as loss of processes and presence of α-synuclein aggregates in the dorsal striatum have been reported in aged constitutive D2-KO mice [69]. In the present work, we did not produce nor examine aged D2-cKO mice, but it is possible that similar pathology would be observed.
In conclusion, this work demonstrates for the first time that SNc DA neurons in the intact brain possess a larger axonal arbour size compared to VTA DA neurons. This work also provides strong additional supportive evidence for the hypothesis that a very large axonal arbour places DA neurons at increased risk in PD.
All procedures involving animals were conducted in strict accordance with the Guide to care and use of experimental animals (2nd Ed.) of the Canadian Council on Animal Care. The experimental protocols were approved by the animal ethics committee (CDEA) of the Université de Montréal.
Housing was at a constant temperature (21°C) and humidity (60%), under a fixed 12h light/dark cycle and free access to food and water. Initial comparisons of the axonal arborization size of SNc and VTA DA neurons was performed using DAT-Cre knock-in mice [70]. The rest of the experiments were performed using DATIREScre mice obtained from Jackson Labs [71] and crossed with Drd2loxP mice [48]. Mouse background was mixed 129SV/C57BL6 and both males and females were used.
All animals were genotyped using a KAPA2G Fast HotStart DNA Polymerase kit from Kapa Biosystem. Primer used were:
DAT-Cre
DATIREScre
Drd2loxP
Two-month-old DAT-Cre or DATIREScre positive mice were anesthetized with isoflurane (Aerrane; Baxter, Deerfield, IL, USA) and fixed on a stereotaxic frame (Stoelting,Wood Dale, IL, USA). Fur on top of the head was trimmed, and the surgical area was disinfected with iodine alcohol. Throughout the entire procedure, eye gel (Lubrital, CDMV, Canada) was applied to the eyes, and a heat pad was placed under the animal and kept warm. Next, bupivacaine (5 mg/ml and 2 mg/kg, Marcaine; Hospira, Lake Forest, IL, USA) was subcutaneously injected at the surgical site, an incision of about 1 cm made with a scalpel blade, and the cranium was exposed. Using a dental burr, one hole of 1 mm diameter was drilled above the site of injection [AP (anterior–posterior; ML (medial–lateral); DV (dorsal-ventral), from bregma]. The following injection coordinates were used:
Note that the coordinates for SNc and VTA injections were purposely 0.3 mm anterior to the center of the targetted region. These coordinates were adjusted to prevent infection of RRF, rostral linear nucleus (RLI) or caudal linear nucleus (CLI) DA neurons. Next, borosilicate pipettes were pulled using a Sutter Instrument, P-2000 puller, coupled to a 10 μL Hamilton syringe (Hamilton, 701RN) using a RN adaptor (Hamilton, 55750–01) and the whole setup was filled with mineral oil. Using a Quintessential Stereotaxic Injector (Stoelting), solutions to be injected were pulled up in the glass pipet. For the axonal arborization size quantification, 0.1 μL (SNc) or 0.05 μL (VTA) of sterile NaCl containing 1.15x1012 viral genome particles/mL of AAV2-EF1a-DIO-eYFP (UNC Vector Core, Chapel Hill, NC, USA) was injected. For α-synuclein over-expression, 0.8 μL of AAV2-CBA-alpha-Syn (3.8x1012 viral genome particles/mL, MJF Foundation, USA) or AAV2-CBA-eGFP (2.0x1012 viral genome particles/mL MJF Foundation, USA) was injected. For 6-OHDA lesions, 0.5 μL of 6-OHDA (3 mg/mL) in 0.2% ascorbic acid solution was injected. Forty minutes prior to 6-OHDA injections, the norepinephrine transporter blocker desipramine (25mg/Kg) was injected intraperitoneally to the animals to prevent lesions of the noradrenergic fibers. After the unilateral injection, the pipette was left in place for 10 min to allow diffusion and then slowly withdrawn. Finally, the scalp skin was sutured and a subcutaneous injection of the anti-inflammatory drug carprofen (Rimadyl, 50 mg/mL) was given. Animals recovered in their home cage and were closely monitored for 24h. A second dose of carprofen (5 mg/kg) was given if deemed necessary. The brains were collected 1 month after the 6-OHDA injection (P90), 2 months after viral injection for axonal arborization labeling (P120) or 3 months after viral injection for α-synuclein overexpression studies (P150).
Mice were anesthetized using pentobarbital NaCl saline solution (7 mg/mL) injected intraperitoneally and then were perfused with 50mL of PBS followed by 100 mL of paraformaldehyde (PFA) 4% using an intracardiac needle at a rate of 25 mL/min. The brains were extracted, placed 48h in PFA followed by 48h in a 30% sucrose solution and frozen in isopentane at -30°C for 1 minute. 40 microns thick coronal sections were then produced using a cryostat (Leica CM1800) and placed in antifreeze solution at -20 oC until used.
One out of every 6th slice was used for immunofluorescence. After a PBS wash, the tissue was permeabilized, nonspecific binding sites were blocked and slices were incubated overnight with a rabbit anti-TH antibody (1:1000, AB152, Millipore Sigma, USA), a rat anti-DAT antibody (1:1000, MAB369; MilliporeSigma, USA), a chicken anti-GFP (1:2000, GFP-1020; Aves Labs, USA), a mouse anti-p-S129-α-synuclein (1:2000, 328100, Invitrogen, USA), a chicken anti-α-synuclein (1:2000, AB190376, Cedarlane, USA) and/or rabbit anti-VMAT2 (1:2000, kindly provided by Dr. G.W. Miller [72]) Primary antibodies were subsequently detected with a rabbit or chicken Alexa Fluor-488–conjugated secondary antibody, a rabbit Alexa Fluor-546–conjugated secondary antibody, and/or a rat Alexa Fluor-647–conjugated secondary antibody (1:400; Thermo Fisher Scientific).
One out of every 6th slice was used for DAB immunostaining. After a PBS wash, the tissue was incubated for 10 min with 0.9% H2O2 solution, then washed with PBS again and incubated for 48h with a rabbit anti-TH antibody (1:1000, AB152, Millipore Sigma, USA) at 4°C, 12h with goat anti-rabbit biotin-SP-AffiniPure secondary antibody (111-065-003, Jackson ImmunoResearch Laboratories, USA) at 4°C and 3h with horseradish peroxidase streptavidin (016-030-084, Cedarlane, USA). The DAB reaction was carried out for 45s, then stopped by incubation with 0.1M acetate buffer and slices were mounted on Superfrost/Plus microscope slides. They were left to dry for 96h after which they were stained with cresyl violet and went through subsequent incubations with increasing concentrations of alcohol. After short isopropanol and xylene baths, slides were sealed with Permount mounting medium (SP15-100, Fisher, USA) using glass coverslips.
All of the imaging quantification analyses were performed on images captured using confocal microscopy. Images were acquired using an Olympus Fluoview FV1000 microscope (Olympus). Images acquired using 488 and 546 nm laser excitation were scanned sequentially to reduce nonspecific bleed-through signal. For each slice, up to 4 images were acquired in the dorsal striatum and up to 2 in the ventral striatum. All image quantifications were performed using ImageJ (National Institutes of Health) software. We first applied a background correction and then measured the area and intensity of the signal.
For quantification of TH, DAT and VMAT2 positive terminals in the ventral or dorsal striatum, images were acquired using a 60x oil-immersion objective and averaged from slices at bregma 1.18, 0.14 and -0.94 mm. For axonal arborization size quantification with eYFP viral expression, images were acquired on one out of every 6th slice from bregma -2.2 to 1.94 mm using a 20x water immersion objective since the fibers were easily distinguishable at lower magnification. The proportion of the area covered by eYFP fibers was extrapolated to the size of the striatum for each slice based on The Mouse Brain in Stereotaxic Coordinates 3rd Edition by George Paxinos [73] normalized by the number of infected neurons counted manually (300–1000 neurons) and plotted in relation to the bregma coordinates. Stereological counting was not used for this quantification since the number of neurons was too low to get a reliable count using random sampling. The volume of eYFP positive axonal arborization was then approximated using the area under the curve.
The number of striosomes and their size was also quantified using the integrated particles analyzer in Image J. Colocalization measurements were performed using the Jacop plugin for ImageJ on 60x confocal images [74]. Mander’s M1 and M2 coefficients were obtained after manual thresholding of the images to remove background. A mask of the YFP signal was applied to the other signals for measurement of their colocalization inside YFP fibers.
TH-immunoreactive neurons were counted in one out of every sixth section using a 100x oil-immersion objective on a Leica microscope equipped with a motorized stage. A 60 x 60 μm2 counting frame was used in the Stereo Investigator (MBF Bioscience) sampling software with a 12 μm optical dissector (2 μm guard zones) and counting site intervals of 150 μm after a random start (100 μm intervals for unilateral lesion). Mesencephalic DA nuclei, including the VTA, SNc and RRF were examined. Stereological estimates of the total number of TH-immunoreactive neurons within each nucleus were obtained. The number of TH-negative neurons was also estimated similarly in each region based on cresyl violet staining.
Acute brain slices from 3-month-old mice were obtained using a protective slicing method [75]. Matched pairs of CTL and D2-cKO mice were used on each experimental day. After intracardiac perfusion, brains were quickly dissected, submersed in ice-cold NMDG cutting solution and coronal striatal brain slices of 300 μm (from bregma AP 1.34 to 0.98 mm) were prepared with a Leica VT1000S vibrating microtome in ice-cold (0 to 4°C) NMDG protective cutting solution. Slices recovered for 12 min in 32° NMDG solution and were then transferred to oxygenated HEPES-buffered resting solution at RT for at least 1h. For recordings, slices were put in a custom-made recording chamber superfused with artificial cerebral spinal fluid (aCSF) at 1 mL/min and maintained at 32°C. All solutions were adjusted at pH 7.35–7.4, 300 mOsm/kg and saturated with 95% O2-5% CO2 at least 30 min prior to each experiment.
Electrically induced DA release was measured by fast-scan cyclic voltammetry (FSCV) using a 7 μm diameter carbon-fiber electrode placed into the dorsal or ventral striatum ∼100 μm below the surface and a bipolar electrode (Plastics One, Roanoke, VA, USA) placed ∼200 μm away. Carbon-fiber electrodes were fabricated as previously described [76]. Electrodes were polished and filled with 4M potassium acetate and 150 mM potassium chloride. Carbon fibers were then cut using a scalpel blade to obtain maximal basal currents of 100 to 180 nA. Electrodes were finally selected for their sensitivity to DA using in vitro calibration with 1μM DA in aCSF before each experiment. Before and after use, electrodes were cleaned with isopropyl alcohol. The potential of the carbon fiber electrode was scanned at a rate of 300 V/s according to a 10 ms triangular voltage wave (−400 to 1000 mV vs Ag/AgCl) with a 100 ms sampling interval, using a CV203BU headstage preamplifier (Axon instrument, Union City, CA)) and an Axopatch 200B amplifier (Axon Instruments). Data were acquired using a Digidata 1440A analog to digital converter board (Axon Instruments) connected to a computer using Clampex (Axon Instruments). Slices were left to stabilize for 20 min before any electrochemical recordings.
After positioning of the bipolar stimulation and carbon fiber electrodes in the striatum, single pulses (400 μA, 1ms) were applied to the nucleus accumbens core (referred to as ventral striatum) and then to the dorso-lateral part of the dorsal striatum to trigger DA release. Stimulations were applied every 2 min. After recording in the dorsal striatum, the media was changed to ACSF containing 5 μM of nomifensine (Sigma) and single stimuli were applied to the dorsal striatum. Electrode calibration was performed before and after the recording of each slices and the average value for the current at the peak oxidation potential was used to normalize the recorded ex vivo current signals to DA concentrations.
DA release was analyzed as the peak height of DA concentrations and DA reuptake was determined from the clearance rate of DA which was assumed to follow Michaelis-Menten kinetics. A nonlinear least square optimization was applied to fit a three-parameter exponential function with baseline shift to the reuptake phase of the DA response. Uptake parameters (tau and Vmax) were calculated based on the exponential fitting. To determine whether DAT-mediated DA uptake was compromised in D2-cKO mice, the initial portion of the falling phase of single pulse evoked [DA]o curves was used to calculate the Vmax (maximal rate of DA uptake) after setting the Km parameter to 0.2 μM, based on the affinity of DA for the DAT, measured in mouse synaptosome preparations [77] and with the assumption that the Km is not altered in the KO mouse line.
Surface biotinylation experiments were carried using a protocol modified from Rickhag et al. 2013 [78]. Brains from 3-month-old conditional D2-cKO mice and CTL littermates were rapidly dissected and submerged in pre-oxygenated (95% O2 and 5% CO2) ice-cold sucrose buffered artificial cerebrospinal fluid. Coronal striatal sections (300 μm) were obtained using a vibrating blade microtome (Leica VT1000). The slices were allowed to recover in oxygenated aCSF (without sucrose) for 1h at room temperature. After surface biotinylation, slices were rinsed twice and excess biotin was quenched by two washes in glycine in oxygenated aCSF (4°C). The biotinylated slices from individual mice were pooled and homogenized in lysis buffer containing protease and phosphatases inhibitor. The homogenates were quickly incubated, gently mixed and centrifuged to remove debris (4°C).
Protein concentrations were measured and adjusted to 1ug/ml, and 100 μl of total lysates were stored to allow determination of the total protein input. Biotinylated proteins were isolated by loading equal amounts of protein onto 175ul avidin beads (Thermo Scientific) followed by overnight incubation at 4°C. Beads were washed in lysis buffer before elution of biotinylated proteins. Avidin beads were removed by filtration, and surface and total DAT levels were evaluated by western blot analysis. Protein samples were separated by SDS-PAGE and transferred to membranes. The membranes were blocked and then incubated subsequently with antibodies against DAT (Millipore MAB369, 1:1000) and with horseradish peroxidase (HRP)-conjugated anti-rat antibodies. Surface DAT protein bands were visualized by chemiluminescence. Blots of surface protein samples were reprobed for Na+/K+-ATPase (Abcam 1:500) to account for variation in biotinylated input while actin (HRP-conjugated actin (1:10000, A3854, mouse monoclonal, Sigma) was used as loading control for the total lysates. Band intensities were quantified using ImageJ gel analysis software.
Basal superoxide anion production and NADPH oxidase activity in brain tissues were measured using the lucigenin‐enhanced chemiluminescence method with a low concentration (5 μmol/L) of lucigenin, as described previously [79]. The tissues from control and D2-cKO mice were washed in oxygenated Krebs HEPES buffer and placed in scintillation vials containing lucigenin solution, and the emitted luminescence was measured with a liquid scintillation counter (Wallac 1409; Perkin Elmer Life Science) for 10 minutes. The average luminescence value was estimated, the background value was subtracted, and the result was divided by the total weight of tissue in each sample. The NADPH oxidase activity in the samples was assessed by adding 10 to 4 mol/L NADH (Sigma‐Aldrich) in the vials before counting. Basal superoxide–induced luminescence was then subtracted from the luminescence value induced by NADH.
All mice were habituated to the user by handling them once a day during 3 consecutive days before experiments. Mice were moved to the experimental room 1h before the test.
Mice first went through a stepping test recorded with a digital camera (DMK 22BUC03, ImagingSource) and IC Capture 2.4 software. Mice were gently lifted by the base of the tail at one end of a 1-meter corridor leaving only forepaws touching the surface and were pulled backward for 4s over a distance of 1-meter. Recordings were then watched in slow motion and the number of steps of each forepaw was counted. After 1h of rest, animals were placed in a 4L beaker with the digital camera recording their movements from underneath to assess rotation. After 20 min, amphetamine 5 mg/kg was intraperitoneally injected and mice were placed back in the beaker for 40 min. Recordings were then watched to count the ipsilateral and contralateral rotations made by the mice during the first (basal) and the last (amphetamine) 20 min.
All experiments were performed blind to the experimental groups, from surgeries to image analysis. Parametric statistical tests were used because samples contained data with normal distributions. Data were presented as mean ± SEM. The level of statistical significance was established at p < 0.05 in one or two-way ANOVAs or two-tailed t-tests with Welch’s correction when needed. A ROUT outlier analysis was performed when required (Q = 1%). Statistical analyses were performed with the Prism 7 software (GraphPad Software, p < 0.05 = *, p < 0.01 = **, p < 0.001 = ***, p < 0.0001 = ****). The Tukey post-hoc test was used when all the means were compared to each other and the Sidak post-hoc test was used when only subsets of means were compared.
|
10.1371/journal.pgen.1002440 | Contribution of Intragenic DNA Methylation in Mouse Gametic DNA Methylomes to Establish Oocyte-Specific Heritable Marks | Genome-wide dynamic changes in DNA methylation are indispensable for germline development and genomic imprinting in mammals. Here, we report single-base resolution DNA methylome and transcriptome maps of mouse germ cells, generated using whole-genome shotgun bisulfite sequencing and cDNA sequencing (mRNA-seq). Oocyte genomes showed a significant positive correlation between mRNA transcript levels and methylation of the transcribed region. Sperm genomes had nearly complete coverage of methylation, except in the CpG-rich regions, and showed a significant negative correlation between gene expression and promoter methylation. Thus, these methylome maps revealed that oocytes and sperms are widely different in the extent and distribution of DNA methylation. Furthermore, a comparison of oocyte and sperm methylomes identified more than 1,600 CpG islands differentially methylated in oocytes and sperm (germline differentially methylated regions, gDMRs), in addition to the known imprinting control regions (ICRs). About half of these differentially methylated DNA sequences appear to be at least partially resistant to the global DNA demethylation that occurs during preimplantation development. In the absence of Dnmt3L, neither methylation of most oocyte-methylated gDMRs nor intragenic methylation was observed. There was also genome-wide hypomethylation, and partial methylation at particular retrotransposons, while maintaining global gene expression, in oocytes. Along with the identification of the many Dnmt3L-dependent gDMRs at intragenic regions, the present results suggest that oocyte methylation can be divided into 2 types: Dnmt3L-dependent methylation, which is required for maternal methylation imprinting, and Dnmt3L-independent methylation, which might be essential for endogenous retroviral DNA silencing. The present data provide entirely new perspectives on the evaluation of epigenetic markers in germline cells.
| In mammals, germ-cell–specific methylation patterns and genomic imprints are established throughout large-scale de novo DNA methylation in oogenesis and spermatogenesis. These steps are required for normal germline differentiation and embryonic development; however, current DNA methylation analyses only provide us a partial picture of germ cell methylome. To the best of our knowledge, this is the first study to generate comprehensive maps of DNA methylomes and transcriptomes at single base resolution for mouse germ cells. These methylome maps revealed genome-wide opposing DNA methylation patterns and differential correlation between methylation and gene expression levels in oocyte and sperm genomes. In addition, our results indicate the presence of 2 types of methylation patterns in the oocytes: (i) methylation across the transcribed regions, which might be required for the establishment of maternal methylation imprints and normal embryogenesis, and (ii) retroviral methylation, which might be essential for silencing of retrotransposons and normal oogenesis. We believe that an extension of this work would lead to a better understanding of the epigenetic reprogramming in germline cells and of the role for gene regulations.
| Throughout mammalian gametogenesis, dynamic DNA methylation changes occur in a sex- and sequence-specific manner. These changes result in the establishment of oocyte- and sperm-specific genomic imprints and unique methylation patterns of repetitive elements via DNA methyltransferase activity [1]–[4]. This process is indispensable for functional gamete and embryo development. For example, sex-specific methylation imprints are maintained throughout cell division after fertilization, despite genome-wide demethylation and de novo methylation during embryogenesis. These imprints control parent-of-origin specific monoallelic expression of a subset of genes, which are known as imprinted genes [5]–[9]. In addition, DNA methylation during spermatogenesis plays a crucial role in meiotic progression and retrotransposon silencing [10]–[14]. However, little is known about the profile and functional role of DNA methylation during oogenesis, except for the establishment of genomic imprints.
Recently, the epigenetic modifications which are responsible for regulating cell differentiation and embryo development have been studied in detail by using high-throughput sequencing: bisulfite sequencing (“BS-seq”); “Methyl-seq” with a methyl-sensitive restriction enzyme; “MeDIP-seq” with methylated DNA immunoprecipitation; and “MBD-seq” with a methyl-DNA binding domain protein antibody [15]–[26]. However, a major limitation of epigenomic studies is the lack of a standard methodology for DNA methylome analysis. Ideally, the gold standard is high resolution and genome-wide methylome analysis of germ cells. However, genome-wide methylome analysis of female germ cells has almost never been performed due to the limited availability of samples. Shotgun bisulfite sequencing (SBS) may be able to overcome this limitation and enable the determination of the cytosine methylation status of individual CpG sites at a whole-genome level without a bias toward CpG-rich regions [22], [23], [26] and with only relatively small-scale DNA samples [24], [27]. As a result, in this study, an improved SBS method for small-scale DNA samples was used to analyze the DNA methylome of mouse germ cells. In addition, the mouse germ cell transcriptome was investigated using high-throughput cDNA sequencing (mRNA-seq) to reveal relationships between DNA methylation and gene transcription in both male and female germ cells.
We performed SBS analysis by using MethylC-seq [22] and a new SBS method called “whole bisulfitome-amplified DNA sequencing” (WBA-seq). The MethylC-seq and WBA-seq libraries were generated as shown in Figure S1. The MethylC-seq method generated 1010 and 1085 million tags (reads) from germinal vesicle (GV) stage oocytes and epididymal sperm, respectively. Oocyte DNA libraries generated by MethylC-seq showed higher redundancies than sperm DNA libraries. For example, 33.0% and 81.7% of the 21 million cytosines of CpGs in the mouse genome were covered by at least 1 sequence read from GV oocytes and sperm, respectively; whereas the average read depth (i.e., the number of hits of reads that were mapped to a given position) was over 10× for both germ cells (Figure S2). The WBA-seq method generated 307 and 397 million tags from GV oocytes obtained from wild-type and Dnmt3L-deficient (Dnmt3L−/−) mice, respectively. WBA-seq libraries for GV oocytes showed higher genome coverage (60% of genomic CpGs were covered by at least 1 read) but with smaller average read depth (7.4×) than MethylC-seq library. Some reads from the oocyte libraries strongly matched mitochondrial DNA (mtDNA), satellite, low complexity, or simple repeat sequences (Figure S3), which might have been due to a distinct genomic copy number bias in the mitochondria of germ cells or an over-amplification bias. Thus, SBS results were simplified by removing the redundancy information (only mtDNA was separately examined for DNA methylation) and combining MethylC-seq and WBA-seq results for wild-type oocytes. Consequently, the average read depth was 18.8×, 4.4×, and 12.5× for wild-type and Dnmt3L−/− oocytes, and sperm, respectively, and 70.8%, 45.6%, and 79.9% of genomic CpGs were covered by at least 1 sequence read from each cell type (Table 1 and Figure S3). Furthermore, the average read depths of MethylC-seq of mouse blastocysts and embryonic stem cells (ESCs), which served as zygote and stem cell controls, were 12.8× and 6.1×, respectively (Table 1).
The average methylation level of wild-type oocytes (40.0%) was less than half that of sperm (89.4%) (Figure S4). This difference in global DNA methylation between male and female germ cells was consistent with results from the previous studies [28], [29]. The Dnmt3L−/− oocyte genome was observed to be hypomethylated, exhibiting a methylation level of only 5.5%. Furthermore, blastocysts showed a lesser extent of methylation (21.3%) than did wild-type oocytes; ESCs, on the other hand, showed relatively high levels of methylation (70.6%). To elucidate the distribution of methylation levels on CpG sites, on regional and genome-wide scales, we created dot plots of CpG methylation for individual chromosomes and histograms of the methylation levels for all CpGs. These graphs revealed that hypermethylated CpGs in oocytes tended to cluster in transcribed regions of particular genes (e.g., Kcnq1 or Rlim genes, known to be expressed in oocytes [30], [31]); the sperm genome was almost entirely hypermethylated, except at most CpG-rich regions (Figure 1 and Figure S5). Specifically, 55.7% of the CpGs in the oocyte genome exhibited <10% methylation, whereas another 32.0% of CpGs exhibited ≥90% methylation (Figure 2A). The Dnmt3L−/− oocyte genome was also hypomethylated in almost all chromosomal regions (Figure S6). The methylation level of the mtDNA genome in Dnmt3L−/− oocytes (4.4%) was lower than that observed in wild-type oocytes (6.6%). Sperm methylation levels, by comparison, were relatively high (14.7%), whereas those of the blastocysts and ESCs were quite low (1.3% and 2.1%, respectively) (Figure S4).
Since previous studies revealed a significant correlation between CpG frequency and methylation within intra- and intergenic regions in somatic cells [32], [33], the CpG density and methylation levels were compared to identify genome-wide differential methylation patterns in germ cells. CpG density was defined as the number of CpG dinucleotides in 200 nucleotide (nt) windows (e.g., 1 CpG dinucleotide per 200 nt corresponds to a density of 0.005). At low CpG densities (range, 0.005–0.05), the oocyte genome was about 50% methylated, whereas the sperm genome was 80–90% methylated. At moderate to high CpG densities (range, 0.05–0.2), both male and female germ cells were hypomethylated (Figure 2B). Furthermore, 4 families of transposable elements (long interspersed nuclear elements (LINEs), short interspersed nuclear elements (SINEs), long terminal repeats (LTRs), and DNA transposons) were moderately methylated in oocyte genomes but were hypermethylated in sperm. In addition, a general trend towards higher methylation levels at higher CpG densities in the oocyte genome occurred in LTRs. Conversely, a trend toward lower CpG methylation levels at higher CpG densities in the wild-type oocyte and sperm genomes was observed in SINEs and DNA transposons. In contrast, all of these transposable elements were hypomethylated in Dnmt3L−/− oocytes. Interestingly, however, there was partial CpG methylation in LINEs and LTRs at relatively high CpG densities (range, 0.03–0.1). These complete or partial undermethylations were confirmed by bisulfite sequencing in L1 LINEs, B1/Alu SINEs, and intracisternal A particle (IAP) LTRs (Figure S7). These results suggested that each germ cell has a unique sequence- and CpG-density-dependent methylation pattern. In addition, oocyte CpG methylation, except in a subset of retrotransposons, appears to be Dnmt3L dependent.
We also characterized the methylation patterns of 15 germline-differentially methylated regions (gDMRs). The differential (between oocyte and sperm) methylation occurs at imprinted gene loci (also called imprinting control regions (ICRs)). The ICRs of maternally methylated imprinted genes (e.g., Nespas-Gnas) were shown to be hypermethylated in oocytes but hypomethylated in sperm, while the converse was true in ICRs of paternally-methylated imprinted genes (e.g., H19) (Figure 3 and Figure S8). Interestingly, only the Snrpn gDMR was partially methylated (35.7%), whereas all other maternal ICRs were hypomethylated in Dnmt3L−/− oocytes (Table 2). This residual methylation might result in the stochastic acquisition of the maternal imprint in the progeny of Dnmt3L−/− females [34]. These results strongly suggested that the methylation level of individual CpGs can be determined from DNA methylome maps with a high degree of accuracy.
The study of mammalian DNA methylation patterns has previously suggested that methylation predominantly occurs at CpG sites; however, more recent studies, based on SBS methods, have indicated that methylation at non-CpG sites also occurs in human ESCs [22], [23]. Detection of non-CpG methylation is one of the applications of the bisulfite-based methylation analysis but is problematic due to the incomplete conversion of cytosine, and overestimates of such cytosine by PCR amplification, which cannot be discriminated from true methylation. In order to evaluate the methylation status of non-CpG sites and avoid these problems, additional SBS analysis of mouse GV oocytes, sperm, blastocysts, and ESCs was performed by a non-amplification technique, termed Post-Bisulfite Adapter Tagging (PBAT) [Miura F. & Ito T, personal communication]. All C (originally methylated cytosine) and T (originally unmethylated cytosine) that mapped to genomic CpG and CpH sites (H = A, T, or C) were counted. The PBAT results showed CpG methylation ratios (C ratios = 0.395, 0.748, 0.137, 0.615 in oocytes, sperm. blastocysts, and ESCs) which are similar to the average methylation levels of individual DNA methylome maps obtained by MethylC-seq and WBA-seq among all examined cells. Interestingly, a relatively high fold enrichment of non-CpG methylation was observed in GV oocytes (C ratio = 0.034–0.038), but not in the other cell types, including mouse ESCs (C ratio <0.01) (Figure S11).
To elucidate the interaction between intragenic DNA methylation and gene transcription, the correlation between promoter and gene-body methylation and expression levels for 20,854 different genes was examined. The mRNA-seq profiles for germ cells and ESCs are shown in Table S1. The results showed that mRNA transcript levels in oocytes were strongly correlated to gene-body methylation levels (Spearman's ρ>0.5, p<1×10−9) but were not significantly correlated to promoter methylation levels (|ρ|<0.1) (Figure 4A). For example, the regions +2 to +5 kb from the transcription start site (TSS) and 0 to −5 kb from the transcription termination site (TTS) were hypermethylated (60–90% methylation) for the top 20% of expressed genes but were hypomethylated (10–30% methylation) for the bottom 20% of expressed genes. However, areas near the TSS (±500 base pairs (bp)) were hypomethylated (10–20% methylation) in all genes, regardless of their expression level. In contrast, in the Dnmt3L−/− oocyte genome, the correlation between gene expression and gene-body methylation was very weak (|ρ|<0.1) (Figure 4B). In the sperm genome, promoter methylation was negatively correlated (Spearman's ρ = −0.36, p<1×10−9) with gene expression, whereas gene-body methylation was positively correlated (Spearman's ρ = 0.14–0.16, p<1×10−9) to gene expression; the latter correlation was weaker than that observed in the oocyte genome (Figure 4C).
Further investigation of gene expression patterns in oocyte genomes revealed that the mRNA transcript levels between wild-type and Dnmt3L−/− oocytes were very highly correlated (R2 = 0.9611) (Figure 5A). In fact, there were no significant differences in the expression levels of representative oocyte-specific genes (e.g., Gdf9, Bmp15, Bcl2l10, Zp1, Zp2, Zp3, Zar1, Npm2, Nlrp5, and Dppa3, which are responsible for ovarian follicle formation, reproduction, and early development [35]) and DNA methyltransferase genes (e.g., Dnmt1, a maintenance methyltransferase, and Dnmt3a and Dnmt3b de novo methyltransferases); the expected difference in the expression level of Dnmt3L between wild-type and Dnmt3L−/− oocytes was observed (Figure 5B, 5C). These results suggested that changes in gene expression did not occur during oogenesis, despite global intragenic hypomethylation in Dnmt3L−/− oocytes. Furthermore, the expression levels and exon patterns of maternally-methylated imprinted genes across each ICR were not altered in Dnmt3L−/− oocytes (Figure 3 and Figure 5D). This result suggested that the disruption of maternal methylation imprints in the Dnmt3L−/− oocyte genome was not due to the lack of their transcription [36]. On the other hand, maternal methylation imprints at ICRs (and many other hypermethylations at transcribed regions) in wild-type oocyte genomes might be the result of gene transcription via Dnmt3L-mediated intragenic methylation.
Surprisingly, gene expression in ESC genomes was negatively correlated with promoter methylation and was not positively correlated with gene-body methylation (Figure S12). Meanwhile, these ESCs showed the apparent expression of all DNA methyltransferase gene families including Dnmt3L (Figure S13). Previous studies indicated that the zygotic and somatic functioning of Dnmt3L is not essential for global methylation in ESCs in mice [6]. Thus, unlike oocytes, the functional role of Dnmt3L in gene-body methylation after fertilization is unclear. However, the expression of pluripotency-associated genes, Pou5f1, Klf4, Sox2, Myc, Nanog, and Lin28a, was clearly observed in ESCs. The expression of Pou5f1, Lin28a, and Glis1, recently identified as maternal reprogramming factors, were also observed in oocytes (Figure S14). While differential expression of the pluripotency genes among germ and stem cells was observed, the promoter regions of these genes demonstrated low-level methylation in almost all of the examined cells. In sperm cells, only the Nanog promoter was hypermethylated (this result was similar to a previous study [29]).
To identify gDMRs, the average CpG methylation levels of individual CpG islands (CGIs), which are CpG-rich genomic regions often lacking DNA methylation, were calculated. Recently, Illingworth et al. determined the number of CGIs by deep sequencing of isolated, unmethylated DNA clusters [37]. Among the 23,021 mouse CGIs (22,974 CGIs were informative in both oocytes and sperm), 2014 were highly methylated (≥80% methylation) in oocytes, 818 were highly methylated in sperm, and 377 were highly methylated in both germ cells (Figure 6A). Furthermore, we also identified 1678 gDMRs (≥80% methylation in 1 gamete and ≤20% in the other), 1329 of which were oocyte-specific methylated CGIs, while the remaining 349 were sperm-specific methylated CGIs (Figure 6A, Figure S6, and Table S2). Among these gDMRs, 646 gDMRs were confirmed to show a differential methylation status between GV oocytes and sperm (by similar criteria: ≥75% methylation in 1 gamete and ≤25% in the other); the methylation status was previously examined by performing large-scale bisulfite sequencing of CpG-rich regions of the genome (reduced representation bisulfite sequencing: RRBS) (Table S3) [38]. Additionally, almost all known ICRs except Zdbf2 DMRs (which do not have any CGIs) were re-identified from our gDMR list (Table S2).
A total of 78% oocyte-methylated gDMRs (n = 1045) were located within the intragenic regions. Approximately 25% of the oocyte-methylated gDMRs (n = 322) overlap with either the first exon or the proximal promoter regions of the genes, as has been observed with most of the described maternal ICRs [39]; only 5% of the sperm-methylated gDMR (n = 18) showed such overlap. Alternatively, 34% of sperm-methylated gDMRs (n = 120) overlap with intergenic regions, as in all known paternal ICRs (Figure 6B). Interestingly, oocyte-methylated gDMRs in transcribed regions tended to be more abundant within highly expressed genes, but such a trend was not observed in the sperm genome (Figure 6C). Oocyte-methylated gDMRs were also identified in non-imprinted genes, such as the DNA methyltransferase genes (e.g., Dnmt1 and Dnmt3b) and some male germline-specific genes (e.g., Piwil1, Spag1, Ggnbp2, Tbpl1, Spata16, Ggn, Acrbp, and Cd46). The oocyte-methylated gDMR in Dnmt1 was located in spermatocyte- and somatic-specific exons, while oocyte-specific exons were hypomethylated in oocytes (Figure S9). Dnmt3L−/− oocytes also showed hypomethylation in most of these gDMRs. Significant changes in the expression levels of genes with alternative splicing patterns were not observed in the Dnmt3L−/− oocyte genome (Figure 3, Figure 5E, and Figure S9). These results indicate that these oocyte-specific methylated gDMRs do not regulate gene expression or alternative splicing during the oocyte stage.
To determine whether or not these germ cell-specific methylations are maintained after fertilization, when the genomes undergo global demethylation, the individual CGI methylation levels in blastocyst genomes were calculated. In blastocysts, all ICRs demonstrated low to moderate methylation (25.1–64.3%), whereas many gDMRs were demethylated (0–20%) (Figure 6D). Furthermore, 817 oocyte-methylated gDMRs (including Piwil1, despite being a non-imprinted gene locus) and 34 sperm-specific gDMRs were resistant to demethylation during early embryogenesis (≥20% methylation in blastocysts) (Figure 6D and Table S2). Among the demethylation-resistant gDMRs, a novel gDMR in the intron of Gpr1 (Figure S10) was found to be a tissue-specific, paternally-expressed imprinted gene [40]. Bisulfite sequencing analysis showed that this gDMR was hypomethylated in Dnmt3L−/− oocytes and maternal allele-specific methylation was detected in this region in blastocysts (Figure 6E). Methylation profiles in ESCs showed that 26% (n = 213) of the demethylation-resistant gDMRs became less methylated (0–20%) whereas the other gDMRs maintained or increased DNA methylation (Figure S15). Among ICRs, only Gnas exon1A ICR was demethylated (7.8%), whereas the other ICRs developed partial or high methylation levels (range, 38.6–83.1%) in ESCs (Table 2). Among other demethylation-sensitive gDMRs, which were demethylated (<20% methylation) in blastocysts, many (76%, n = 264) sperm-methylated gDMRs were re-methylated (≥20% methylation); most (81%, n = 416) of the oocyte-methylated gDMRs maintained low methylation (0–20%) in ESCs (Figure S15). Finally, out of 704 demethylation-resistant (in blastocysts) oocyte-methylated gDMRs which were informative in Dnmt3L−/− oocytes, only 4 remained hypermethylated (80–100% methylation) in the Dnmt3L−/− oocyte genome. However, almost all other oocyte-specific methylation marks at gDMRs were Dnmt3L-dependent (Figure 6F). These results suggest that Dnmt3L-mediated methylation during oogenesis regulates the establishment of most heritable oocyte-specific marks, including genomic imprints.
To the best of our knowledge, this is the first study to generate single-base resolution maps of DNA methylomes spanning the entire genome of mouse germ cells. The oocyte maps are particularly valuable and informative because, in the past, such an analysis was prohibitive due to the need for large quantities of DNA. Recently, Smallwood et al. [38] reported large-scale DNA methylation patterns in mouse germ cells by using the RRBS method, which targets only CpG-rich regions. However, our more comprehensive results provide strong evidence that gene expression was positively correlated to Dnmt3L-dependent intragenic methylation in oocytes, and that methylation patterns in oocytes differed from those in sperm and non-germline cells.
The functional role of gene-body methylation has been an enigma despite its conservation in plants and animals [41]–[43]. Maunakea et al. [44] suggested that gene-body methylation is involved in the regulation of alternative splicing events. Although methylated gDMRs were detected in the alternative exons of Dnmt1 and Gnas in mouse oocytes, loss of oocyte-specific methylation marks in the Dnmt3L−/− oocytes did not affect the expression patterns of alternatively spliced transcripts. Therefore, our results indicate that gene-body methylation is not involved in alternative splicing in oocytes.
Previously, Chotalia et al. [36] showed that transcription during the oocyte stage is required for the establishment of maternal methylation marks on an imprinted gene. The present results show that Dnmt3L−/− oocytes lost almost all of their maternal methylation imprints while maintaining a constant amount of mRNA through each ICR despite the global loss of intragenic methylation. Thus, these results strongly suggest that the establishment of genomic imprints via transcription is mediated by Dnmt3L-dependent intragenic methylation.
A possible mechanism for gene-body methylation involves the exposure of intragenic regions to DNA methyltransferases, considering that RNA polymerase disrupts the chromatin structure during transcription. However, not all transcripts across gDMRs corresponded to highly expressed genes in oocytes (Figure 6C). Therefore, other epigenetic marks with an open chromatin structure might also be important for DNA methylation in oocytes. For instance, a recent knockout study showed that Kdm1b, which encodes histone H3K4 demethylase, is required for the establishment of some maternal methylation imprints [45]. Thus, several factors, including transcriptional and epigenetic modifications, might be involved in Dnmt3L-mediated intragenic methylation.
The results of this study show that gene-body methylation was correlated to gene expression in sperm. However, the extent of that correlation is much less than in oocytes due to genome-wide hypermethylation, including in low-CpG-density regions. In male germline cells, global methylation acquisition begins during late embryonic development and before birth [3]. To more clearly show this correlation, analysis of early-stage germ cells in fetal or neonatal animals might be required. Surprisingly, a positive correlation between mRNA expression and gene-body methylation was not observed in mouse ESCs. In addition, the accumulation of non-CpG methylation was not observed in mouse ESCs. These results contradict the results of another study, which showed that active transcription was associated with intragenic DNA methylation with non-CpG methylation in human ESCs [22], [23]. This discrepancy might reflect the differences between human and mouse ESCs, the precise cell derivations or culture conditions [46], [47]. However, further comparative studies on germ cell epigenomes from other species are required to further elucidate the functional role of epigenetic marking systems.
In this study, a large number of heritable oocyte-specific methylation marks were identified within a set of novel CpG islands [37]. The difference in the number of oocyte- and sperm-specific gDMRs reflects the fact that only 3 or 4 paternally-methylated imprinted loci were observed, as compared to approximately 20 maternally-methylated imprinted loci. The reason for the relative abundance of oocyte-specific methylated CGIs might be related to the intragenic methylation of CpG-rich regions, which are hypomethylated in sperm. The results show that most of the oocyte-specific marks are Dnmt3L-dependent, similar to results recently obtained by RRBS-based analysis [38]. However, whether all of these CpG-rich regions serve as imprinting methylation marks is unclear. For instance, although many genes with oocyte-specific methylation marks were identified (Figure 6B), the evidence that these genes were imprinted was lacking (e.g., Piwil1 and Dnmt1). These methylation marks might not be involved in the formation of a fertile oocyte but might play crucial roles in gene expression after fertilization. Furthermore, ESC methylomes showed that many gDMRs, especially sperm-specific gDMRs, acquired new methylation patterns after implantation. Methylation of these CGIs might control tissue-specific gene expression [48], [49]. Partial alternation of imprinted methylation patterns in ESCs were observed in the present study, potentially caused by significant differences in the extent of the ICRs during embryo development [39]. A fuller understanding of epigenetic stability will require further methylome profiling during early embryogenesis and stem cell differentiation. The present study also identified a gDMR as a novel ICR candidate in the intron of the imprinted Gpr1 gene. Thus, traditional promoter arrays may not identify all ICRs. However, further analyses are needed to determine which gDMRs, identified in the CpG methylome maps, are true ICRs at the imprinted Gpr1-Zdbf2 locus [40], [50].
mRNA-seq results showed that the expression levels of most genes in the wild-type and Dnmt3L−/− oocytes were similar. For instance, the expression level of almost all oocyte-specific genes, which regulate ovarian follicle formation, reproduction, and early development, were not significantly altered (Figure 5B and Table S1). These results are consistent with the findings of previous studies, which showed that Dnmt3L−/− female mice were capable of producing fertile oocytes (however, their offspring were not viable due to the lack of imprinting) [5], [6]. Thus, regulation of oocyte-specific genes must be beyond the control of Dnmt3L-dependent cytosine methylation.
Although Dnmt3L−/− oocytes showed global hypomethylation at low to high CpG densities, some families of retrotransposons, such as LINEs and LTRs, were partially methylated at moderate to high CpG densities. Therefore, Dnmt3L-independent methylation might be involved in the silencing of retrotransposons and completion of oocyte meiosis. Previously, De La Fuente et al. [51] showed that Hells (also known as Lsh), which encodes a member of the sucrose non-fermenter 2 (SNF2) family of chromatin remodeling proteins, is required for DNA methylation of IAP and pericentromeric satellite repeats as well as repression of IAP retrotransposition in pachytene oocytes. Unfortunately, measurement of the methylation levels of satellite DNA, which is abundant in the pericentromeric regions, was not possible because these sequences were excluded from our analysis. However, a previous sequencing study showed that methylation levels of satellite DNA did not differ between the wild-type and Dnmt3L−/− oocytes [52]. Combined, these results suggest the presence of 2 types of oocyte methylation patterns: (i) Dnmt3L-mediated intragenic methylation that is essential for early embryogenesis and (ii) Dnmt3L-independent retroviral and pericentromeric methylation, which may be mediated by Hells activity, is crucial for oocyte meiosis [51]. Further studies on Hells-mediated oocyte methylation are required to elucidate the details of this mechanism.
Previous studies on the cytosine methylation of mtDNA have been highly controversial. A recent study by Shock et al. [53] reported cytosine methylation and hydroxymethylation in mammalian mitochondria. Our results indicated that mtDNA is unmethylated in blastocysts and ESCs, but is partially methylated in germ cells. Whether or not 5-hydroxymethylcytosine (5-hmC) exists in mitochondrial or genomic chromosomes of germ cells remains unclear. Meanwhile, rapid hydroxylation of 5-methylcytosine (5-mC) in the paternal pronucleus during zygotic development was also recently reported [54], [55]. Currently, it is difficult to assess hydroxymethylation profiles in oocyte genomes due to the limited DNA recovery. Further investigation of cytosine modification during germ cell and zygote development will be required in the future to better understand this process.
The DNA methylome maps of mouse germ cells, in this study, were derived from SBS data and, therefore, accurately represent methylation levels of individual CpGs on a whole-genome level. The adaptation of the SBS method for small-scale DNA analysis, described in the present report, has the potential to enable further analyses of germline lineages. The current work examined SBS library construction using 3 methods, MethylC-seq, WBA-seq, and PBAT. MethylC-seq basically required only microograms of DNA [22], [23], [56], thus over amplification might cause redundancy in oocyte libraries. The latter methods allow comprehensive methylome analysis in samples with low amounts of starting DNA by avoiding DNA damage due to sodium bisulfite treatment (after adapter ligation, in the case of MethylC-Seq). Recent studies using BS sequencing have shown that methylated cytosine is abundant in the non-CpG regions of human pluripotent stem cells and mouse oocytes [22], [23], [39], [56]; however, the function of non-CpG methylation in mammalian genomes remains unclear. The PBAT results also showed an abundance of non-CpG methylation in oocytes, with results similar to a previous sequencing study on imprinted loci [39]. However, accurate assessment of non-CpG methylation is required using increased sequencing depths because methylation levels of the non-CpG sites were much lower than those of the CpG sites. SBS library construction was conducted by WBA-seq from 2000 fully matured (metaphase II stage) oocytes; sufficient quantities for sequencing were not obtained. During oogenesis, most of the oocyte specific imprinted methylation marks were established during the GV stage. This contrasted to a previous study where a continuous increase in methylation levels was observed [38]. Further improvement of SBS methods, requiring smaller amounts of DNA, is needed to provide complete germ cell methylome maps and to elucidate the exact function of non-CpG methylation in germ cells.
In conclusion, we constructed the first extensive, high-resolution maps of DNA methylomes of mouse oocytes and sperm. These maps described the epigenetic properties of these DNA methylomes. Our data could serve as a platform for future studies to elucidate the role of epigenetic modifications in the development and functioning of germ and stem cells. Such studies are anticipated to improve our understanding of epigenetic reprogramming.
Five thousand germinal vesicle (GV)-stage oocytes were collected from the ovarian follicles of adult (7- to 9-week-old) female C57BL/6N mice (Clea Japan, Tokyo, Japan) 44–48 h after they were injected with equine chorionic gonadotropin. Three hundred blastocysts at embryonic day 3.5 were obtained from superovulated adult female C57BL/6N mice by flushing the uterus. Genomic DNA was extracted using the QIAamp DNA Mini Kit (Qiagen, Valencia, CA). Sperm were released from the cauda epididymises of adult male C57BL/6N mice. Sperm DNA was isolated by a standard phenol-chloroform extraction procedure with dithiothreitol (DTT). Genomic DNA from 2 lines of ESCs derived from C57BL/6J mice (Clea Japan) was extracted using the DNeasy Blood & Tissue Kit (Qiagen). DNA samples were sheared into 100-bp fragments in oocytes and 200-bp fragments in other samples using the Covaris S2 focused acoustic system (Covaris, Woburn, MA). Cytosine-methylated adapters (Illumina, San Diego, CA) were ligated to DNA by using the Paired-End DNA Sample Prep Kit or ChIP-Seq DNA Sample Prep Kit (Illumina). DNA fragments were isolated by 2–3% agarose gel electrophoresis and purified using the QIAquick Gel Extraction Kit (Qiagen). Sodium bisulfite conversion was performed using the Epitect Bisulfite Kit (Qiagen).
All bisulfite-converted DNA molecules were polymerase chain reaction (PCR)-amplified as follows: 2.5 U of Hot Start Taq polymerase (TaKaRa, Tokyo, Japan), 5 µL 10× PCR buffer, 25 µM dNTPs, 1 µL of each PCR Primer PE 1.0 and 2.0 (Illumina) (50 µL final). Thermocycling parameters were: initial denaturation at 94°C for 1 min, 15–25 cycles of denaturation at 94°C for 30 s, annealing at 65°C for 30 s, and extension at 72°C for 30 s, followed by a final extension at 72°C for 5 min. PCR reaction products were purified using the QIAquick kit (Qiagen).
Two thousand GV-stage oocytes were collected from 7- to 9-week-old female C57BL/6N mice (Clea Japan) and, 2300 GV-stage oocytes were collected from 7–15-week-old Dnmt3L−/− female mice (129SvJae×C57BL/6N hybrid genetic background) [6], [57]. Genomic DNA was extracted using the QIAamp DNA Mini Kit (Qiagen), and then bisulfite-treated with Epitect Bisulfite Kit (Qiagen). Subsequently, the bisulfite-converted DNA was amplified using Epitect Whole Bisulfitome Kit (Qiagen). The collected DNA was sheared into 200-bp fragments using Covaris S2. Unmodified Paired-End adapters (Illumina) were ligated to the DNA by using the Paired-End DNA Sample Prep Kit (Illumina). DNA fragments were isolated by 2% agarose gel electrophoresis and purified using the QIAquick Kit (Qiagen). All DNA was PCR amplified and purified in the same manner as the MethylC-seq method, except the number of PCR cycles was reduced to 7.
GV-stage oocytes (400) and blastocysts (100) were obtained from 7- to 9-week-old female C57BL/6N mice (Clea Japan), and genomic DNA was extracted using the QIAamp DNA Mini Kit (Qiagen). The isolated oocyte and blastocyst genomic DNA and 100 ng of genomic DNA from sperm, blastocysts, and ESCs containing 1∶200 amount of unmethylated lambda DNA (Invitrogen, Carlsbad, CA) were bisulfite-treated using the MethylCode Bisulfite Conversion Kit (Invitrogen). Details of the PBAT method are unpublished [Miura F & Ito T, personal communication]. Briefly, bisulfite-treated DNA were double-stranded using Klenow Fragments (3′- 5′ exo-) (New England Biolabs, Ipswich, MA) with random primers containing 5′ biotin tags and Illumina PE adaptors. The biotinylated molecules (first strand) were captured using Dynabeads M280 Streptavidin (Invitrogen) and double-stranded using Klenow Fragments (3′-5′ exo-) with random primers containing Illumina PE adaptors (second strand). Finally, template DNA strands were synthesized as complementary DNA with a second strand (unmethylated C is converted to T) using Phusion Hot Start High-Fidelity DNA Polymerase (New England Biolabs) with PCR Primer PE 1.0 (Illumina).
Total RNA from 1000 wild-type GV oocytes, 500 Dnmt3L−/− GV oocytes, sperm, and ESCs was extracted using the RNeasy Mini Kit (Qiagen) and treated with DNase I (Promega, Madison, WI). RNA-Seq libraries were constructed using the mRNA-Seq Sample Preparation Kit (Illumina).
The MethylC-seq for blastocysts, WBA-seq, and PBAT libraries were sequenced on a HiSeq 2000 sequencing system (Illumina); the other MethylC-seq and mRNA-seq libraries were sequenced on a Genome Analyzer II (Illumina). Sample preparation, cluster generation, and sequencing were performed using the Paired-End Cluster Generation Kit-HS and the TruSeq SBS Kit-HS for the HiSeq 2000. Similarly, the Paired-End Cluster Generation Kits v2 and v4 and 18- and 36-Cycle Sequencing Kits v3 and v4 were used for the Genome Analyzer II. All kits were from Illumina.
All sequenced reads were processed using the standard Illumina base-calling pipeline (v1.4–1.7). Generated sequence tags were mapped onto the mouse genome (mm9, UCSC Genome Browser, July 2007, Build 37.1) by using the Illumina ELAND program.
MethylC-seq tags (36 or 76 nt) were mapped with a custom Perl program, as described previously [17], [22]. Briefly, all cytosines in the tags were replaced by thymines. Next, these tags were aligned to 2 mouse genome reference sequences (mm9), such that the antisense strand had cytosines replaced by thymines and the sense strand had guanines replaced by adenines. Finally, all tags (32–76 nt) that mapped uniquely without any mismatches to both strands were compiled and used for further analyses.
The 76 nt WBA-seq tags were mapped as follows. All tags were converted to 2 types of reads; in 1 read (“For” read), cytosines were replaced by thymines and in the other read (“Rev” read), guanines were replaced by adenines. Both “For” and “Rev” reads were aligned to sense and antisense mm9 strands. A total of 793, 397, 948, 480, and 238 million tags were aligned in wild-type oocytes, Dnmt3L−/− oocytes, sperm, blastocysts, and ESC genomes, respectively. To avoid bias, tags mapped with multiple hits or matched chromosome M (mitochondria), chromosome Y, or 3 types of repetitive sequences (simple repeat, low complexity repeat, and satellite DNA sequences) were omitted from further analyses.
The 47 nt PBAT tags (trimmed first 4 nt and last 1 nt) were mapped as follows. All guanidines in the tags were replaced by adenines, and these tags were aligned to sense and antisense strands mm9.
For gene-level analysis, the concentrations of the perfectly matching 35 nt (trimmed first nt) mRNA-seq tags from wild-type oocytes, Dnmt3L−/− oocytes, sperm, and ESCs were calculated for the genomic regions corresponding to those covered by the RefSeq transcript models. The expression level of 20,854 unique genes was ranked by expression levels (calculated as RPKM values) in each library (Table S1). A total of 33, 28, 23, and 25 tags were aligned in 4 mRNA-seq libraries, respectively. mRNA-seq data analysis was performed and visualized using GenomeStudio Data Analysis software (Illumina).
The percentage of individual cytosines methylated at all CpG sites covered by at least 1 read was calculated as 100×(number of aligned cytosines (methylated cytosines))/(total number of aligned cytosines and thymines (originally unmethylated cytosines)). All genomic CpG methylation data are available on our website (http://www.nodai-genome.org/mouse_en.html). The CpG and non-CpG (CpH) methylation levels determined by PBAT results were calculated as the ratio between the total read C and the total read T mapped to genomic cytosines. Bisulfite conversion failure rates were calculated by read C∶T ratios from lambda DNA mapping data. The failure rates were as follows: GV oocyte, 0.009; sperm, 0.008; blastocysts, 0.011; and ESCs, 0.006. Locations of transposable elements in the mouse genome (mm9) were obtained from the UCSC Genome Browser, and the average methylation levels of the whole genome and each transposable element were recalculated from the ratio of the aligned cytosines and thymines in each sequence. Lists of 23,021 CGIs were obtained from a previous report [37]. Around the TSS and TTS (±5 kb), genomic regions were divided into 20-bp bins. For each bin, the average methylation value was calculated for each gene. The expression level of 20,854 genes was divided into 5 percentile groups ranked by RPKM values, and the average methylation level for each group was mapped onto the gene structure model. These computational analyses were performed using a custom Perl program. Supercomputing resources were provided by the Human Genome Center, Institute of Medical Science, University of Tokyo.
Correlations between gene expression ranks and average methylation levels in the promoter (±500 bp from the TTS) or gene-body regions (gene-body 1: +2 to +5 kb from the TSS; gene-body 2: 0 to −5 kb from the TTS) were calculated using Spearman's rank correlation coefficient (ρ). An R-squared value (R2) was calculated to evaluate the correlation of RPKM values between wild-type and Dnmt3L−/− oocytes. Statistical analysis was performed using the R statistical package.
To analyze the methylation of the three transposable elements (L1 LINE, B1/Alu SINE, and IAP LTR), 20 wild-type GV oocytes were obtained from adult female C57BL/6N mice. Bisulfite sequencing conditions and primer sets for the three transposable elements were described, previously [52]. To analyze the methylation of the Gpr1 locus, 10 blastocysts were obtained from BJF1 (C57BL/6N×JF1) and Dnmt3Lmat−/− (Dnmt3L−/−×JF1) mice [6], [57]. Genomic DNA from blastocysts was isolated using the QIAamp DNA Mini Kit (Qiagen) and treated with sodium bisulfite with the Epitect Bisulfite Kit (Qiagen). The Gpr1 gDMR sequence was amplified with 2 rounds of nested PCR. The first-round PCR reaction contained 1 U of Hot Start Taq polymerase (TaKaRa), 1× PCR buffer, 200 µM dNTPs, 1 µM forward primer, and 1 µM reverse primer (20 µL final). Thermocycling parameters were as follows: initial denaturation at 94°C for 1 min, 35 cycles of denaturation at 94°C for 30 s, annealing at 50°C for 30 s, and extension at 72°C for 30 s, followed by a final extension at 72°C for 5 min. Subsequently, 2 µL of the product was used as the input for the second-round PCR, which was performed in the same manner. Primer sets for the nested PCR were as follows: Gpr1-BSF1 (5′-GATTAGATTAGGTTAGTTTGGAA-3′) and Gpr1-BSR1 (5′-ACTAAAACACTAATCACCAAATA-3′) for the first round; Gpr1-BSF2 (5′-AGATTAGGTTAGTTTGGAATT-3′) and Gpr1-BSR2 (5′-AACACTAATCACCAAATAATTC-3′) for the second round. The second-round PCR product was subcloned and sequenced, as described previously [50]. The percentage methylation was calculated as 100×(number of methylated CpG dinucleotides)/(total number of CpGs). At least 10 clones from each parental allele were sequenced. Sequence data were analyzed using the QUMA quantification tool for methylation analysis [58].
The MethylC-seq, WBA-seq, PBAT, and mRNA-seq data in this study have been deposited in the DNA Data Bank of Japan (DDBJ) under accession number DRA000484.
|
10.1371/journal.pmed.1002197 | Histological Transformation and Progression in Follicular Lymphoma: A Clonal Evolution Study | Follicular lymphoma (FL) is an indolent, yet incurable B cell malignancy. A subset of patients experience an increased mortality rate driven by two distinct clinical end points: histological transformation and early progression after immunochemotherapy. The nature of tumor clonal dynamics leading to these clinical end points is poorly understood, and previously determined genetic alterations do not explain the majority of transformed cases or accurately predict early progressive disease. We contend that detailed knowledge of the expansion patterns of specific cell populations plus their associated mutations would provide insight into therapeutic strategies and disease biology over the time course of FL clinical histories.
Using a combination of whole genome sequencing, targeted deep sequencing, and digital droplet PCR on matched diagnostic and relapse specimens, we deciphered the constituent clonal populations in 15 transformation cases and 6 progression cases, and measured the change in clonal population abundance over time. We observed widely divergent patterns of clonal dynamics in transformed cases relative to progressed cases. Transformation specimens were generally composed of clones that were rare or absent in diagnostic specimens, consistent with dramatic clonal expansions that came to dominate the transformation specimens. This pattern was independent of time to transformation and treatment modality. By contrast, early progression specimens were composed of clones that were already present in the diagnostic specimens and exhibited only moderate clonal dynamics, even in the presence of immunochemotherapy. Analysis of somatic mutations impacting 94 genes was undertaken in an extension cohort consisting of 395 samples from 277 patients in order to decipher disrupted biology in the two clinical end points. We found 12 genes that were more commonly mutated in transformed samples than in the preceding FL tumors, including TP53, B2M, CCND3, GNA13, S1PR2, and P2RY8. Moreover, ten genes were more commonly mutated in diagnostic specimens of patients with early progression, including TP53, BTG1, MKI67, and XBP1.
Our results illuminate contrasting modes of evolution shaping the clinical histories of transformation and progression. They have implications for interpretation of evolutionary dynamics in the context of treatment-induced selective pressures, and indicate that transformation and progression will require different clinical management strategies.
| Follicular lymphoma (FL) is a largely incurable malignancy in which early progression and transformation have consistently been linked to lymphoma-related mortality.
We contended that detailed characterization of clonal dynamics would reveal fundamental biological properties with implications for future patient management strategies relating to both transformation and progression.
We also sought to identify recurrent gene mutations associated with transformation and/or early progression in a large patient cohort.
Using whole genome sequencing, deep allelic sampling by amplicon sequencing, and digital droplet PCR, we found dramatic clonal expansions in transformed disease, whereby dominant clones in transformation samples emerged from extremely low prevalence clones or from clones that were not detected in the diagnostic samples.
The dynamics of disease progression during treatment in the absence of transformation showed markedly different characteristics, with much of the clonal architecture preserved from diagnostic to relapse specimens.
Targeted capture-based sequencing in a large extension cohort then established genetic variants associated with transformation and early progression in the broader patient population.
Taken together, our findings illuminate previously undescribed patterns of clonal expansion underpinning FL clinical histories suggesting that contrasting management strategies will be necessary across the FL patient population.
We uncovered novel associations of gene mutations with early progression that could inform future prognostic assay development.
| Follicular lymphoma (FL) is the second most common subtype of non-Hodgkin lymphoma and the most frequent indolent lymphoma, accounting for 22%–32% of all new non-Hodgkin lymphoma diagnoses in Western countries [1,2]. Patient outcomes are favorable, with median overall survival extending well beyond 10 y [3–5]. However, FL remains an incurable malignancy as most patients eventually experience progressive disease. A subset of patients are at risk of early lymphoma-related mortality due to early progression after immunochemotherapy or to histological transformation to aggressive lymphoma (2%–3% of patients per year), both of which lead to shortened survival [6–13]. Hence, mutational profiling of FL specimens at the temporal boundaries of clinical inflection points represents a compelling opportunity to study the evolutionary dynamics underpinning FL disease progression.
To infer evolutionary properties, deconvolution of malignant tissues into constituent clones is required. Clonal decomposition is accomplished through analysis of allelic measurements, under the assumption that the prevalence of specific alleles in a DNA mixture extracted from a tumor quantitatively represents clonal population abundance. Here, we brought to bear targeted amplicon sequencing plus ultra-sensitive digital droplet PCR in order to measure the changing prevalence of alleles at unprecedented resolution over FL disease progression. With precise measurements of alleles, computational inference can then determine clonal composition and the phylogenetic topology of clones, yielding insight into temporal mutation acquisition and genotypes giving rise to clonal expansions over time. With this approach, longitudinal comparison of the clonal composition of tumors sampled at different time points in patient’s clinical history can be performed, deciphering which constituent populations were present at diagnosis, and which populations constituted the relapse. Thus, the degree to which a tumor is evolving and the contributions of specific clones to the evolutionary process (collectively termed clonal dynamics) can be quantitatively assessed.
To varying levels of resolution, related approaches have been applied to a variety of progression scenarios in hematologic and solid malignancies [14–17]. For example, secondary acute myeloid leukemia from underlying myelodysplastic syndrome and Richter syndrome from chronic lymphocytic leukemia arise without significant branched evolution [18,19]. By contrast, transformation of FL has most commonly been described as divergent branched evolution from a common progenitor [20,21]. The nature of clonal trajectories leading to transformation or early progression are poorly understood; it is unknown if similar, or contrasting, modes of selection underpin these clinical end points.
Discrete transformation-associated genetic alterations have been described involving CDKN2A, MYC, TP53, CD58, or B2M [20–29]. However, these events alone cannot explain the majority of transformed cases, leaving a discovery gap for genetic drivers of transformation. Similarly, progression has been described to occur more frequently in the presence of selected, recurrent cytogenetic aberrations or single gene mutations [30–36]. Recently, a clinicogenetic risk model (m7-FLIPI), including the mutational status of seven genes, the Follicular Lymphoma International Prognostic Index (FLIPI), and performance status, was shown to improve outcome prediction for patients requiring immunochemotherapy [37]. Nonetheless, the m7-FLIPI imperfectly captures determinants of early progression [13]. A newer prognostic model, named POD24-PI, was developed using the original m7-FLIPI data to specifically predict early progression. The POD24-PI has better sensitivity but lower specificity for the prediction of early progression [13], raising the question of whether progressive disease might be attributed to genetic lesions that are not captured by either prognostic model. Furthermore, the mechanisms underlying resistance to immunochemotherapy remain elusive; genetic profiling of early progression cases has the potential to uncover novel genetic lesions in molecular pathways leading to treatment resistance.
To address these questions, we set out to compare the clonal dynamics of tumors leading to transformation and those associated with early progression. We executed in-depth, high-resolution genome-wide profiling of mutant alleles. In addition, we aimed to establish the patient population prevalence of genetic events associated with transformation and early progression through targeted sequencing of a large cohort of samples with accompanying clinical outcome data.
Patient specimens were collected as part of research projects approved by the research ethics boards of the University of British Columbia–British Columbia Cancer Agency (H13-01765), UZ Leuven (S-55498), or the Mayo Clinic (08–005005).
We assembled a cohort of tumor and normal specimens from 41 patients selected for whole genome sequencing (WGS) (Figs 1 and S1). Samples were acquired from the BC Cancer Agency lymphoma tumor bank, and patients were grouped according to three clinical end points: patients who presented with transformation (transformed FL [TFL], n = 15), those whose disease progressed without evidence of transformation (progressed FL [PFL], n = 6), and those whose lymphoma displayed no evidence of transformation or progression for more than 5 y after initial diagnosis (non-progressed FL [NPFL], n = 20). Paired tumor samples from fresh frozen blocks or cell suspensions consisting of diagnostic and relapse specimens for TFL and PFL patients and single diagnostic specimens for NPFL patients were acquired. We refer to samples from the primary time point as T1 samples, and those from the time of transformation (TFL cases) or progression (PFL cases) as T2 samples. Tumor and normal specimens comprising 103 WGS libraries in total were sequenced, yielding 62 tumor samples sequenced to an average 62.7-fold ± 23.0-fold coverage, and matching germline DNA sequenced to an average 35.2-fold ± 12.8-fold coverage (S2 Fig; S1 Table). Somatic single nucleotide variants (sSNVs), somatic small insertions and deletions (sIndels), somatic copy number alterations (sCNAs), and structural rearrangements were predicted for each tumor sample as described in S1 Supporting Appendix.
We next constructed a larger extension cohort consisting of samples from 277 patients used for targeted, capture-based sequencing (Fig 1). Patients were grouped into three categories: patients presenting with transformation (n = 159) and patients presenting with either early disease progression (n = 41) or late/never progression (n = 84). Early progression was defined as progression occurring within 2.5 y after starting treatment, which was intended to consist of rituximab chemotherapy followed by rituximab maintenance. Late/never progression was defined as no progression of lymphoma for at least 5 y after initiation of either observation or rituximab chemotherapy and rituximab maintenance. The majority of samples (96%) were acquired from the BC Cancer Agency lymphoma tumor bank, and a smaller number from the Mayo Clinic (2%) and the University of Leuven (2%). DNA from the extension cohort was subjected to a hybrid-capture-based panel of 94 genes and was sequenced to 1046.96-fold ± 229.88-fold coverage for fresh frozen samples and to 192.56-fold ± 120.49-fold coverage for formalin-fixed paraffin-embedded tissue samples.
Complete information on patient cohorts and sample preparation can be found in S2 Table and in S1 Supporting Appendix. All sequencing data are available for download through the European Genome-phenome Archive under accession number EGAS00001001709.
Detailed bioinformatics methods are presented in S1 Supporting Appendix. Briefly, WGS data were processed to provide sSNV, sIndel, sCNA, and structural rearrangement predictions. For inference of clonal structure, we selected ≥192 sSNVs or sIndels per patient and performed targeted deep amplicon sequencing, providing precise allelic measurements. A subset of mutations were profiled using digital droplet PCR. Those data, together with copy number status and tumor content, were used as an input for inference of clonal dynamics using previously described computational techniques [38,39].
In the extension cohort, we sequenced, using capture-based sequencing, the coding sequence of 86 genes as well as the 5′ regions of 20 genes that are targets of somatic hypermutation (12 genes overlapping with the 86 previously mentioned genes, i.e., 94 genes in total). sSNVs and sIndels were called as described in S1 Supporting Appendix. The proportions of samples harboring somatic mutations were compared between clinical groups using Bayesian proportion tests.
We began our analysis by comparing mutational burden over time in T1 and T2 samples from the WGS cohort. At T1, the average number of alterations was 7,133.29 ± 3,107.02 (range 2,184–21,802) for sSNVs, 512.63 ± 296.67 (range 70–1,801) for sIndels, and 26.24 ± 21.28 (range 4–112) for structural rearrangements across all WGS tumor samples (Fig 2). The mutational burden was significantly higher in T2 than in T1 samples for all mutation types (in both TFL and PFL patients) (Fig 3A) and was independent of the time interval between sampling (S3 Fig).
When comparing the three clinical groups (TFL, PFL, and NPFL), baseline mutation rates at T1 did not differ for sSNVs, sIndels, and sCNAs (T1 facet of Fig 3B), suggesting that the increase in mutation rate for TFL and PFL cases was acquired after diagnosis. However, the number of structural rearrangements was higher in TFL T1 samples (31.33 ± 23.29) than in PFL T1 samples (17.00 ± 8.88) and NPFL samples (16.90 ± 13.76) (Kruskal-Wallis p = 0.026; T1 facet of Fig 3B), consistent with TFL cases at diagnosis harboring an increased propensity to accumulate translocations. Comparison of TFL and PFL T2 samples revealed a higher number of sIndels (one-tailed Wilcox p < 0.001), a higher proportion of the genome altered by sCNAs (one-tailed Wilcox p = 0.018), and a higher number of rearrangements in TFL samples compared to PFL samples (one-tailed Wilcox p = 0.028; T2 facet of Fig 3B), suggesting that histological transformation is associated with a higher mutational rate in the structural genome relative to samples that progressed on therapy. Overall, a higher mutational burden in T2 samples relative to T1 samples was observed, with a more pronounced effect in TFL cases.
We next profiled the clonal composition and evolutionary changes of T1 and T2 samples. All T1–T2 pairs exhibited uniclonal origin by virtue of shared mutations comprising an ancestral clone in addition to a substantial fraction of T1-specific (0.175 ± 0.105 [minimum–maximum, 0.038–0.431]) and T2-specific mutations (0.366 ± 0.166 [0.063–0.664]) (S4 Fig, contour density on T1 and T2 axes). However, comparative analysis of the clonal structure of T1 and T2 samples (S1 Supporting Appendix; S4 Table) revealed dramatic clonal dynamics in 13 of 15 TFL patients (87%). In these 13 patients, T2 samples were composed primarily of divergent clones (or phylogenetic lineages) that were extremely rare (<1%) in T1 samples (Figs 4 and S5). This defined a characteristic mode of evolution with massive expansion of clones in T2 samples that were rare or detectably absent in T1 samples. This suggests that diagnostic samples are not likely to possess reliable predictors of transformation in the majority of cases, and that the clonal dynamics occurring after diagnosis likely underpin histological change. This pattern was independent of time to transformation. For example, the T2 sample from FL1007 (transformed after 14.57 y; Fig 4), characterized by FOXO1 and BCL6 mutations in the ancestral clone (cluster 1), was entirely composed of a clonal lineage harboring B2M and CCND3 mutations (clusters 2 and 3) that were near zero prevalence levels in the T1 sample. Notably, these clones were mutually exclusive to the clonal lineage dominating the T1 sample (clusters 4, 7, 6, and 5). The T2 sample from FL1017 (transformed after 0.42 y; Fig 4), characterized by CREBBP and KMT2D mutations in its ancestral clone, harbored a T2-specific lineage containing EZH2 and FOXO1 mutations (clusters 2 and 1), exhibiting a distribution of clones similar to that of FL1007. This pattern of clonal dynamics was independent of treatment regimen and was found in untreated cases (observation alone; FL1007, FL1006, FL1012, FL1014, and FL1019) and in cases treated with rituximab and/or chemotherapy (FL1001, FL1004, FL1005, FL1008, FL1013, FL1016, and FL1017). The pattern of expansion from undetectable or extremely rare clones (<1%) was validated using orthogonal digital droplet PCR technology (S1 Supporting Appendix) in 3/3 TFL cases attempted, confirming that a clone as rare as 2 out of approximately 105 cells at diagnosis came to dominate the transformed specimen (Figs 5A, 5B and S6A–S6C). We also observed this signal in the extension cohort in 18 cases out of 32 (56%) that were available for analysis and not overlapping with our WGS cohort (S7 Fig). These observations were made from a sparse sampling of only 94 genes and yet still yielded similar patterns where at least one mutation exhibited increased prevalence from near zero in the T1 sample to dominant levels in the T2 sample.
Two TFL cases (13%) exhibited clonal dynamics that contrasted with the dominant pattern. In these cases (FL1009 and FL1020—both untreated and both with relatively short times to transformation, 0.39 y and 0.78 y, respectively), the dynamic properties showed conserved clonal architecture (FL1009) or only modest dynamics (FL1020). Thus, a small minority of cases may already contain the properties driving transformation at the time of diagnosis. Together, these results reveal a striking pattern of clonal dynamics underpinning histological transformation in the majority of TFL cases, independent of time to transformation and treatment regimen.
Progressed samples exhibited patterns of clonal dynamics markedly different from those of transformed cases (Fig 6). Four cases (FL2002, FL2005, FL2007, and FL2008) harbored readily detectable clones at T1, which expanded to full clonal prevalence during treatment with immunochemotherapy. This suggests that clones harboring treatment resistance properties were already present at diagnosis, and that symptomatic disease progression may be attributed to selection of clones that were major constituents of the diagnostic sample. This mode of progression is reminiscent of the clonal evolution described in chronic lymphocytic leukemia, another mature, incurable, and typically relapsing lymphoid malignancy [15,16]. FL2006 showed a slightly different pattern whereby the ancestral clone dominated the T1 and T2 samples but was accompanied by modest dynamics, including expansion of a low prevalence clone (cluster 2) in the T2 sample. An exceptional case (FL2001) in the PFL group exhibited dynamics similar to those of TFL cases (validated with digital droplet PCR; Figs 5C, S6D and S6E), with a T2-specific lineage with ARID1A mutation (clusters 2 and 3) coming to dominate the relapse sample and with no evidence of the T1 clones (clusters 4 and 5). This patient initially presented with indolent FL, received single agent rituximab, and presented 4 y after diagnosis with symptomatic, progressive lymphoma unresponsive to three lines of systemic therapy, leading to the patient’s death. In this case, the phylogenetic structure was analogous to the TFL pattern, yet the biopsy from T2 showed no evidence of large cell transformation. Thus, treatment resistance patterns accompanied by significant clonal dynamics can occur in FL in the absence of overt transformation. PFL clonal dynamics suggest that progression on therapy is driven by a starkly different mode of evolution than what was observed for TFL. These two clinical end points are likely underpinned by non-overlapping evolutionary mechanisms, with PFL harboring intrinsically resistant properties at diagnosis and TFL generally acquiring the dominant transformation phenotype after diagnosis.
We next sought to quantify the statistical likelihood of observing the clonal expansion of an extremely rare clone at T1 (<1%) into a dominant clone at T2 (>50%) under the assumption of neutral evolutionary dynamics for TFL patients. We modeled drift in 1,000 independent simulations under the Wright-Fisher process, to simulate the pattern of allelic “drift” without selection in asexually reproducing systems. The majority (88.1%) of the simulations resulted in an eventual loss of the mutant allele (cluster 1 of Fig 7A). Conversely, only six (0.6%; cluster 2) of the simulations exhibited a trajectory similar to the clonal expansion patterns observed in the TFL patients. As such, observing this clonal expansion pattern in 13 out of 15 TFL patients is statistically unlikely (binomial exact test p < 0.001) when assuming 0.6% as the expected trajectory rate. Modeling drift in PFL starting with a dominant clone at T1, the simulations demonstrate trajectories that are consistent with the observed patterns of evolution in PFL patients (Fig 7B). These results are consistent with the notion that histological transformation is driven through positive selection in the T1–T2 interval in TFL patients. In contrast, the clonal dynamic patterns between T1 and T2 in PFL patients are consistent with Wright-Fisher allelic drift without selection, suggesting that clones at T1 are not expanding under positive selection, despite treatment intervention.
Evolutionary analysis suggested several patterns of mutation acquisition (S8 Fig) including TNFRSF14, CREBBP, and GNA13 mutations as predominantly ancestral (in the top level node of the clone phylogeny in 7/7, 12/13, and 5/5 mutations, respectively). KMT2D, BCL6, HIST1H1E, and EZH2 mutations showed evidence of being ancestral in some cases and descendant (lower than the ancestral node) in others. The plasticity across ancestral and descendant states for recurrent gene mutations prompted us to resolve their etiology in a larger series of cases (extension cohort; 395 genomic DNA samples [T1 or T2] from 277 patients) and assess their roles in transformation and early progression (S5 and S6 Tables). Ninety-four genes were sequenced in this cohort (see S1 Supporting Appendix).
We first compared T1 (n = 128) and T2 (n = 149) samples of transformed cases from 159 patients (118 paired biopsies). Similar to our findings from WGS (Fig 2B), mutational load in 86 genes in which the entire coding sequence was assessed was higher in T2 than in T1 samples (mean number of mutated genes 12.47 ± 6.80 versus 9.39 ± 5.74, Student t test p < 0.001) (S9 Fig). Mutation burden in the 5′ regions of 20 genes that are targets of somatic hypermutation did not significantly differ between the T1 and T2 samples, with the exception of MYC and TMSB4X (S10 Fig). We determined which genes had a higher likelihood of being mutated in T2 compared to T1 using a Bayesian proportion test and found 12 genes to be more commonly altered in transformed lymphoma (Fig 8A). These included previously described genes associated with transformation, such as TP53, B2M, MYC, and EBF1, as well as novel genes (e.g., EZH2, CCND3, PIM1, and ITPKB). B2M mutations were associated with a significantly reduced CD8+ T cell infiltrate in transformed lymphoma biopsies (Fig 8B and 8C). Moreover, mutations in GNA13, S1PR2, and P2RY8, all implicated in dissemination of germinal center B cells [40], were enriched in T2 samples. These findings suggest that defective DNA damage response, increased proliferation, escape from immune surveillance, and loss of confinement within the germinal center represent key features that drive histological transformation from indolent to aggressive lymphoma.
We next overlaid mutation status with detailed histological annotation. Composite morphology was associated with a lower prevalence of TP53 mutations (8% versus 37%, Fisher p = 0.007) relative to diffuse large B cell lymphoma morphology. In addition, cell-of-origin classification was available for 108 cases with diffuse large B cell lymphoma histology, 18 and 90 of which were ABC and GCB subtype, respectively. More BCL10 (16% versus 1%, Fisher p = 0.004), CD79B (22% versus 4%, Fisher p = 0.005), and MYD88 mutations (28% versus 9%, p = 0.006) were observed in ABC TFL relative to GCB TFL (Fig 8D), suggesting that B cell receptor and NF-κB signaling are important contributors to the ABC phenotype in TFL.
Next, we assessed the association of gene mutations with patient outcome, contrasting patients with early progression (<2.5 y after starting rituximab chemotherapy) (n = 41) and patients with late/never progression (no progression for >5 y) (n = 84). Samples from patients with early progression were enriched for high-risk clinical factors including poor performance status, tumor mass ≥ 7 cm, elevated lactate dehydrogenase, and high-risk FLIPI score (S6 Table). Median overall survival was extremely poor in these patients (3.01 y versus not reached in patients with late/never progression, log-rank p < 0.001; S11 Fig), highlighting the critical need for identifying these patients upfront.
Overall, the burden of somatic hypermutation was not significantly different in samples from patients with early versus late/never progression, but samples from patients with early progression had more mutations per sample in BACH2, BTG2, RHOH, and SOCS1, and fewer mutations in LTB, when compared to patients with late/never progression (S12 Fig). Patients with early progression had, on average, a higher mutation load in those 86 genes in which the entire coding sequence was assessed, when compared to patients with late/never progression (13.44 ± 9.17 versus 9.75 ± 5.71, Student t test p = 0.022) (S13 Fig). Ten genes were mutated more commonly in patients with early progression than in patients with late/never progression, including KMT2C, TP53, BTG1, MKI67, XBP1, and SOCS1 (Fig 9A). Only MEF2C was more commonly mutated in patients with late/never progression. Overall, 33 out of 41 patients with early progression (80%) had mutations in any of the ten early-progression-associated genes, but none of the early-progression-associated genes were mutated at a frequency > 27% (Fig 9B). Thus, early progression appears to be related to relatively infrequent genetic alterations. Furthermore, none of the early-progression-associated gene mutations form part of the m7-FLIPI outcome predictor, and, in our cohort that was enriched for clinical extremes, the m7-FLIPI was similarly associated with early progression when compared with the FLIPI, but not superior, having better specificity (88% versus 76%) but worse sensitivity (36% versus 63%). Taken together, our results identify early progression as a distinct clinicogenetic disease category that is imperfectly captured by traditional prognostic tools.
We established that transformation and progression in FL are driven by disparate modes of evolutionary change. Shown schematically in Fig 10, TFL is characterized by the emergence of clones that become dominant at T2 and that typically lie below the detection limit of even highly sensitive methods at the T1 (FL) time point (Fig 10A), implying that the aggressive phenotypes emerge after diagnosis. By contrast, early progression of FL commonly results from prevalent clones at T1, such that much of the clonal architecture is maintained despite treatment, implying that resistant properties are well established at diagnosis (Fig 10B). The content of gene mutations associated with transformation and early progression also differed. We found novel associations of gene mutations with transformation (including CCND3, GNA13, S1PR2, and P2RY8 mutations) and showed that TFL is molecularly heterogeneous, with, for example, the ABC subtype of TFL being enriched for BCL10, CD79B, and MYD88 mutations. Genes with recurrent mutations associated with early progression included KMT2C, TP53, BTG1, and MKI67. Thus, transformation and progression can be attributed to disruption of different biological processes.
Our study has several limitations. The small number of cases assessed in the WGS cohort, the lack of reliable copy number information in our extension cohort, and the absence of an additional validation cohort to confirm the prognostic implication of gene mutations associated with early progression provide direction for future complementary follow-up studies. Our findings are nonetheless of critical translational relevance. The divergent modes of evolution of PFL and TFL mirror distinct differences in the clinical presentation of these entities, with transformation being uniquely associated with rapid onset of tumor growth and systemic symptoms, suggesting an underlying shift in tumor biology. As the nature of expansion appears to correlate with rapid-onset symptoms, more granular monitoring of these patients would help to determine the exact timing of the evolutionary inflection point. Furthermore, the defining genetic features of transformation may remain elusive at diagnosis, and at best will require ultra-sensitive detection techniques in order to develop predictive assays. Technical improvements in limits of detection may yet reveal that T2 alleles are always detectable in T1 samples, but our deeply sampled data presented here indicate this may remain a challenge and cannot rule out the possibility that T2 clones arise after diagnosis in some cases. Conversely, primary resistance to upfront combined modality therapy generally occurs by the selection of resistant clones readily found at diagnosis, suggesting that their detection may predict resistance to treatment. In that regard, samples from patients who experience early progression harbor relatively uncommon gene mutations that are associated with early progression (e.g., KMT2C, TP53, BTG1, MKI67, and XBP1 mutations), most of which have not previously been described to predict progression.
Our results have fundamental implications for the study of tumor evolution. Paradoxically, several patients who were managed solely with observation exhibited punctuated clonal dynamics, whereas PFL patients who were treated with multi-agent therapy exhibited relative stability in their clonal make-up. This implies that the evolutionary processes driving FL may be independent of selective pressures imposed by treatment regimens. The association of known driver events (such as CCND3 mutations) with transformation suggests that such punctuated expansions typical of transformation are under positive selection. This argues against fixation under neutral selection models, which would suggest gradual shifts over protracted periods, for example, under the assumption of emergent neutrality [41]. Rather, in transformation, it is likely that specific alleles overcome offsetting interactions between beneficial and deleterious mutations acquired over time due to increased fitness. Indeed, three cases in the WGS cohort with CCND3 descendant mutations had widely varying time to transformation: 14.57 y, 5.05 y, and 2.56 y. These mutations showed variant allele frequencies of 0, 0.002, and 0, respectively, at diagnosis and thus emerged from extremely rare populations. Learning precisely how alleles such as CCND3 mutations exhibit epistatic interactions and modify the effect of founder events such as the t(14:18) translocation to confer higher fitness will be critical to elucidating the mechanism of histological transformation. The pattern is dramatically different in progression, where we might expect clonal dynamics in the presence of a shifting fitness landscape induced by therapy. Rather, clonal architecture at diagnosis remains relatively constant, suggesting that fitness could be attributed to non-genetic factors or that these tumors acquire resistance properties very early in their evolutionary histories and in the absence of therapeutic selective pressure.
Our results place transformation and progression in FL at the extremes of the clonal population dynamics spectrum, at once informing future management strategies and stimulating deeper questions on how FLs mechanistically navigate varied fitness landscapes.
|
10.1371/journal.ppat.1002626 | Methicillin Resistance Alters the Biofilm Phenotype and Attenuates Virulence in Staphylococcus aureus Device-Associated Infections | Clinical isolates of Staphylococcus aureus can express biofilm phenotypes promoted by the major cell wall autolysin and the fibronectin-binding proteins or the icaADBC-encoded polysaccharide intercellular adhesin/poly-N-acetylglucosamine (PIA/PNAG). Biofilm production in methicillin-susceptible S. aureus (MSSA) strains is typically dependent on PIA/PNAG whereas methicillin-resistant isolates express an Atl/FnBP-mediated biofilm phenotype suggesting a relationship between susceptibility to β-lactam antibiotics and biofilm. By introducing the methicillin resistance gene mecA into the PNAG-producing laboratory strain 8325-4 we generated a heterogeneously resistant (HeR) strain, from which a homogeneous, high-level resistant (HoR) derivative was isolated following exposure to oxacillin. The HoR phenotype was associated with a R602H substitution in the DHHA1 domain of GdpP, a recently identified c-di-AMP phosphodiesterase with roles in resistance/tolerance to β-lactam antibiotics and cell envelope stress. Transcription of icaADBC and PNAG production were impaired in the 8325-4 HoR derivative, which instead produced a proteinaceous biofilm that was significantly inhibited by antibodies against the mecA-encoded penicillin binding protein 2a (PBP2a). Conversely excision of the SCCmec element in the MRSA strain BH1CC resulted in oxacillin susceptibility and reduced biofilm production, both of which were complemented by mecA alone. Transcriptional activity of the accessory gene regulator locus was also repressed in the 8325-4 HoR strain, which in turn was accompanied by reduced protease production and significantly reduced virulence in a mouse model of device infection. Thus, homogeneous methicillin resistance has the potential to affect agr- and icaADBC-mediated phenotypes, including altered biofilm expression and virulence, which together are consistent with the adaptation of healthcare-associated MRSA strains to the antibiotic-rich hospital environment in which they are frequently responsible for device-related infections in immuno-compromised patients.
| The acquisition of mecA, which encodes penicillin binding protein 2a (PBP2a) and methicillin resistance, by Staphylococcus aureus has added to an already impressive array of virulence mechanisms including enzyme and toxin production, biofilm forming capacity and immune evasion. And yet clinical data does not indicate that healthcare-associated methicillin resistant S. aureus (MRSA) strains are more virulent than their methicillin-susceptible counterparts. Here our findings suggest that MRSA sacrifices virulence potential for antibiotic resistance and that expression of methicillin resistance alters the biofilm phenotype but does not interfere with the colonization of implanted medical devices in vivo. High level expression of PBP2a, which was associated with a mutation in the c-di-AMP phosphodiesterase gene gdpP, resulted in these pleiotrophic effects by blocking icaADBC-dependent polysaccharide type biofilm development and promoting an alternative PBP2a-mediated biofilm, repressing the accessory gene regulator and extracellular protease production, and attenuating virulence in a mouse device-infection model. Thus the adaptation of MRSA to the hospital environment has apparently focused on the acquisition of antibiotic resistance and retention of biofilm forming capacity, which are likely to be more advantageous than metabolically-expensive enzyme and toxin production in immunocompromised patients with implanted medical devices offering a route to infection.
| Infections caused by healthcare-associated Staphylococcus aureus and methicillin resistant S. aureus (MRSA) pose a major threat to hospital patients. A significant risk factor for these healthcare-associated infections is the extensive use of implanted prosthetic biomaterials for diagnostic and therapeutic purposes, which can be colonized by staphylococci giving rise to device-related infections (DVIs) involving biofilms [1]. In addition to resistance to β-lactam antibiotics such as oxacillin, current chemotherapeutics for DVIs have limited effectiveness against biofilms.
The challenge of developing therapeutics to treat staphylococcal biofilm infections is compounded by the existence of multiple biofilm mechanisms in both S. aureus and S. epidermidis. Thus, although production of the exopolysaccharide polysaccharide intercellular adhesin (PIA) or polymeric N-acetyl-glucosamine (PNAG) synthesized and exported by proteins encoded by the icaADBC genes is common among clinical isolates of both species [2], [3], [4], [5], [6], ica-independent biofilm production has also been described under in vitro conditions [1]. Using clinical isolates of S. aureus, we reported that methicillin resistant S. aureus (MRSA) strains express an icaADBC-independent biofilm phenotype in vitro [3], [4], which is instead dependent on the fibronectin binding proteins (FnBPA and FnBPB) and the major autolysin (Atl) [6], [7]. Atl-dependent autolytic activity and extracellular DNA release are involved in the early stages of biofilm production by these MRSA isolates, whereas the FnBPs promote subsequent intercellular accumulation and biofilm maturation [6], [7]. Unlike MRSA, clinical isolates of methicillin susceptible S. aureus (MSSA) express a PNAG-dependent biofilm phenotype on hydrophilic surfaces and an Atl/PNAG-dependent biofilm on hydrophobic surfaces. Other staphylococcal surface proteins implicated in biofilm include the biofilm-associated protein (Bap, in bovine S. aureus isolates), accumulation-associated protein (Aap) of S. epidermidis and its S. aureus homologue SasG [8], [9], [10], [11], protein A [12], SasC [13] and the extracellular matrix binding protein (Embp) of S. epidermidis [14]. The growing number of bacterial factors involved in staphylococcal biofilm development underscores the importance of this phenotype to the pathogen and suggests that there may be redundancy between biofilm mechanisms in different clinical isolates or on different surfaces.
The methicillin resistance gene mecA encodes the low affinity penicillin binding protein 2a carried on a mobile staphylococcal cassette chromosomal mec element (SCCmec) [15] of which eight different types have been characterized to date [16]. Heterogeneity is a feature of S. aureus methicillin resistance [17], [18], [19]. Many S. aureus clinical isolates exhibit heterogeneous methicillin resistance (HeR) under laboratory growth conditions. In a HeR strain the majority of cells grown in the presence of a β-lactam antibiotic are susceptible to low concentrations of the drug, with only a subpopulation expressing higher-level resistance. However HeR strains become capable of expressing homogeneous resistance (HoR) after selection on elevated concentrations of β-lactam antibiotics or under specific growth conditions [20]. This transition from HeR to HoR is complex with mutations at the fem (factor essential for methicillin resistance), aux (auxiliary) and tagO loci all being implicated [21], [22], [23]. In addition, an oxacillin-induced increased SOS response was shown to increase the mutation rate during HeR to HoR selection in a mechanism dependent on the accessory gene regulator Agr [24], [25]. Nevertheless because HoR clinical isolates are not deficient in any of these accessory factors and because mutations at these loci alone are insufficient to explain HeR to HoR selection, the mechanism underpinning this phenomenon is clearly complex.
SCCmec elements can also carry resistance genes for other antibiotics and heavy metals as well as the psm-mec locus, which encodes a cytolysin termed phenol-soluble modulin-mec (PSM-mec) [26]. Carriage of the psm-mec locus from type II SCCmec elements attenuates virulence, suppresses colony spreading activity, reduces expression of the chromosomally encoded PSMα and promotes biofilm formation [26], [27], [28]. Furthermore both the psm-mec encoded RNA and the PSM-mec peptide contribute to the pleiotropic role of this locus [27], [28].
Our analysis of S. aureus clinical isolates identified a novel biofilm phenotype expressed by MRSA clinical isolates in which the major cell wall autolysin Atl and the fibronectin-binding proteins FnBPA and FnBPB have fundamental functions [3], [6], [7]. The Atl/FnBP biofilm phenotype appears to be absent or less prevalent among methicillin-susceptible S. aureus (MSSA) isolates, which produce PNAG-dependent biofilms in vitro. Interestingly the psm-mec locus from a type II SCCmec element increased expression of FnBPA in the MSSA strain Newman [28]. However because clinical MRSA isolates that produce an FnBP-dependent biofilm [4], [6] can contain either type II (psm-mec+) or type IV (psm-mec−) SCCmec elements [27], [28], it seems unlikely that carriage of the psm-mec locus alone can explain the expression of Atl/FnBP- or PNAG-dependent biofilm phenotypes by MRSA and MSSA clinical isolates, respectively. Furthermore when considered with earlier reports suggesting a correlation between β-lactam resistance and biofilm [29], [30], [31], [32] our data raise the question as to whether methicillin susceptibility itself influences biofilm and if so, how.
Here we investigated the impact of acquisition of PBP2a-induced homogeneous oxacillin resistance on biofilm and virulence in the laboratory MSSA strain 8325-4. Genetic changes associated with the HoR phenotype in 8325-4 were identified by whole genome sequencing. The biofilm phenotypes of 8325-4 and its HoR derivative were compared and the impact of HoR oxacillin resistance on transcription of the icaADBC and agr loci examined. The impact of loss of SCCmec and methicillin susceptibility on the biofilm phenotype of the MRSA strain BH1CC was also examined. Extracellular protease production by 8325-4 and 8325-4 HoR was measured and the virulence of both strains compared in a mouse model of device-related infection. Our data reveal that expression of homogeneous methicillin resistance in S. aureus influences the biofilm phenotype and attenuates virulence.
To investigate the relationship between susceptibility to β-lactam antibiotics and the biofilm phenotype we used plasmid pSR2 carrying the ccrAB recombinase genes to promote excision of SCCmec in the MRSA clinical isolate BH1CC, which contains a type II SCCmec element and produces an Atl/FnBP-dependent biofilm in growth media supplemented with glucose [6], [7]. Excision of the SCCmec element in BH1CC resulted in a reduction in the oxacillin MIC from >100 µg/ml to <1 µg/ml (data not shown). Biofilm assays revealed a significant reduction in biofilm production by BH1CC ΔSCCmec in BHI glucose compared to BH1CC (Figure 1A). Complementation of the ΔSCCmec mutant with pmecA restored both oxacillin resistance (data not shown) and biofilm production in BHI glucose to near wild type levels (Figure 1A). In contrast pmecAS403A, in which the serine residue in the PBP2a active site was replaced with alanine (as described in the Materials and Methods), failed to restore oxacillin resistance (data not shown) or biofilm production in the BH1CC ΔSCCmec mutant (Figure 1A). RT-PCR analysis demonstrated that mecA mRNA levels were similar in the BH1CC SCCmec mutant carrying pmecA and pmecAS403A (Figure 1B) indicating that functional PBP2a and not mecA mRNA was responsible for the observed phenotypes.
Protease activity in the culture supernatant of BH1CC ΔSCCmec was reduced by approximately 50% compared to BH1CC and was restored to wild type levels by complementation with pmecA (Figure 1C). In addition the addition of the serine protease inhibitor dichloroisocoumarin to the growth media restored biofilm production by the BH1CC ΔSCCmec mutant strain in a dose-dependent manner (Figure 1D). Taken together these data indicate that oxacillin resistance promotes protein adhesin-dependent biofilm production in BH1CC at least in part by repressing extracellular protease production.
To determine if these findings could be extended to other clinical MRSA isolates, we deleted the SCCmec element from six clinical isolates in our collection that formed robust biofilm, were genetically amenable, had different SCCmec types and were from different clonal complexes. Furthermore we complemented all of the SCCmec mutants with the pmecA plasmid. Our data show that the impact of the SCCmec mutation on biofilm varied between the strains (Figure S1). In four strains biofilm production was significantly reduced whereas in the remaining two strains, biofilm was largely unaffected (Figure S1). However complementation with the pmecA plasmid increased biofilm production in all seven SCCmec mutants (Figure S1). The variable impact of the SCCmec deletions in different strains may reflect the complexity and multiple mechanisms of S. aureus biofilm production but nevertheless these data indicate that high level mecA/PBP2a expression always promoted biofilm production providing further evidence that methicillin resistance influences the biofilm phenotype in S. aureus clinical isolates.
To investigate the impact of oxacillin resistance on icaADBC/PNAG-dependent S. aureus biofilm production, we generated a methicillin (oxacillin) resistant derivative of the laboratory strain 8325-4. 8325-4 was chosen because is it amenable to genetic manipulation and exclusively produces icaADBC/PNAG-dependent biofilms [4]. The related laboratory strain SH1000 (an rsbU-repaired derivative of 8325-4 [33]) is capable of PNAG-independent biofilm production [34], [35], while HG003 [36] (an rsbU- and tcaR-repaired derivative of 8325) exhibited a smooth colony morphology on Congo red agar and did not produce detectable levels of PNAG in our experiments (Figure S2).
Transformation of 8325-4 with plasmid pRB474 carrying the mecA gene expressed from its own promoter (pmecA) was accompanied by heterogeneous resistance (HeR) to oxacillin (data not shown). A homogeneously, high-level oxacillin-resistant (HoR) derivative of 8325-4 pmecA was subsequently isolated on BHI media supplemented with 100 µg/ml oxacillin. Western blot analysis performed using commercial monoclonal antibody against PBP2a from Denka-Seiken (Japan) revealed substantially higher PBP2a levels in 8325-4 pmecA HoR than in 8325-4 pmecA HeR (Figure 2A). To investigate the genetic basis for the switch from expression of heterogeneous to homogeneous oxacillin resistance, the genomes of 8325-4 pmecA HeR and 8325-4 pmecA HoR were sequenced and aligned to the S. aureus NCTC8325 (CP000253) genome sequence as described in the Materials and Methods. Briefly one non-synonymous single nucleotide polymorphism (SNP) was identified in the gdpP (SAOUHSC_00015) gene of the HoR strain and confirmed by PCR amplification followed by capillary electrophoresis sequencing. This SNP results in a R602H substitution in GdpP, which has recently been identified as a c-di-AMP phosphodiesterase and implicated in resistance/tolerance to β-lactam antibiotics, biofilm formation and cell wall architecture [37], [38]. Subsequent characterization of nine independently isolated 8325-4 HoR strains identified a G308D substitution in seven strains, and Δ382–504 and Δ80–174 deletions in the other two strains. The GdpP R602 and G308 residues are highly conserved across multiple S. aureus species. These data revealed a strong correlation between high level PBP2a production and homogeneous oxacillin resistance in 8325-4 and suggest that gdpP mutations are also required for maximal PBP2a expression or stability in the 8325-4 pmecA HoR strain.
Comparison of the biofilm phenotypes of 8325-4, 8325-4 pmecA HeR and 8325-4 pmecA HoR under different growth conditions revealed that homogeneous oxacillin resistance was associated with a substantial reduction in biofilm forming capacity in BHI and BHI NaCl (P<0.0001) but that 8325-4 pmecA HoR retained the capacity to form reduced levels of biofilm in BHI glucose (Figure 2B). Our previous studies have shown that biofilm production by 8325-4 is dependent on PNAG under all growth conditions, whereas glucose-induced biofilm by MRSA isolates is mediated by protein adhesins [4], [6], [7]. Thus the loss of NaCl-induced biofilm was suggestive of impaired PNAG production in 8325-4 pmecA HoR. The reduced levels of biofilm produced by 8325-4 pmecA HoR compared to 8325-4 pmecA HeR in BHI glucose can be attributed to the altered biofilm phenotype expressed by this strain, but as described below, both strains formed similar levels of biofilm under flow conditions in BHI glucose. The biofilm phenotypes of 8325-4 and 8325-4 pmecA HeR were similar (Figure 2B) indicating the heterogeneous oxacillin resistance phenotype was not associated with a change in the biofilm phenotype. Curing the pmecA plasmid from 8325-4 pmecA HoR was associated with a return to oxacillin susceptibility and a restoration of the wild type biofilm phenotype (Figure 2B).
Sodium metaperiodate, which is known to break down PNAG-dependent biofilms, degraded 8325-4 and 8325-4 pmecA HeR biofilms grown in BHI, BHI NaCl and BHI glucose, whereas proteinase K had no significant effect (Figures 2C and D). In contrast 8325-4 pmecA HoR biofilms were completely dispersed by proteinase K (P<0.0001) but not by sodium metaperiodate (Figure 2E). Biofilms produced by 8325-4 pmecA HoR (cured) were dispersed by sodium metaperiodate but not proteinase K (Figure 2F). 8325-4 pmecA HoR biofilms were also inhibited by growth in the presence of DNase I whereas 8325-4 and 8325-4 pmecA HeR were unaffected (Figure 2G). Furthermore polyanethole sodium sulfonate (PAS), which blocks autolysis (and consequently eDNA release) [6], [7], completely blocked biofilm production by 8325-4 pmecA HoR but not 8325-4 and 8325-4 pmecA HeR (Figure 2G). Finally, independently isolated 8325-4 pmecA HoR isolates all exhibited the same biofilm phenotype (data not shown). Taken together these data indicate that acquisition of homogeneous oxacillin resistance is apparently accompanied by a switch from PNAG- to protein-mediated biofilm formation and that in contrast to 8325-4, the 8325-4 pmecA HoR biofilm phenotype is dependent on autolytic activity and eDNA release similar to that observed among clinical MRSA isolates expressing an Atl/FnBP-dependent biofilm [6], [7].
A BioFlux system was used to compare biofilm production by 8325-4 pmecA HeR and 8325-4 pmecA HoR grown in BHI NaCl and BHI glucose under flow conditions. These data showed that both strains formed abundant and similar levels of biofilm in BHI glucose (Figure S3). However neither strain was capable of biofilm production in BHI NaCl media (Figure S3) making it impossible to determine the contribution of PNAG to the biofilm phenotypes of both strains using the BioFlux system. The biofilm negative phenotype in BHI NaCl is likely due to the more hydrophobic surface of the BioFlux flow cell compared to the very hydrophilic, tissue-culture treated polystyrene used in our microtitre plate assay. We have previously reported that 8325-4 is incapable of producing a PNAG-type biofilm on hydrophobic polystyrene when grown in BHI NaCl, but can produce a complex biofilm dependent not only on PNAG but also protein adhesin(s) and eDNA on hydrophobic polystyrene in BHI glucose media [7].
To investigate the genetic basis for protein-mediated biofilm in 8325-4 pmecA HoR, the pmecA plasmid was transformed into fnbAB::Tcr, atl::Cmr and srtA::Tcr derivatives of 8325-4 and homogeneous oxacillin resistant variants were isolated as described above. Biofilm assays revealed that HoR derivatives of 8325-4 icaADBC::Tcr, 8325-4 fnbAB::Tcr, 8325-4 atl::Cmr and 8325-4 srtA::Tcr exhibited a similar biofilm phenotype to 8325-4 pmecA HoR (Figure 3A). These data strongly suggest that the biofilm phenotypic switch associated with acquisition of oxacillin resistance is not dependent on PNAG, LPXTG cell wall anchored proteins (including FnBPA, FnBPB and Protein A) or the major autolysin.
This raised the possibility that overexpression of the PBP2a protein itself may promote biofilm development in BHI glucose. Consistent with this, commercial monoclonal antibodies (Calbiochem) to PBP2a reduced biofilm production by 8325-4 pmecA HoR by up to 50% (P<0.05)(Figure 3B). Biofilm production by 8325-4 Δspa pmecA HoR was also inhibited by the PBP2a antibody in a concentration dependent manner (Figure 3B), ruling out any non-specific interference due to antibody binding to Protein A. In contrast the PBP2a monoclonal antibody had no significant effect on PNAG-dependent biofilm production by the plasmid-cured 8325-4 pmecA HoR strain (Figure 3B). These data directly implicate the PBP2a protein in 8325-4 pmecA HoR biofilm development. However it is important to note that the failure of the pmecAS403A plasmid to complement biofilm production in the BH1CC SCCmec strain (Figure 1A) also suggests that PBP2a-induced oxacillin resistance is required for the PBP2a-mediated biofilm phenotype
Comparison of 8325-4, 8325-4 pmecA HeR and 8325-4 pmecA HoR on Congo red agar revealed that the acquisition of homogeneous resistance was associated with a switch from a crusty to a smooth colony morphology, which is indicative of reduced PNAG production (data not shown). Accordingly immunoassays demonstrated that, unlike 8325-4 and 8325-4 pmecA HeR, PNAG was not produced by 8325-4 pmecA HoR (Figure 4A). Interestingly PNAG production was restored in the cured 8325-4 HoR strain (Figure 4A). Using real time RT-PCR, a >300-fold repression of icaADBC transcription was measured in 8325-4 pmecA HoR compared to 8325-4 (Figure 4B). Furthermore wild type levels of icaADBC transcription were restored in the plasmid-cured 8325-4 pmecA HoR strain (Figure 4B). Transcriptional activity of icaR was similar in both strains (Figure 4C). In addition genome sequencing of 8325-4 pmecA HoR confirmed the absence of any mutations in any known ica operon transcriptional regulator. Thus homogeneous oxacillin resistance alters the biofilm phenotype by repressing icaADBC transcription and PNAG production.
Assays of protease activity in culture supernatants revealed an approximately 2-fold reduction in protease levels in 8325-4 pmecA HoR compared to 8325-4, 8325-4 pmecA HeR and the cured 8325-4 HoR strain (Figure 5A). Because extracellular protease activity is subject to regulation by the accessory gene regulator (Agr) system, we used real time RT-PCR to compare RNAIII transcript levels in these strains. These data revealed that RNAIII expression was significantly repressed in 8325-4 pmecA HoR grown to both the exponential and stationary phases of growth (Figure 5B). Genome sequence analysis confirmed the absence of any mutations in the agr locus of 8325-4 pmecA HoR. Thus the homogeneous oxacillin resistance phenotype in 8325-4 is associated with repression of the Agr system and extracellular protease production, both of which are consistent with PBP2a-mediated biofilm development by 8325-4 pmecA HoR. Furthermore these data correlate with our earlier observation that protease activity was increased in BH1CC ΔSCCmec culture supernatants (Figure 1B).
Because 8325-4 and 8325-4 pmecA HoR express PNAG-dependent and PNAG-independent biofilm phenotypes, respectively, their virulence was compared using an established mouse model of device-related infection [39], [40]. Briefly two 1-cm segments of 14-gauge polyethylene intravenous catheter were implanted subcutaneously per mouse. In total four groups of eight mice were inoculated with bacterial cell suspensions of 1×107 or 1×108 of either 8325-4 or 8325-4 pmecA HoR, injected directly into the catheter. These experiments revealed a significantly higher mortality rate among mice inoculated with 8325-4 compared to 8325-4 pmecA HoR (Figure 6). Seven of the eight mice inoculated with 1×108 8325-4 died or were euthanized by Day 4. Similarly only one of seven mice (one mouse failed to recover from the anesthetic) inoculated with 1×107 8325-4 survived beyond Day 4 (Figure 6). In contrast all eight animals inoculated with 1×107 8325-4 pmecA HoR survived to Day 7 and six of the eight animals inoculated with 1×108 8325-4 pmecA HoR survived to Day 7 (Figure 6).
The high mortality rate associated with infection of 8325-4 made it impossible for us to compare the invasiveness and dissemination of 8325-4 and 8325-4 pmecA HoR. Thus, the above experiment was repeated over an 18-hour time period with an inoculum of 1×107 bacteria after which all animals were sacrificed. The implanted catheter sections were aseptically removed and bacteria associated with the implanted biomaterial were quantitatively cultured on tryptone soya agar (TSA). In addition, peri-catheter tissue, liver, kidneys and spleen were dissected, weighed, homogenized and quantitatively cultured on TSA, as were bacteria present in blood. These data revealed that the numbers of adherent bacteria on implanted catheters were similar for 8325-4 and 8325-4 pmecA HoR (Figure 7A) indicating that the altered biofilm phenotype expressed by 8325-4 pmecA HoR does not diminish its ability to colonize implanted devices. There were significantly more 8325-4 pmecA HoR bacteria in the peri-catheter tissue than 8325-4 bacteria (Figure 7B). However significantly fewer 8325-4 pmecA HoR bacteria were recovered from the liver, blood, spleen and kidneys than 8325-4 bacteria (Figure 7 C–F). Thus, these findings are consistent with the mortality data and indicate that 8325-4 is significantly more invasive than 8325-4 pmecA HoR, which accumulated in the peri-catheter tissue and was less capable of disseminating to other organs.
The immune response of the mice to 8325-4 and 8325-4 pmecA HoR was measured by assaying levels of TNF-α and IL-6 in the peri-catheter tissue. Levels of both pro-inflammatory cytokines were significantly increased in mice inoculated with 8325-4 compared to 8325-4 pmecA HoR (Figure 8), despite the fact that there were more 8325-4 pmecA HoR cells recovered from the tissue (Figure 7B). Taken together these data reveal that homogeneous, high-level resistance to oxacillin significantly attenuated the virulence of 8325-4.
To investigate the impact of homogenous methicillin resistance on biofilm production and Agr activity in clinical S. aureus isolates, we first introduced pmecA into the MSSA strains MSSA476 and 15981 and isolated HoR derivatives on media containing oxacillin 100 µg/ml. As observed in 8325-4, acquisition of the HoR phenotype in MSSA476 and 15981 was associated with repression of polysaccharide-type biofilm production and expression of a protein adhesin-mediated biofilm phenotype (Figure S4). Furthermore the MSSA476 and 15981 HoR strains exhibited reduced δ-hemolytic activity on sheep blood BHI agar (Figure S4). Given that the δ-hemolysin is encoded by the hld gene within the agr locus RNAIII transcript [41], these data indicate that the HoR phenotype in MSSA476 and 15981 is accompanied by repression of Agr activity.
The impact of the HoR phenotype on biofilm production by the CA-MRSA USA300 strain LAC was also examined. In our experiments this strain formed a protein-adhesin type biofilm in BHI glucose media (Figure S5A). A LAC HoR derivative isolated on media containing oxacillin 100 µg/ml produced significantly more biofilm (Figure S5A) and exhibited reduced δ-hemolytic activity on sheep blood BHI agar (Figure S5B). Finally we examined the impact of the HoR phenotype on three CC5 strains with different SCCmec elements exhibiting a heterogeneous pattern of oxacillin resistance, namely DAR173, DAR26 and DAR9. These strains produced a polysaccharide-type biofilm, whereas their HoR derivates isolated on media containing oxacillin 100 µg/ml produced protein adhesin-type biofilms and exhibited reduced δ-hemolytic activity (Figure S5C–K).
In the laboratory strain 8325-4, non-synonymous SNPs in gdpP were associated with the HoR phenotype. Similarly, DNA sequencing of the gdpP gene from the USA300 LAC HoR derivative identified an R450STOP mutation. GdpP amino acid substitutions were also identified in HoR derivatives of the clinical isolates 15981 (P392S, D105N), MSSA476 (V52I, D105N, P392S) and DAR26 (N105D, S392P). However these latter substitutions occur in GdpP residues that are not conserved in multiple S. aureus species and so their significance is unclear. No gdpP mutations were identified in HoR derivatives of the clinical isolates DAR176 and DAR9 suggesting that the HoR phenotypes of these strains may be independent of c-di-AMP or that these strains contain unidentified mutations in c-di-AMP target genes.
Antibiotic resistance, enzyme and toxin production, biofilm forming capacity and immune evasion capability have combined to accelerate the emergence of S. aureus as a globally important human pathogen. Our recent research has focused on the relationship between two of these virulence determinants, antibiotic resistance and biofilm-forming capacity, in clinical S. aureus isolates. Our data show that MSSA clinical isolates are more likely to produce a PNAG-dependent biofilm than MRSA isolates which produce an Atl/FnBP-dependent biofilm [3], [4], [6], [7], suggesting that methicillin susceptibility influences biofilm expression. MRSA strains contain the Staphylococcus cassette chromosome mec (SCCmec) including the mecA gene, which encodes the PBP2a protein that confers resistance to β-lactam antibiotics. To investigate how methicillin resistance influences the biofilm phenotype we introduced a plasmid expressing the mecA gene from its own promoter (pmecA) into the PNAG-producing laboratory strain 8325-4. 8325-4 pmecA exhibited heterogeneous resistance to oxacillin (designated 8325-4 pmecA HeR) from which an homogeneous, high-level oxacillin resistant derivative was isolated, designated 8325-4 pmecA HoR. Western blots show that expression of homogeneous resistance was associated with substantially increased expression of PBP2a. Genome re-sequencing identified a number of amino acid substitutions and small deletions in the gdpP gene of independent 8325-4 HoR strains and in HoR derivatives of the clinical isolates USA300 strain LAC, MSSA476, 15981 and DAR26. GdpP has recently been identified as a phosphodiesterase controlling intracellular levels of the secondary messenger c-di-AMP, which in turn influences phenotypes such as cell wall architecture, biofilm formation and, most relevant to this study, resistance/tolerance to β-lactam antibiotics [37], [38]. Thus c-di-AMP may be involved in the multiple phenotypic changes associated with homogeneous oxacillin resistance. However, given that the GdpP amino acid sequence is identical in 8325-4 (MSSA), USA300 strain LAC (HeR MRSA) and BH1CC (HoR MRSA), mutations in gdpP and/or altered c-di-AMP signaling alone may not be sufficient to explain the HeR to HoR transition. Furthermore no gdpP SNPs were identified in HoR derivatives of two clinical isolates (DAR9 and DAR176) suggesting that either these strains contain mutations in c-di-AMP target genes or that c-di-AMP-independent mechanisms may also be involved in the HoR phenotype. In this context our recent finding that loss of methicillin resistance in the HoR MRSA strain BH1CC reduced biofilm forming capacity and increased virulence in a mouse sepsis model [42], also indicates that altered c-di-AMP levels alone are unlikely to account for the pleiotropic effects of the HoR phenotype. Clearly homogenous methicillin resistance remains a complex phenotype and future research to identify c-di-AMP targets is necessary to gain mechanistic insights into how this cyclic dinucleotide contributes to the co-regulation of antibiotic resistance, biofilm and virulence.
The biofilm phenotype of 8325-4 pmecA HoR was dramatically altered and characterized by the loss of PNAG production and expression of a protein-adhesin mediated biofilm phenotype. Using a series of surface protein mutants, we were able to show that the PBP2a-induced biofilm phenotype was independent of PNAG, Atl, and any of the LPXTG-anchored surface proteins (including the FnBPs and protein A). Furthermore, a monoclonal antibody against PBP2a reduced 8325-4 pmecA HoR biofilm production by approximately 50% implicating PBP2a itself in intercellular accumulation. The inability of the PBP2a antibody to fully inhibit 8325-4 pmecA HoR biofilm may be explained, at least in part, by our observation that this biofilm phenotype was also dependent on autolytic activity and extracellular DNA. We have previously implicated autolytic activity and eDNA in the early stages of PNAG-independent biofilm phenotypes expressed by S. aureus clinical isolates [6], [7]. In contrast biofilm expression by 8325-4 pmecA HeR was similar to wild type, indicating that high level PBP2a expression is required to change the biofilm phenotype. In the MRSA strain BH1CC, excision of SCCmec and induced methicillin susceptibility also resulted in an impaired biofilm phenotype that could be complemented by the wild type mecA gene but not a mecA allele expressing a PBP2a mutant with an S403A substitution in the active site of the enzyme. Taken together, these data suggest that both high level PBP2a expression and methicillin resistance are required for the PBP2a-induced biofilm phenotype. In addition, because Atl/FnBP-dependent biofilm development by BH1CC was unaffected by PBP2a antibody (data not shown), our data also reveal redundancy among S. aureus surface proteins that can promote biofilm development.
How PBP2a promotes biofilm is uncertain but possibly the altered cell wall architecture in methicillin resistant strains expressing high levels of PBP2a may facilitate PBP2a-promoted cell-cell interactions that are not possible between MSSA cells. What is clear, however, is that high level PBP2a expression resulted in significant repression of icaADBC transcription and PNAG production in 8325-4 pmecA HoR (Figure 9). Comparison of acetic acid levels in 8325-4 and 8325-4 pmecA HoR culture supernatants revealed no significant difference (data not shown) suggesting that altered TCA cycle activity, which is known to regulate PIA/PNAG production [43], [44], may not be involved in this phenotype. PBP2a overexpression also resulted in significant repression of the agr locus. These data are supported by our recent findings that excision of the SCCmec element from the MRSA strain BH1CC and loss of oxacillin resistance had the opposite effect and was associated with increased agr transcription [42]. Furthermore our previous data showed that MRSA cells are unable to detect the Agr-encoded auto-inducing peptide (AIP), thus preventing normal activation of the Agr system and concomitant virulence gene regulation [42]. However this does not explain why the icaADBC locus is repressed in strains expressing high levels of PBP2a. Given that both the agr and ica loci are repressed by PBP2a expression, it seems possible that a transcriptional regulator(s) involved in regulating both operons may also play a role in this phenotype. For instance preliminary data indicate that the activity of the virulence and carbon metabolism regulator CcpA, which is known to increase oxacillin resistance levels [45], [46], was activated approx 4-fold in 8325-4 pmecA HoR (unpublished findings). On the other hand CcpA positively regulates both icaADBC and agr expression [45], [46], suggesting that the impact of PBP2a overexpression on biofilm and virulence is complex and experiments are underway to investigate the potential involvement of other transcription factors.
Repression of the Agr locus in response to homogenous methicillin resistance was accompanied by down-regulation of extracellular protease production [35], [47] (Figure 9). Protease activity has previously been shown to alter murein hydrolase activity and in turn biofilm [35], which appears to correlate with our finding implicating autolytic activity in the 8325-4 pmecA HoR biofilm phenotype. Thus repression of extracellular proteases combined with high level PBP2a expression and methicillin resistance appears to be required for PBP2a-mediated biofilm development in 8325-4. Consistent with this, activation of agr following excision of SCCmec from BH1CC increased production of extracellular toxins [42]. Here we have shown that loss of SCCmec also increased extracellular protease production in BH1CC, which in turn correlated with diminished Atl/FnBP-promoted biofilm forming capacity. It is also interesting to note that extracellular protease activity was similar in both 8325-4 pmecA HoR and BH1CC SCCmec pmecA, which both express a mecA-promoted biofilm phenotype and that as mentioned above BH1CC does not display the characteristic stationary phase induction of RNAIII typical of agr+ strains [42]. Taken together these data indicate that the capacity of PBP2a to promote biofilm is dependent, at least in part, on the background levels of extracellular protease activity in individual strains.
The altered biofilm phenotype and repression of agr in 8325-4 pmecA HoR was accompanied by a significant reduction in virulence in a mouse model of device-related infection. This finding correlated with our previous data that the virulence of BH1CC was increased following excision of SCCmec in a mouse sepsis model [42]. Thus using two different strains and two different mouse infection models, our data indicate that expression of methicillin resistance reduces virulence potential in S. aureus. Furthermore although the 8325-4 pmecA HoR strain was equally capable of colonizing implanted biomaterials, it accumulated significantly more in peri-catheter tissue but was significantly less capable of causing invasive disease, resulting in a significantly reduced mortality rate than the wild type 8325-4. An intriguing backdrop to our findings are the recent reports that psm-mec locus, which is located adjacent to the mecA locus on type II SCCmec elements, also represses virulence and promotes the S. aureus biofilm phenotype [26], [27], [28]. Thus both mecA-encoded PBP2a and the psm-mec encoded phenol-soluble modulin-mec (PSM-mec) can independently repress virulence and promote biofilm in MRSA. The psm-mec-encoded RNA is able to repress expression of the chromosomally encoded PSMα, while both the psm-mec RNA and the PSM-mec protein repress colony spreading ability and promote biofilm [26], [27], [28].
Given that healthcare-associated MRSA strains are typically responsible for infections in immunocompromised patients in which numerous implanted medical devices are used for organ and life support and in which the use of antimicrobial drugs is high, these findings reveal a sophisticated level of adaptation by MRSA to the hospital environment. The reduced metabolic costs associated with PBP2a-induced repression of Agr and exoprotein production coupled with PBP2a-mediated biofilm production, may confer advantages on MRSA strains. Agr defective strains are known to be less virulent [48], [49] and a number of studies [50], [51] have suggested that, in immunocompromised patients, Agr defective strains may have advantages including reduced metabolic costs of exoprotein production, increased biofilm production [52] and increased FnBP expression [53] with resulting effects on host cell invasion/immune evasion [54]. Indeed biofilm-forming S. epidermidis strains, which express fewer exoproteins and exotoxins (and are consequently less virulent), are also significant pathogens in this patient group. Thus sacrificing virulence for antibiotic resistance is not necessarily a disadvantage for MRSA and may in fact benefit the pathogen in this clinical setting. Furthermore MRSA strains have retained the capacity for biofilm production, albeit using surface protein adhesins rather than PNAG, which is important given that the majority of bloodstream infections in hospital patients are associated with implanted medical devices.
This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All mouse protocols were reviewed and approved by the Institutional Animal Care and Use Committee at the University of Nebraska Medical Centre. All surgery was performed under anesthesia and all efforts were made to minimize suffering.
The S. aureus strains and the plasmids used in the manipulation of these strains are described in Table 1. Escherichia coli strains were grown at 37°C on LB medium supplemented, when required, with ampicillin (100 µg/ml) or kanamycin (50 µg/ml). S. aureus strains were grown at 30°C or 37°C on Brain-Heart Infusion (BHI) (Oxoid) medium supplemented when required with chloramphenicol (10 µg/ml), tetracycline (10 µg/ml), kanamycin (10 µg/ml) and oxacillin (0–100 µg/ml). BHI broth was supplemented where indicated with 1% glucose or 4% NaCl.
δ-haemolytic activity was visualized on sheep blood BHI agar as described by Traber et al. [41]. Briefly clinical S. aureus isolates were inoculated onto a lawn of S. aureus RN4220, which produces only β-haemolysin. The RN4220-expressed β-haemolysin enhances lysis of red blood cells by δ-haemolysin, while inhibiting α-haemolysin, thus enabling detection of δ-haemolytic activity by clinical S. aureus isolates.
Genomic and plasmid DNA was prepared using Wizard Genomic DNA and plasmid purification kits (Promega). Prior to DNA extraction cells were pre-treated with 5–10 µl of a 1 mg/ml concentration of lysostaphin (Ambi products, New York) in 100 µl 50 mM EDTA to facilitate lysis. Restriction and DNA modifying enzymes (Roche, UK and New England Biolabs, MA) were used according to the manufacturer's instructions. The serine residue at amino acid 403 in the active site of PBP2a was mutated to alanine using Phusion polymerase (NEB) and the primers mecAS403A_For and mecAS403A_Rev (Table 2). Plasmid pmecA was used as the template. Successful mutagenesis of the DNA sequence encoding the S403 residue resulted in the introduction of a new DraIII restriction enzyme site and candidate plasmids harbouring the mutation were digested with this enzyme before being confirmed by sequencing. The mutated plasmid designated pmecAS403A.
Transformations of plasmid DNA into E. coli and Staphylococcus strains were performed as described previously [55]. MWG Biotech, Germany or Sigma-Aldrich, Ireland supplied oligonucleotide primers used for PCR and RT-PCR (Table 2).
Semi-quantitative measurements of biofilm formation were determined using Nunclon tissue culture treated (Δ surface) 96-well polystyrene plates (Nunc, Denmark), based on the methods of Christensen et al. [56] and Ziebuhr et al. [57] with the following modification. Bacteria were grown in individual wells of 96-well plates at 37°C in BHI medium or BHI supplemented with 4% NaCl or 1% glucose. The serine protease inhibitor dichloroisocoumarin was added to BHI glucose media at concentrations of 0.004–0.5 mM, where indicated. After 24 h of growth, the plates were washed vigorously three times with distilled H2O to remove unattached bacteria and dried for 1 hour at 60°C, as recommended by Gelosia et al. [58] as described previously [4], [55]. The absorbance of the adhered, stained biofilms was measured at A492 using a microtitre plate reader. Each strain was tested at least three times and average results are presented. Mouse monoclonal anti-PBP2a antibody (CalBioreagents, CA), DNase I (Sigma) or polyanethole sodium sulfonate (PAS) (Sigma) were added to biofilm cultures at the start of the assay as indicated. Biofilm stability against proteinase K (Sigma), sodium-meta-periodate (Sigma) was tested as described previously [10], [59], [60].
Bacteria were grown on Congo red agar (CRA) plates, which are composed of BHI agar supplemented with 5% sucrose (Sigma) and 0.8 mg of Congo red/ml (Sigma) to distinguish between PNAG-producing (black, dry colony morphology) and non-PNAG-producing (red, smooth colony morphology) phenotypes as described previously [4], [55].
Protease activity in culture supernatants was measured using a protease assay kit (Calbiochem, Germany) according to the manufacturer's instructions.
To analyze biofilm formation under flow conditions, we utilized the BioFlux 1000 microfluidic system (Fluxion Biosciences Inc., South San Francisco, CA) which allows automated image acquisition within specialized multi-well plates. To grow biofilms, the microfluidic channels were primed with 50% BHI supplemented with 4% NaCl or 1% glucose at 10.0 dyn/cm2. Channels were seeded at 2 dyn/cm2 with 107 CFU from overnight cultures of 8325-4 pmecA HeR and 8325-4 pmecA HoR. The plate was then incubated at 37°C for 1 hour to allow cells to adhere. Excess inoculums were removed and 2 ml of 50% BHI supplemented with 4% NaCl or 1% glucose was added to the input wells. Biofilms were grown at 37°C with a flow of fresh media at a constant shear of 0.7 dyn/cm2. Images were taken every 5 minutes for 18 hours at 200× magnification under brightfield.
The pmecA plasmid [61] was introduced by electroporation into 8325-4. The 8325-4 pmecA strain exhibited heterogeneous oxacillin resistance (HeR) characterized by a minority of cells with an MIC>100 µg/ml and the majority of cells with an MIC<1 µg/ml. To obtain high-level oxacillin resistant derivatives, dilutions of 8325-4 pmecA cultures were grown on BHI agar containing increasing concentrations of oxacillin (0–100 µg/ml) and the number of colony forming units counted after 24 h growth at 37°C. High-level, homogeneous oxacillin resistant derivatives (HoR) were subcultured from oxacillin 100 µg/ml plates. Curing the pmecA plasmid from 8325-4 pmecA HoR was achieved by 48 h growth in antibiotic free media at 45°C and isolation of chloramphenicol susceptible colonies followed by plasmid profile analysis.
The genomes of the 8325-4 pmecA HeR and 8325-4 pmecA HoR were sequenced using an Illumina Genome Analyzer as per manufacturer's instructions (Illumina, San Diego, CA, USA) generating 5.7 million and 7.5 million reads for 8325-4 pmecA HoR and 8325-4 pmecA HeR, respectively. The genome sequences were then mapped back to the S. aureus NCTC8325 (CP000253) genome sequence using the short read aligner Bowtie (http://bowtie-bio.sourceforge.net/index.shtml & PMID:19261174) allowing up to 2 mismatches per uniquely mapped read. Single nucleotide polymorphisms (SNPs) were identified using samtools programs [PMID:19505943] in regions with a read depth of greater than 4. Identified SNPs were confirmed by PCR amplification followed by capillary electrophoresis sequencing of all candidate polymorphic regions from 8325-4 pmecA HeR and 8325-4 pmecA HoR.
Cultures were grown in BHI glucose after which cells were collected and immediately stored at −20°C in RNAlater (Ambion) to ensure maintenance of RNA integrity prior to purification. Cells were pelleted and resuspended in 100 µl of 200 mM Tris HCl (pH 7.8) supplemented with 800 µg/ml lysostaphin for 2 min to weaken the cell wall prior to lysis. Total RNA was subsequently isolated using the Qiagen RNeasy mini kit (Qiagen) according to the manufacturer's instructions. Residual DNA present in the RNA preparations was removed using Ambion recombinant Turbo DNase. Purified RNA was eluted and stored in RNAsecure resuspension solution (Ambion) and the integrity of the rRNA confirmed by agarose gel electrophoresis. RNA concentration was determined using a Nanodrop spectrophotometer.
Real-time reverse transcription-PCR (RT-PCR) was performed on a LightCycler instrument using the RNA amplification kit Sybr Green I (Roche Biochemicals, Switzerland) following the manufacturer's recommended protocol. RT was performed at 61°C for 30 min, followed by a denaturation step at 95°C for 30 sec and 35 amplification cycles of 95°C for 20 sec, 50°C for 20 sec and 72°C for 20 sec. Melting curve analysis was performed at 45°C to 95°C (temperature transition, 0.1°C per sec) with stepwise fluorescence detection. For LightCycler RT-PCR, RelQuant software (Roche Biochemicals) was used to measure relative expression of target genes. 16S rRNA was used as an internal standard in real-time RT-PCR experiments. Each experiment was performed at least three times and average data with standard deviations are presented.
PNAG assays were performed as described elsewhere [62]. Briefly 5 ml overnight cultures (approximately 5×109 bacteria) were collected by centrifugation, resuspended in 500 µl of 0.5 M EDTA and boiled for 5 min. The cell debris was again centrifuged and the supernatant treated with 20 µg proteinase K at 65°C for 1 h. The proteinase K was inactivated by boiling for 5 min and the samples diluted as appropriate before application onto nitrocellulose (pre-wetted in TBS) using a vacuum blotter. The blots were dried, re-wet in TBS, and blocked for 1 h in 5% skimmed milk. The primary antibody (1∶5,000 dilution of rabbit anti-PNAG (a kind gift from Tomas Maira Litran) in TBST+1% skimmed milk) was then applied to the membrane for 1 h. Horseradish peroxidase linked anti-rabbit IgG secondary antibody (1∶5,000 dilution in TBST+1% skimmed milk) was then incubated with the membrane for 1 h. A chemiluminescence kit (Amersham) was used to generate light via the HRP-catalyzed breakdown of luminal and detected using a BioRad Fluor-S Max CCD camera system.
Western blot analysis of PBP2a was performed as described elsewhere [63] using overnight bacterial cultures (20 ml) grown at 37°C in BHI glucose. The A600 of the overnight cultures was measured to ensure that similar cell densities were present prior to protein extraction. The cells were harvested by centrifugation, washed with 50 mM Tris (Sigma), 150 mM NaCl and 5 mM MgCl (pH 7.5) and resuspended in the same buffer. Lysostaphin (200 µg/µl), RNase (10 µg/µl) and DNase (20 µg/µl) were added to the cell suspension and incubated at 37°C for 30 min. The cells were disrupted by sonication on ice and the insoluble cell fraction was pelleted by ultracentrifugation at 80000× g for 40 min before being resuspended in 50 mM sodium phosphate (pH 7.0) containing 6 M urea.
The membrane proteins (25 µg) were separated using 10% sodium dodecyl sulfate-polyacrylamide gel electrophoresis and electrophoretically transferred onto Immobilon-P membranes (Millipore) at 12 V for 30 mins. The primary antibody (1∶5000 dilution of a mouse anti-PBP2a monoclonal antibody (Denka Seiken) in TBS+0.1% Tween) was applied to the membranes overnight. A horseradish peroxidase conjugated anti-mouse IgG secondary antibody (1∶1000 in TBST+1% skim milk) was incubated with the membranes for 1 hour. A chemiluminescence kit (Amersham) was used as described above.
A mouse model of device-related infection [39] was used to compare virulence of S. aureus 8325-4 and 8325-4 pmecA HoR. Briefly, the flanks of anesthetized 6-week old male C57BL/6 mice were shaved, and the skin cleansed with povidone-iodine. Using aseptic technique, a 1-cm segment of 14-gauge polyethylene intravenous catheter was implanted into the subcutaneous space (two per mouse and the incision closed with Vetbond (3M, Minneapolis, MN). Next, 1×107 or 1×108 S. aureus were injected into the catheter lumen. Eight mice were used to test each inoculum and strain. Survival over 7 days was measured. Animals were euthanized before the end of the experiment using the following indices for evaluating whether a moribund state had been achieved: extreme lethargy, failure to demonstrate typical avoidance behaviour when handled, ulceration of the infection site through the skin, excessive loss of body weight (i.e. >20%), and/or labored breathing.
The high level of mortality associated with infection of 8325-4 prompted us to repeat the above experiment over 18 hours using an inoculum of 1×107 bacteria before sacrificing all animals and enumerating the numbers of bacteria associated with the catheter, surrounding tissue, blood, liver, spleen and kidneys. To do this, the catheters were aseptically removed, placed in sterile microcentrifuge tubes with 1 ml of phosphate buffered saline (PBS), vortexed for one minute, and quantitatively cultured on tryptone soya agar (TSA). In addition, peri-catheter tissue, liver, kidneys and spleen were dissected, weighed, homogenized and quantitatively cultured on TSA. Finally, bacteria present in blood were also quantitatively cultured on TSA.
TNF-α and IL-6 levels were measured in the peri-catheter tissue using TNF-α (OptiEIA, BD Bioscience) or IL-6 (Duoset; R&D Systems) ELISA kits according to the manufacturer's instructions. Results were normalized to the total amount of tissue recovered.
Two-tailed, two-sample equal variance Student's t-tests (Microsoft Excel 2007) were used to determine statistically significant differences in biofilm forming capacity and relative gene expression.
For the animal experiments descriptive statistics (including mean, standard deviation, median minimum and maximum) values were calculated for each strain at each location i.e. catheter, peri-catheter tissue, blood, liver, spleen and kidneys. Log10 transformation of CFU data from catheter and tissue was used to ensure normal distribution. A two-sample t-test was conducted to compare the log-transformed CFU values for 8325-4 and 8325-4 pmecA HoR from either catheter or peri-catheter tissue. Nonparametric Wilcoxon rank sum tests were used to compare 8325-4 and 8325-4 pmecA HoR CFU data from blood, liver, spleen and kidneys.
|
10.1371/journal.pcbi.1000242 | Dependence of Bacterial Chemotaxis on Gradient Shape and Adaptation Rate | Simulation of cellular behavior on multiple scales requires models that are sufficiently detailed to capture central intracellular processes but at the same time enable the simulation of entire cell populations in a computationally cheap way. In this paper we present RapidCell, a hybrid model of chemotactic Escherichia coli that combines the Monod-Wyman-Changeux signal processing by mixed chemoreceptor clusters, the adaptation dynamics described by ordinary differential equations, and a detailed model of cell tumbling. Our model dramatically reduces computational costs and allows the highly efficient simulation of E. coli chemotaxis. We use the model to investigate chemotaxis in different gradients, and suggest a new, constant-activity type of gradient to systematically study chemotactic behavior of virtual bacteria. Using the unique properties of this gradient, we show that optimal chemotaxis is observed in a narrow range of CheA kinase activity, where concentration of the response regulator CheY-P falls into the operating range of flagellar motors. Our simulations also confirm that the CheB phosphorylation feedback improves chemotactic efficiency by shifting the average CheY-P concentration to fit the motor operating range. Our results suggest that in liquid media the variability in adaptation times among cells may be evolutionary favorable to ensure coexistence of subpopulations that will be optimally tactic in different gradients. However, in a porous medium (agar) such variability appears to be less important, because agar structure poses mainly negative selection against subpopulations with low levels of adaptation enzymes. RapidCell is available from the authors upon request.
| Chemotaxis plays an important role in bacterial lifestyle, providing bacteria with the ability to actively search for an optimal growth environment. The chemotaxis system is likely to be highly optimized, because on the evolutionary time scale even a modest enhancement of its efficiency can give cells a large competitive advantage. In this study, we use up-to-date experimental and modeling information to construct a new computational model of chemotactic E. coli and implement it in a computationally efficient way to simulate large bacterial populations. Our simulations are performed in a new type of attractant gradient that ensures a constant level of chemotactic excitation at any position. We show that optimal chemotactic movement in a gradient results from a fine balance between excitation and adaptation. As a consequence, steeper gradients require higher optimal rates of adaptation. Simulations demonstrate that the observed intercellular variability of adaptation times, which is caused by gene expression noise, may play a positive role for the bacterial population as a whole, by allowing its subpopulations to be optimally tactic in gradients of different strengths. We further show that optimal chemotactic properties in a porous medium (agar) are different from those in a liquid.
| One of the central questions of modern systems biology is the influence of microscopic parameters of a single cell on the behavior of a cell population, a common problem in multi-scale modeling. In terms of bacterial chemotaxis, this issue can be formulated as the influence of signaling network parameters on the spatiotemporal dynamics of a population in various gradients of chemoattractants. The problem of efficient multi-scale simulation imposes strict requirements on the model: it should be maximally detailed to grasp the main features of the signaling network yet computationally cheap to simulate large numbers of bacteria.
Chemotaxis plays an important role in microbial population dynamics. Chemotactic bacteria in a nonmixed environment—that is in presence of nutrient gradients—have significant growth advantages, as shown experimentally for different bacterial species [1]–[4]. Modeling of microbial population dynamics indicates that motility and chemotactic ability can be as important for evolutionary competition as cell growth rate [5],[6].
Escherichia coli is an ideal organism for chemotaxis modeling, because of the rich experimental information collected over years of extensive research. In common with many other bacteria, E. coli can migrate towards high concentrations of attractants and away from repellents. In the adapted state, cells perform a random walk, which becomes biased in the presence of a spatial gradient. This swimming bias is based on temporal comparisons of attractant concentrations during cell runs. If the direction of a run is favorable, i.e. up the attractant gradient or down the repellent gradient, the run become longer. Between runs, the cell tumbles and reorients for the next run [7].
Chemotaxis in E. coli is mediated by an atypical two-component signal transduction pathway (for recent reviews see [8],[9]). Ligand molecules bind to clusters of transmembrane receptors, which are in complex with the histidine kinase CheA and the adaptor CheW. Each receptor can be either active or inactive, depending on ligand binding and the methylation level. The active receptor activates CheA, eliciting downstream phosphorylation of the response regulator CheY. Phosphorylated CheY (CheY-P) is dephosphorylated by CheZ. Receptors can be methylated by the methyltransferase CheR and demethylated by the methylesterase CheB, and methylation regulates the receptor activity. The methylation of receptors provides a sort of chemical ‘memory’, which allows the cell to compare the current ligand concentration with the past. Phosphorylation of CheB by CheA provides a negative feedback to the system, although it appears nonessential for exact adaptation [10],[11]. Phosphorylated molecules of CheY-P freely diffuse through the cytoplasm and bind to the flagellar motor protein FliM, causing motors to switch from CCW to CW rotation. Switching of the motors to the CW state results in a tumble and reorientation, whereas the CCW rotation corresponds to straight runs.
A number of mathematical models of chemotaxis have been proposed [10], [12]–[18], including two recent programs that simulate cell motion along with the intracellular pathway dynamics: AgentCell [19], which is based on the StochSim pathway simulator [20]–[22], and E. solo [23], which is based on the BCT simulator [24]–[26]. The current version of AgentCell (2.0) simulates the whole pathway stochastically, making it thus computationally very expensive. The E. solo program simulates the pathway using about 90 ordinary differential equations (ODEs). However, simulation of large bacterial populations on long time scales requires computationally cheaper models.
It was recently shown using fluorescence resonance energy transfer (FRET) that the amplitude of the initial CheY-P response can be described by a Hill function of a relative change in receptor occupancy during stepwise ligand stimulation [27]. Recent modeling efforts [12],[28],[29] show that a mixed-cluster Monod-Wyman-Changeux (MWC) model of strongly coupled receptors is consistent with the FRET data, and can account for the observed sensitivity and precise adaptation over a wide range of ligand concentrations. The amplitude of pathway excitation can therefore be determined using several algebraic equations describing the free energy of the cluster.
In our model (Figure 1A), we employed the MWC model for a mixed receptor cluster [12] with a mean-field approximation for adaptation kinetics [30]. Due to its hybrid approach, the model allowed us to reduce the computational costs dramatically, while keeping the main quantitative characteristics of the cell response (methylation level, relative CheY-P concentration, motor bias) consistent with experimental data. To couple the bias of individual motors to the probability of tumbling, we applied a voting model for several independent motors, based on detailed experimental investigation of tumbling mechanics [31].
These components were combined into a new simulator for E. coli chemotaxis—RapidCell, which uses a hybrid pathway simulation instead of a fully stochastic or ODE approach, and is therefore computationally cheap. This allows the simulation of populations of 104–105 cells on a time scale of hours using a desktop computer.
To study the dependence of chemotaxis on gradient strength in a systematic way, we propose a new—constant-activity—gradient which ensures a constant average CheY-P level and cellular drift velocity along the gradient, in contrast to commonly used Gaussian and linear gradients. We show that the MWC model gives an approximately constant response over a wide range of ligand concentrations. Though purely theoretical, such a gradient serves as a perfect in silico assay to study the chemotactic properties of cells.
The chemotaxis pathway is robust to changes in network parameters and intracellular protein concentrations [10],[15],[32]. This enables efficient chemotaxis with varying levels of intracellular components and under perturbations from extracellular environment. However, adaptation time is not robust [10],[11],[33],[34] and varies even among genetically identical cells in a population because of stochastic variations in gene expression and low copy numbers of the adaptation enzymes.
Our simulations predict that in liquid media for any given gradient steepness, there is an optimal adaptation rate that provides the highest cellular drift velocity. We suggest a simple mechanism for this phenomenon: the optimal rate of adaptation is observed in a narrow range of kinase activity, where the average CheY-P level fits the operating range of the flagellar motor. In this range, the relation between CheY-P and motor bias is approximately linear, and cells perform chemotaxis with the highest efficiency.
The situation is different for cells swimming in agar. Here, the optimal range of motor bias appears to be very narrow and just slightly higher than in the non-stimulated state. Due to the porous structure of agar, cells with a higher CCW motor bias stay trapped for a longer time, thus negating advantage in chemotactic efficiency. This leads to a strong selection against cells which adapt slowly and therefore tend to overreact to chemotactic stimulation. On the other hand, chemotaxis in agar poses only a weak selection against cells with a high adaptation rate.
Our simulations suggest that in liquid media the variability in protein levels among cells may be advantageous for bacterial populations on a long time scales. In a nonmixed environment with different food sources and gradient intensities, such variability can help the whole population to respond to different gradients more readily, due to positive selection of subpopulations with optimal levels of adaptation enzymes in a given gradient.
During a run, the cell is assumed to move with a constant speed v = 20 µm/s, while the direction of motion is affected by rotational diffusion [7],[42]. After each time step, the running direction is changed by adding a stochastic component with normal distribution and diffusion coefficient Dr = 0.062 rad2 s−1 [42].
The virtual cells are swimming in a 2D environment with a predefined attractant concentration field S(x, y, t). The domain geometry is either rectangular or circular, with reflecting walls. The simulation time was chosen to be short enough to avoid boundary effects. The rectangular domain is within (0, xmax)×(0, ymax), and the circular domain within (0, rmax), with xmax = ymax = 2rmax = 20 mm.
The output file of the RapidCell program contains the key characteristics of the intracellular state (CheY-P level, methylation state, motor bias) and the geometric characteristics of cell motion (position and orientation). The model was implemented using Java classes similar to AgentCell [19], but with simplified architecture. The algorithm is implemented as a discrete-time Monte Carlo scheme with time step Δt = 0.01 s. For random-number generation, we used external Java libraries [49],[50]. The code was written using Eclipse SDK (www.eclipse.org). The output data were analyzed with MATLAB (The MathWorks, MA).
To test our model (Figure 1A), we compared cellular behavior in the proposed universal constant-activity gradient with other gradients, observing the single cell swimming (Figure 1B) and the behavior of large populations. The key characteristics we consider are the CheY-P concentration and the drift velocity along the gradient.
We used the constant-activity gradient to study the effect of adaptation rate on chemotactic efficiency. For this purpose, we simulated homogeneous populations consisting of cells with the same adaptation rate. In a fixed constant-activity gradient, the population drift velocity depends on adaptation rate in a unimodal manner (Figure 5A). A zero level of adaptation enzymes (non-adapting cells) results in a low drift velocity, though it is clearly distinguishable from non-chemotactic behavior. A proportional increase of adaptation rate improves cellular drift velocity up to a certain maximum, after which it slowly declines again. Extremely high adaptation rates, more than 100 times higher than wild-type, make the cells non-chemotactic (Figure 5A).
To study chemotactic efficiency as a function of gradient steepness, cells were simulated in six constant-activity gradients with the steepness changing 64-fold, from 1.14 to 72.88×10−3, (Figure 5B). In each gradient, we determined the optimal adaptation rate, at which cellular drift velocity reaches its maximum. The simulated drift velocities are in the same range as those measured experimentally for E. coli in steep gradients (7 µm s−1) [53]. Our simulations indicate that experimental cell-drift velocities are inlikely to exceed 15 µm s−1, corresponding to an extremely steep and short-scale gradient. In very weak gradients, the drift velocity can be as low as 2.5 µm s−1, still distinguishable from the non-chemotactic cellular drift (0.8 µm s−1). Interestingly, we observed that the optimal adaptation rate rises in proportion with the gradient steepness (Figure 5B).
To investigate the latter effect in more detail, we varied the adaptation rate from 0 to 10-fold relative to the wild-type. In steeper gradients, the optimal adaptation rate is indeed higher (Figure 6A), and the peak of the drift velocity becomes less sharp. To find the reason for the observed dependence between the gradient steepness and optimal adaptation rate, we tracked the average CheY phosphorylation levels of the virtual cells. As one can see in Figure 6A and 6B, in all gradients the 90%-intervals around the velocity peaks correspond to adaptation rate intervals [0.1,0.5], [0.4,1.5], [1,3], respectively. These three intervals fall into to the same interval [0.80≤CheY-P≤0.97], within the error of estimation. The optimal adaptation rates which give maximal drift velocities correspond to an average [CheY-P]∼0.9. In steep gradients, the profile of average CheY-P flattens, and the optimal adaptation rate becomes higher (Figure 6B).
The reason why the interval [0.80≤CheY-P≤0.97] corresponds to optimal chemotaxis is evident from the profile of motor bias as a function of CheY-P (Figure 6C). The interval [0.80≤CheY-P≤0.97] corresponds to the operating range of the motor [0.95≥mb≥0.72], where the dependence between mb and CheY-P is approximately linear. In this interval, chemotactic behavior is most efficient in liquid media. The optimal adaptation rate therefore sets the CheY-P level to fit the motor operating range. In steep gradients, the adaptation rate must be high enough to balance the strong excitation and set CheY-P within this optimal interval. In shallow gradients, adaptation must be slow enough to allow excitation, otherwise the cells become adapted before they are able to respond.
The effect of varying the [CheR] to [CheB] ratio was studied at fixed [CheB] in three constant-activity gradients N1, N2, and N3 in a liquid medium. The chemotactic efficiency dramatically decreases above [CheR] = 1 (Figure 7), because the resulting higher steady-state CheY-P level produces tumbling behavior. Below [CheR] = 1, chemotactic efficiency decreases slowly for N3, or goes up for the N1 and N2 gradients. The latter effect is caused by a shift of average CheY-P level to the optimal interval, where the chemotactic sensitivity is the highest due to a more optimal fit to the motor operating range.
We have further studied the effect of CheB phosphorylation feedback on chemotactic efficiency in a liquid medium. Under the assumption that [CheR] and [CheB] perfectly match each other (A* = 1/3), the CheBp-effect is positive when the adaptation rate is lower than the optimum, and negative when the adaptation rate is higher, in the given gradient (Figure 8A). This effect is caused by the reduction of CheB activity relative to CheR, when the kinase activity A is below the steady-state level (A* = 1/3), as described in the section ‘Model of E. coli Signaling Network’. The average CheY-P level is thus shifted up, which results in a positive or negative effect of CheB phosphorylation, depending on the rate of adaptation (Figure 8B).
The positive role of phosphorylation can be significantly increased when the ratio of [CheR] to [CheB] is non-perfect (Figure 8C). For example, 25%-active CheB can significantly counteract the strong negative effect of [CheR] = 1.25 in the N3 gradient—the drift velocity rises from 1.8 to 2.8 µm s−1 (55%). At [CheR] = 0.75 the effect is not so dramatic, but remains significant—the average drift velocities increase by about 10–15% in all three gradients. This suggests that CheB phosphorylation helps to maintain chemotaxis at fluctuating concentrations of CheR and CheB, when their ratio is not perfect due to gene-expression noise.
In the swarm assay in soft agar, bacteria consume an attractant, thereby creating a local gradient, and follow it in the form of a growing ring [54],[55]. We assume that the intensity of the moving gradient remains constant, and use the constant-activity gradient as a simple model for the swarm assay simulation. The constant-activity gradient provides a constant cellular-drift velocity at any distance from the center of the plate. This property allows us to use it as a stationary model of the real moving gradient of attractant.
In swarm assays, bacteria move in a labyrinth of agar filaments, with obstacles and traps along the cell's path. The cell can encounter traps during its run, and stays trapped until it makes the next tumble, as observed by Wolfe and Berg [55]. Therefore, non-adapting cells and non-tumbling mutants form the smallest rings. To simulate motility in such a porous medium as agar, we have introduced a new state of the cell, corresponding to a stop in a trap during a run (Figure 9). The positions of traps are not fixed in space. Instead, it is assumed that each cell encounters traps in an exponentially distributed time series, which mimics the random collisions of the cell with agar filaments. The mean free time between traps is set to 2.0 s to achieve biologically realistic drift velocities (about 1 µm s−1). While it is trapped, the cell remains stationary until it makes a tumble, whereupon normal run and tumble behavior resumes until the next stop occures [55].
In our model, we assumed that the levels of the adaptation enzymes CheR and CheB vary in a coordinated manner, leaving the [CheR]/[CheB] ratio the same as in the wild type. The ratio of CheR to CheB can be assumed to remain largely fixed because their genes are adjacent and transcriptionally coupled in the meche operon. The adaptation rate in our model is thus proportional to the level of co-expression of CheR and CheB, which will be further denoted as [CheR,CheB].
In order to study chemotactic efficiency at different adaptation rates in agar, we have experimentally measured chemotactic efficiency on swarm plates. In these experiments, CheR and CheB-YFP were co-expressed from one operon under control of a pBAD promoter and native ribosome-binding sites. The pBAD promoter gives expression levels lower or higher than the wild-type value, depending on the strength of arabinose induction. Mean protein levels in the population at a given induction were determined as described in Experimental Methods.
Experiment and simulations show that cells with [CheR,CheB] above a certain threshold perform chemotaxis equally efficiently (Figure 10A and 10B). However, the cells with [CheR,CheB] below the threshold have severely impaired chemotactic behavior. According to the simulations, cells with low [CheR,CheB] tend to run without tumbling and stay trapped most of the time. On the other hand, cells with extremely high [CheR,CheB] loose their sensitivity to the gradient and also have poor chemotactic efficiency (Figure S5).
This suggests a positive selection for cells with optimal [CheR,CheB] in liquid media—such cells can reach the nutrient source faster and have more available substrates for growth. In contrast, swimming in agar poses mainly negative selection—cells with low [CheR,CheB] are filtered out from the chemotactic population. The limits of motor bias for optimal chemotaxis in agar are also different from those in liquid media. As one can see in Figure 10C, the average CCW motor bias of successful cells is just slightly higher than the steady-state mb0. Cells with higher motor bias would drift faster in liquid media, but not in agar, because the period of time they remain trapped also increases with CCW motor bias.
To model swarm assays more realistically, we simulated cell populations with a log-normal distribution of [CheR,CheB] values. The mean (1.6) and standard deviation (0.48) are fitted to reproduce the variability of adaptation times observed for wild-type cells [33]: Tad = 311±150 s in response to a 0–10−3 M MeAsp step.
The scatter plot of distances travelled by cells along the gradient N2 in a liquid medium shows that a subpopulation with optimal [CheR,CheB] levels drifts more rapidly than other cells (Figure 11A). Simulations in the N3 gradient in agar show that cells with low [CheR,CheB] levels are hindered by agar traps, while other cells drift successfully (Figure 11B). In Figure 11C and 11D the same cells are colored from deep blue to red, according to their [CheR,CheB]. The outer edge of the bacterial ring in a liquid medium contains many blue cells with [CheR,CheB] between 0.5 and 2. In contrast, the outer edge in the agar contains a uniform mixture of cells with different [CheR,CheB] levels, while deep blue cells with low [CheR,CheB] tend to be left behind.
To confirm that chemotactic cells are selected for their [CheR,CheB] levels in swarm plates, cells expressing CheR and CheB-YFP from one operon were taken from two positions in the swarm ring—at the center and at the outer edge—and protein levels in individual cells were determined using fluorescence imaging. The cells collected near the center at a standard agar concentration (0.27%) have on average lower copy numbers of adaptation enzymes than cells at the outer edge, confirming the predicted selection against low copy numbers (Figure 12A). As expected, in the swarm plates with a reduced agar concentration (0.20%), the difference between center and outer edge is much smaller (Figure 12B), suggesting that there is no strong selection against low copy numbers in liquid media. It should be noted that agar concentrations below 0.20% do not produce a stable gel structure, and therefore that is probably the most liquid agar that can be used for swarm plate experiments.
Our simulations and additional experiments with a pTrc promoter, which gives much higher basal expression level of [CheR,CheB], show that very high levels of the adaptation enzymes, over 20-fold, can again decrease chemotactic efficiency in agar (Figures S5 and S6).
In this paper, we present RapidCell—a model of chemotactic E. coli, which allows us to study the effect of chemotaxis network properties on the behavior of large bacterial populations. RapidCell uses a hybrid model for pathway simulation, with mixed algebraic and ODE description instead of a fully stochastic model, AgentCell [19], or a complete system of ordinary differential equations, E. solo [23]. Our model allowed us to dramatically decrease in computational costs. Though many molecular details are skipped or modeled in a rapid-equilibrium (algebraic) approximation, the key steps of the network are reproduced in agreement with up-to-date experimental data. In contrast to detailed single-cell simulation programs which reproduce the noisy behavior of individual cells [19],[56], RapidCell is aimed at predicting the averaged behavior of bacterial populations, and to investigate how it is affected by the signaling network parameters, neglecting the intrinsic noise coming from molecular reactions. However, artificial sources of noise can be further added in the deterministic model of the signaling pathway. In the present version of RapidCell, the noise arises only from rotational diffusion and stochastic switching of the motors.
For the receptor cluster simulation, we used the mixed-receptor cluster MWC model [12],[28],[30], which accounts for the observed broad range of sensitivity and reproduces the recent in vivo FRET data [27]. Adaptation is modeled according to the mean-field approximation of the assistance-neighborhood model, with the assumption that the average methylation level of multiple receptors can be represented as a continuous rather than a discrete variable [30]. In contrast to the other reactions, methylation and demethylation are relatively slow and therefore described by an ODE. The ODE is integrated by the first-order Euler scheme to ensure high computational speed of the program, while the time step is chosen as 0.01 s to keep the simulation error low.
Taking into account the available experimental studies on tumble mechanics [31],[57], we use a voting model of run-tumble switching [13],[31],[44]. The model is in a good agreement with experimentally measured run and tumble times. However, more high-resolution experimental data on the interplay among multiple flagella during the run and the tumble would be necessary for a detailed model of run-tumble cellular behavior.
There are several types of gradients usually applied in computer models of chemotaxis. The linear gradient arises between stationary source and adsorber, and can often be observed under natural conditions. The Gaussian, another commonly used gradient, appears when a limited amount of molecules is injected into the medium from a micropipette or a similar source [42]. Other gradients that arise from general models of diffusion have hyperbolic or exponential shapes. However, all commonly used gradients have a ‘blind’ zone where receptors are saturated and cells do not respond. When cells drift along these gradients, the average profile of CheY-P changes dramatically, from a steep fall at low concentrations to a weakly stimulated state at high concentrations (Figure 3C). This makes it difficult to compare long-term chemotactic efficiency, because the average CheY-P and drift velocity are non-stable along the gradient.
To study chemotaxis systematically, we propose a new—constant-activity—type of gradient. This gradient has the unique property of providing the same CheY-P level and cellular-drift velocity over a wide range of ligand concentrations. The stability of the CheY-P level allows us to study properties of virtual chemotactic cells systematically, and to compare chemotactic behavior over long time periods and concentration ranges.
The form of the constant-activity gradient is derived from the MWC model, by formulating the differential equation for the gradient shape which will give a constant rate of receptor free energy change due to ligand binding. In earlier work, the condition of constant chemotactic response was studied using a phenomenological model of ligand binding, with a single dissociation constant KD [48]. The study of Block and co-authors showed that such a model can be simplified, and as a result an exponential ramp of ligand should give a constant response in the range between Cmin = 0.31KD and Cmax = 3.2KD, a prediction that was supported by their experiments [48].
In our study, we show that the differential equation for the constant-response gradient proposed in [48] is the result of the MWC model. We further solve this differential equation analytically, and find the exact form of the constant-activity gradient. This gradient grows similarly to the exponential function at moderate ligand concentrations, and increases faster than exponential at low and high concentrations (Figure 2A).
Our simulations show that the chemotactic response of the MWC model in the constant-activity gradient remains stable over four orders of ligand concentration—between 0.1 and 1000KD, in the case when Tsr receptors are fully insensitive to the ligand. However, in the case of (Me)-Asp, the Tsr receptors are able to respond non-specifically to high ligand concentrations, therefore above 100KD the cluster activity drops to zero in a mixed-receptor cluster [12],[27]. However, our simulations of population behavior consider only moderate Asp concentrations, so the cluster activity remains nearly constant in all observed cases.
The exponential ramp also gives nearly constant response in the MWC model, but over a much smaller range—between 0.5 and 3.0KD, in agreement with [48] and the recent study of Tu et al. [58].
We also show that the apparent dissociation constant KD can be estimated by either the arithmetic or geometric mean of Koff and Kon, but the geometric mean gives a better approximation over a wide range of ligand concentrations.
The shape of the constant-activity gradient is also close to a hyperbolic gradient, with the change of variables, KDCx/(1−Cx) = KD(1/y−1)∼KD/y, (y = 1−Cx, KD≪1). The hyperbolic gradient arises from simple models of diffusion, when ligand molecules are emitted from a spherical source into the surrounding medium. In nature, such conditions can be observed, for example, in aquatic ecosystems where microalgae leak organic matter attractive for bacteria [59]. This suggests that hyperbolic and exponential gradients with appropriate parameters can be good approximations for the constant-activity gradient.
In our model, the adaptation rate is assumed to be proportional to the co-varied concentration of the adaptation enzymes [CheR,CheB], and we use both terms to denote the rate of adaptation. However, increasing expression of the adaptation enzymes may lead to saturation of the adaptation rate at some point, because the enzymes will start working out of saturation kinetics. For these reasons, it is more correct to consider our results in terms of adaptation-rate effects on chemotaxis, whatever the origins of adaptation-rate variability may be.
The effect of adaptation rate on chemotaxis agrees in many respects with the results reported in [13] for optimal noise filtering of the chemotaxis signaling system. In their work, the authors demonstrated the existence of an optimal cutoff frequency, an analog of the adaptation rate in our study, for efficient chemotaxis. For a fixed linear gradient, they show the same shape of chemotactic efficiency as a function of cutoff frequency (Figure 3B in [13]) as we found in our simulations (Figure 5A). The authors also show that the optimal cutoff frequency depends on gradient steepness in a linear manner (Figure 5A in [13]), consistent with our results (Figure 5B) for steep gradients.
Our simulations in the constant-activity gradient suggest a simple biological mechanism that determines the optimal adaptation rate for a given gradient steepness. Different optimal adaptation rates correspond to a single CheY-P interval, which fits the linear range of the motor-response function. This means that the highest drift velocity in liquid media is observed when the CheY-P level is in the narrow interval fitting the operating range of the motor. In this range, the dependence between CheY-P and mb is approximately linear (Figure 6C).
We found that the CheB phosphorylation feedback can have either a positive or negative effect on chemotactic efficiency, depending on how it shifts the average CheY-P level relative to the region of linear motor response. In the case of non-perfect ratio of CheR to CheB, the CheB phosphorylation mechanism can partially counteract the negative effect of unbalanced [CheR]/[CheB], by shifting the average CheY-P towards the optimal region. This confirms that CheB phosphorylation can improve the chemotactic properties of cells with deviations in the ratio of [CheR]/[CheB], as well as in the ratios of other proteins, from the optimum [32].
Chemotactic behavior in liquid media differs from that in agar. We simulated agar effects using traps randomly distributed over time - a cell can encounter traps during its run, and stays trapped until it makes the next tumble, as observed by Wolfe and Berg [55]. This restricts cellular motility—cells that are highly biased towards running remain in traps longer. In agar, the region of optimal motor bias is very narrow and is just above the unstimulated state mb0, because higher bias increases the period of time cells remain in traps.
In our model, we did not take into account the growth of a bacterial populations. The typical swarm plate experiments last several hours, and cells grow and divide during the experiment, leading to variations in protein levels and to redistribution of proteins from generation to generation. However, the effect of different adaptation rates in our simulations is clearly visible already within one cell generation over 1000 s of model time (Figure 11B). The selection thus works on a time scale that is shorter than the generation time, which, in our opinion, justifies using a fixed protein distribution. Therefore, the addition of cell growth should not change our results qualitatively. In experiments, daughter cells with sub-optimal levels of CheR and CheB will rapidly fall behind the spreading swarm ring in the vicinity of the division site, while the subpopulation with optimal adaptation rates will be always at the front edge of the ring.
In most of our simulations, we assume that the CheR and CheB ratio is constant due to the genetic coupling between the two respective genes, and that cell-to-cell variation in adaptation rates arises from concerted variation in the levels of both enzymes [32]. We also investigated the effects of variation in the [CheR]/[CheB] ratio, which results from translational noise, and affect both the adaptation rate and the steady-state motor bias. In addition to these investigated sources of noise, there is intrinsic noise in the pathway activity which arises from the stochastic nature of (de-)methylation events. The latter sort of noise can also have positive effects on the spreading of cells in a ligand-free medium [56], and even on chemotactic drift in weak gradients [60]. Superposition of variable noise effects on chemotactic efficiency in variable gradients would be an interesting issue for further study.
In this work, we have estimated the variability in concerted CheR and CheB concentrations using available experimental data on cell-to-cell variability in adaptation times [33]. We assumed a log-normal distribution for protein concentrations, which also gives a log-normal distribution of adaptation times to a step-wise stimulus from 0 to 10−3 M MeAsp [33]. There are also other experimental estimates of cell-to-cell variation in adaptation times [34] and related simulations [61], but the adaptation rates observed in those experiments were several times higher, presumably due to different culture growth conditions.
Our simulations suggest some evolutionary implications. In liquid media with variable food sources and gradient intensities, variability in adaptation times (protein levels) among cells can help the whole population to respond to different gradients more readily, due to positive selection of cells with optimal [CheR,CheB]. In other words, for any given gradient steepness, there will be a subpopulation which has the best [CheR,CheB] to follow this gradient. In contrast, agar poses mainly negative selection on cell populations - cells with low [CheR,CheB] are filtered out from competition, while all other cells travel with approximately equal efficiency.
Inspired by the implementation of AgentCell, RapidCell focuses on highly efficient computation of large populations over long periods, keeping cell-response properties consistent with experimental data. The first version of RapidCell allows us to simulate E. coli populations of size 104–105 cells over a time scale of hours, while tracking the signal network dynamics of individual cells. These capabilities permit the modeling of cellular behavior on a macroscopic scale, as in swarm-plate experiments, and the prediction of properties of heterogeneous populations with noisy components of the signaling network.
|
10.1371/journal.pgen.1002708 | Inactivation of a Novel FGF23 Regulator, FAM20C, Leads to Hypophosphatemic Rickets in Mice | Family with sequence similarity 20,-member C (FAM20C) is highly expressed in the mineralized tissues of mammals. Genetic studies showed that the loss-of-function mutations in FAM20C were associated with human lethal osteosclerotic bone dysplasia (Raine Syndrome), implying an inhibitory role of this molecule in bone formation. However, in vitro gain- and loss-of-function studies suggested that FAM20C promotes the differentiation and mineralization of mouse mesenchymal cells and odontoblasts. Recently, we generated Fam20c conditional knockout (cKO) mice in which Fam20c was globally inactivated (by crossbreeding with Sox2-Cre mice) or inactivated specifically in the mineralized tissues (by crossbreeding with 3.6 kb Col 1a1-Cre mice). Fam20c transgenic mice were also generated and crossbred with Fam20c cKO mice to introduce the transgene in the knockout background. In vitro gain- and loss-of-function were examined by adding recombinant FAM20C to MC3T3-E1 cells and by lentiviral shRNA–mediated knockdown of FAM20C in human and mouse osteogenic cell lines. Surprisingly, both the global and mineralized tissue-specific cKO mice developed hypophosphatemic rickets (but not osteosclerosis), along with a significant downregulation of osteoblast differentiation markers and a dramatic elevation of fibroblast growth factor 23 (FGF23) in the serum and bone. The mice expressing the Fam20c transgene in the wild-type background showed no abnormalities, while the expression of the Fam20c transgene fully rescued the skeletal defects in the cKO mice. Recombinant FAM20C promoted the differentiation and mineralization of MC3T3-E1 cells. Knockdown of FAM20C led to a remarkable downregulation of DMP1, along with a significant upregulation of FGF23 in both human and mouse osteogenic cell lines. These results indicate that FAM20C is a bone formation “promoter” but not an “inhibitor” in mouse osteogenesis. We conclude that FAM20C may regulate osteogenesis through its direct role in facilitating osteoblast differentiation and its systemic regulation of phosphate homeostasis via the mediation of FGF23.
| A recent study demonstrated that the inactivating mutations in the FAM20C gene were associated with lethal osteosclerotic bone dysplasia characterized by a generalized hardening of all bones; this observation implied an inhibitory role of FAM20C during bone formation. However, in vitro studies revealed a contradictory finding that FAM20C accelerated the differentiation of cells forming the mineralized tissues. Here we generated Fam20c conditional knockout (cKO) mice, in which the gene was inactivated either in all tissues or specifically in the mineralized tissues. We also generated recombinant FAM20C protein and Fam20c transgenic mice. The cKO mice did not mimic the human skeleton abnormalities of osteosclerotic bone dysplasia, but exhibited rickets (softer bone) along with a significant reduction of serum phosphate level and a remarkable elevation of serum FGF23, a hormone known to promote phosphate wasting. A number of differentiation markers of the bone-forming cells were downregulated in the cKO mice. Recombinant FAM20C promoted the differentiation of mouse preosteoblasts. Introducing the Fam20c transgene did not lead to any abnormalities but rescued the bone defects of the cKO mice. Taken together, we conclude that FAM20C promotes the differentiation of osteoblast lineages and regulates phosphate homeostasis via the mediation of FGF23.
| FAM20C is a member of the “family with sequence similarity 20”. In mammals, this evolutionarily conserved protein family consists of three members: FAM20A, FAM20B and FAM20C. FAM20A was originally observed in the lung and liver and displays obvious differential expression in hematopoietic cells undergoing myeloid differentiation [1]. A viral mRNA transgenic mouse line with an accidental deletion of a 58-kb fragment in chromosome 11E1 encompassing part of the Fam20a gene and its upstream region showed growth disorder [2]. Recently, it was found that FAM20A is also expressed in ameloblasts and its mutations are associated with human amelogenesis imperfecta and gingival hyperplasia syndrome [3]. More recently, FAM20B was shown to be involved in cartilage matrix production and the ultimate regulation on the timing of skeletal development [4]. FAM20C is highly expressed in the mineralized tissues and identified as the causal gene for lethal osteosclerotic bone dysplasia (Raine Syndrome, OMIM 259775) [1], [5]–[7]. Given the high level of conservation in the C-terminal domains among the three FAM20 members and the their roles observed in the hard tissues, it is tempting to speculate that this evolutionarily conserved family might be a new cluster of molecules performing important functions in the development of the mineralized tissues.
Mouse FAM20C, also known as “dentin matrix protein 4” (DMP4) [5], contains 579 amino acid residues, including a putative 26-amino acid signal peptide at the N-terminus. A C-terminal region of approximately 350 amino acids (corresponding to residue218-residue569 in the mouse FAM20C sequence) has been named the “conserved C-terminal domain” (CCD), which is highly conserved among different species [1].
In a previous study, we systematically analyzed the expression and distribution of FAM20C in mouse bone and tooth using in situ hybridization (ISH) and immunohistochemistry (IHC) methods [7], which showed that FAM20C was highly expressed in the mineralized tissues; it was detected in the osteoblasts/osteocytes, odontoblasts, ameloblasts, and cementoblasts, as well as in the matrices of bone, dentin, and enamel. FAM20C was also detected in the epithelium of early-stage tooth germs and in the chondrogenic cells of long bones. The high expression levels of FAM20C in the mineralized tissues strongly suggest that it may play an important role in the formation and/or mineralization of these tissues.
Hao et al. showed that overexpression of mouse FAM20C accelerated the odontoblast differentiation process and silencing this molecule by siRNA inhibited cell differentiation, implying that this protein may be a factor promoting odontoblast differentiation [5]. Subsequently, Simpson et al. reported that the loss-of-function mutations in the FAM20C gene were associated with lethal/non-lethal osteosclerotic bone dysplasia (Raine Syndrome) [6], [8], an autosomal recessive disorder characterized by a generalized increase in the density of all bones; these data indicated that FAM20C might be a down-regulator of biomineralization, which apparently contradicts the mineralization-promoting properties of FAM20C observed by Hao et al.
In this study, we sought to determine the biological functions of FAM20C via generation and characterization of Fam20c conditional knockout (cKO) mice. Our data showed remarkable skeletal defects, along with a significant reduction of serum phosphate and a dramatic elevation of serum fibroblast growth factor 23 (FGF23) in the homozygous Fam20c cKO mice. The phenotypic profiles of the Fam20c-deficient mice resemble those of hereditary hypophosphatemic rickets in humans and rodents resulting from mutations in molecules affecting the regulation of FGF23 [9]–[15].
The mouse Fam20c gene consists of 10 exons and spans approximately 55-kb. To generate a conditional knockout allele for Fam20c, we constructed a targeting vector with loxP sites floxing exons 6∼9 which are highly conserved across species (Figure 1A); a number of mutations were identified in this region of the human FAM20C gene in patients with lethal osteosclerotic bone dysplasia [6]. The correct targeting events were confirmed by polymerase chain reaction (PCR) screening, and the presence of 5′ and 3′ loxP sites was determined by PCR product sequencing. Two correctly targeted ES cell clones were identified (Figure 1B, Clones 286 and 297), and both went through germline transmission. F1 Fam20cflox/+ heterozygous mice were crossbred with Sox2 promoter-Cre transgenic mice to generate “Sox2-Cre-Fam20cΔ/Δ” mice, in which exons 6∼9 were removed from both alleles of the Fam20c gene in the epiblasts at post coitum day 6.5 (E6.5). The presence of the floxed alleles and the absence of exons 6∼9 in the null alleles were confirmed by PCR genotyping (Figure 1C).
The lack of Fam20c mRNA in the Sox2-Cre-Fam20cΔ/Δ mice was shown by reverse transcription PCR (RT-PCR) performed with two sets of primers using mRNA extracted from the long bones (Figure 2A), as well as by in situ hybridization (ISH) carried out on the long bones (Figure 2B). The lack of FAM20C protein was determined by immunohistochemistry (IHC) analyses (Figure 2C) performed on the long bones using an affinity-purified anti-FAM20C polyclonal antibody [7].
Both male and female Sox2-Cre-Fam20cΔ/Δ (homozygous cKO) mice are infertile, while the Sox2-Cre-Fam20cΔ/+ (heterozygous cKO) mice have normal fertility. The Fam20cflox/flox mice and the heterozygous cKO mice did not demonstrate any phenotypic changes compared with their wild type (WT) littermates (data not shown), while the homozygous cKO mice displayed remarkable skeletal defects, indicating that the haploinsufficiency of Fam20c has no significant effects on the bone formation. We also bred the Fam20cflox/flox mice with the 3.6 kb Col 1a1-Cre mice to generate Col1a1-Cre-Fam20cΔ/Δ mice, which displayed skeletal defects similar to those observed in the Sox2-Cre-Fam20cΔ/Δ mice. In this report, we described in detail the analyses of phenotypic changes in the Sox2-Cre-Fam20cΔ/Δ mice while the X-ray and histology data of the long bone from the Col1a1-Cre-Fam20cΔ/Δ mice were included in one set of the figures to show the similarity between the global and mineralized tissue-specific cKO mice. The data regarding the Fam20c cKO mice refer to the analyses of the Sox2-Cre-Fam20cΔ/Δ mice unless otherwise stated.
The osteocytes in Fam20cΔ/Δ mice lost normal morphology and appeared immature as shown by resin-casted scanning electron microscopy (SEM) analyses (Figure 8A and 8B), indicating a faulty maturation process from osteoblasts to osteocytes. Backscatter SEM analyses revealed periosteocytic lesions (“halo”) surrounding the osteocytes in the Fam20c cKO mouse bone (Figure 8C and 8D). To determine the molecular changes associated with the immaturity of osteoblasts/osteocytes, we examined their terminal differentiation markers: type Ia collagen, dentin matrix protein 1 (DMP1), and osteocalcin (OCN). ISH (Figure 8E–8J), and real-time PCR analyses (Table 2) revealed a significant downregulation of these markers in the Sox2-Cre-Fam20cΔ/Δ mice. Microarray analyses using total RNA extracted from the calvaria of 3-week-old Sox2-Cre-Fam20cΔ/Δ mice and their WT littermates indicated that among the ∼45,000 molecules evaluated, 350 genes were upregulated by over 2.0 folds and 185 were downregulated. Real-time PCR analyses on selected genes confirmed the significant changes in a number of biomineralization regulators and key players in the Wnt and TGF-β signaling pathway associated with cell differentiation (Table 2) [9]–[15], [19]–[27], suggesting an essential role of FAM20C in the differentiation and mineralization of osteogenic cells. Notably, the most striking transcriptional alteration was FGF23 (upregulated by ∼110 folds), a phosphorus regulator mainly produced by osteoblasts/osteocytes [11], [28], [29]. Immunohistochemistry against FGF23 confirmed the dramatic elevation in the bone cells and bone matrix of Fam20c cKO mice (Figure 8K and 8L). The transcript levels of the above genes in the Sox2-Cre-Fam20cΔ/+ (heterozygous cKO) mice showed no difference from the WT mice (data not shown). Given the many similarities among the Fam20c cKO mice, Dmp1 KO mice and Hyp mice, we examined the expression levels of Fam20c in the Dmp1- and Phex- deficient mice, and the levels of Dmp1 and Phex in the Fam20c cKO mice by real-time PCR analyses. The Fam20c expression was not altered in the Dmp1 KO mice and Hyp mice (data not shown). The expression of Dmp1 was significantly downregulated (Table 2, Figure 8) while that of Phex was not affected (data not shown) in the Fam20c cKO mice. Ectonucleotide pyrophosphatase/phosphodiesterase (Enpp1), another molecule involved in regulating phosphorus homeostasis was slightly downregulated in the bone of the Fam20c cKO mice, but the change (∼1.4 folds) was not statistically significant from the WT (data not shown).
The bone phenotypes in the Fam20c cKO mice appear opposite to those observed in the patients associated with the human FAM20C mutations [6]. These contradictory results raise the question of whether FAM20C functions differently between the two species. The lentiviral shRNA-mediated “knockdown” of FAM20C in mouse preosteoblasts MC3T3-E1 cells, human Saos-2 cells (osteoblasts isolated from human osteosarcoma) and human mesenchymal stem cells (hMSC) revealed a remarkable downregulation of DMP1 (Figure 9A–9C), along with a significant upregulation of FGF23 in both the human and mouse cell lines (Figure 9D–9F), indicating that FAM20C may function similarly in humans and mice.
To examine the role of FAM20C during osteoblast proliferation and differentiation, recombinant mouse FAM20C protein was generated by insect cells using a Bac-to-Bac baculovirus system. The recombinant FAM20C added to the culture of MC3T3-E1 preosteoblasts promoted the mineral deposition (nodule formation) in a dose-dependent manner (Figure 10A), and significantly enhanced the transcription of DMP1, osteocalcin (OCN), and bone sialoprotein (BSP) (Figure 10B). Adding recombinant FAM20C to MC3TC-E1 cells did not alter the expression of FGF23 and the proliferation rate of the cells at all tested concentrations (data not shown).
Seeing the significant elevation of FGF23 in the bone cells of Fam20c cKO mice, we performed serum biochemistry analyses in 18-day- and 42-day-old mice (Table 3). The circulating FGF23 level was remarkably elevated in both the 18-day-old (∼200 folds) and 42-day-old (∼60 folds) cKO mice. Accordingly, the serum phosphorus level significantly decreased at both ages (∼2.5 folds in 18-day-old cKO, ∼2 folds in the 42-day-old cKO mice). The circulating PTH level was significantly elevated in both the18-day-old (∼8 folds) and the 42-day-old (∼5 folds) cKO mice. The serum 1,25(OH)2D3 level was significantly reduced (∼2 folds) in the 18-day-old cKO mice, while the serum 1,25(OH)2D3 level in the 42-day-old cKO mice was slightly higher than that of the WT, but the change in the older mice was not statistically significant. The serum calcium level slightly decreased in the 18-day- and 42-day-old cKO mice, but the reduction was not statistically significant from the WT mice. The blood urea nitrogen (BUN) level in cKO mice had no statistic difference from that of the WT mice at both ages, indicating that no renal failure was occurring in the cKO mice. The serum biochemistry results of the Sox2-Cre-Fam20cΔ/+ (heterozygous cKO) mice were not different from their WT literates (data not shown).
The transcriptional levels of renal Klotho and NaPi-2a were slightly lower in the 18-day-old Fam20c cKO mice, and significantly downregulated (∼3 folds) in the 42-day-old cKO mice. The renal 1α-hydroxylase level was significantly reduced (∼2 folds) in the cKO mice at both ages. We also observed remarkable upregulation of renal 24-hydroxylase in the 18-day-old cKO mice (∼30 folds) as well as in the 42-day-old cKO mice (∼7 folds) (Table 4). The transcriptional levels of these genes in the heterozygous Fam20c cKO mice had no difference from their WT littermates (data not shown).
Conditional transgenic (cTg) mice expressing the full length FAM20C were generated to test the gain of function in vivo. We obtained 15 lines of cTg mice, and three of them with the transgene expression levels of 4∼8 folds over those of the WT littermates were further analyzed. One line with the highest expression level of the transgene (approximately 8 folds over the WT, Figure 11A) was characterized in detail. No abnormalities were observed in the bone of any of the cTg mice by postnatal 6 weeks (Figure 11B). In addition, by breeding the cTg mice with the Fam20c cKO mice, we obtained mice expressing the transgene in the Fam20c knockout background (designated “Sox2-Cre-Fam20cΔ/Δ-cTg mice”). The Sox2-Cre-Fam20cΔ/Δ-cTg mice had no abnormalities in the skeleton by postnatal six weeks (Figure 11B and 11C), indicating that expressing the transgene fully rescued the defects of the Fam20c-deficient mice. Additionally, IHC staining using the anti-FGF23 antibodies showed that the expression of the Fam20c transgene rescued the altered expression of FGF23 in the Fam20c-deficient bone (Figure 11D). The fact that overexpressing FAM20C in the WT background did not cause defects in the bone, along with the observation that overexpressing the transgene rescued the Fam20c-deficient abnormalities, has provided further evidence that the defects in the hard tissues of the Sox2-Cre-Fam20cΔ/Δ mice resulted from the loss-of-function, and were not due to the gain-of-function.
In summary, the multipronged approaches in this study demonstrated that inactivation of FAM20C in mice led to rickets/osteomalacia, along with altered levels of serum phosphate and FGF23. The manifestations of the Fam20c-deficient mice are consistent with a diagnosis of hypophosphatemic rickets. The Fam20c-deficient cells in the mineralized tissues appeared immature and incapable of forming healthy tissues. It is likely that a combination of cell differentiation failure and hypophosphatemia resulting from the FGF23 excess led to the skeletal defects in the Fam20c-deficient mice.
Little is known about FAM20C, a new molecule. In vitro studies have shown that it promotes the differentiation and mineralization processes of undifferentiated mesenchymal cells and odontoblasts [5], whereas human genetic studies suggested that FAM20C might be a down-regulator (inhibitor) of bone formation and/or mineralization [6]. To answer critical questions regarding the biological roles of FAM20C, we generated Fam20c conditional knockout mice, in which exons 6–9 (majority of the conserved CCD region) were ablated. It is worth noting that most of the mutations identified in the patients with lethal osteosclerotic bone dysplasia were in exons 6–9 [6]. The Fam20c conditional knockout mice developed hypophosphatemic rickets but not osteosclerosis. We believe that the abnormalities in the Fam20c cKO mice resulted from the loss-of-function and were not due to the gain-of-function for this protein. This belief is based on the following observations: 1) deleting exons 6–9 (majority of the CCD) in the Fam20c cKO mice was most likely to inactivate this molecule; 2) the inheritance of the phenotypic changes in Fam20c cKO mice occurred in an autosomal recessive trait, while the gain-of-function is usually inherited in an autosomal dominant manner; 3) transgenic mice overexpressing the Fam20c transgene were normal; 4) the overexpression of the Fam20c transgene fully rescued the phenotypic changes in the Fam20c cKO mice; and 5) recombinant FAM20C promoted the differentiation and mineralization of MC3T3-E1 cells in a dose-dependent manner. These data combined with a significant downregulation of osteoblast differentiation markers in cKO mice suggest that FAM20C is essential to the differentiation of mineralizing cells and promotes the formation and mineralization of hard tissues, and thus, inactivation of this molecule leads to differentiation failure of the cells forming these tissues. Additionally, the Fam20c cKO mice developed hypophosphatemia with a remarkable elevation of the serum FGF23 level. We believe that a combination of cell differentiation failure and hypophosphatemia caused by the increase of serum FGF23 led to the skeletal defects in the Fam20c conditional knockout mice.
In 2007, Simpson et al. reported that the mutations of human FAM20C are associated with an osteosclerotic phenotype in some patients [6]. In a later study by the same group [8], Simpson et al. identified FAM20C mutations in two patients whom they believed were suffering from a “different type of Raine Syndrome”; these two patients did not show a generalized increase in bone density, with one case showed “manifestations consistent with a diagnosis of hypophosphatemic rickets”, as the authors stated. The osteosclerotic phenotype in some patients with FAM20C mutations appears opposite to that observed in the Fam20c-deficient mice. These contradictory results raise questions of whether different domains/fragments of FAM20C protein have different functions or if their functions are different between humans and mice.
In previous reports, the human FAM20C mutations in the osteosclerotic patients include point missense mutations and “splicing” mutations [6], [8]. The point mutations were located in different regions of the gene, including the region encoding the N-terminal portion of the protein and that corresponding to the C-terminal part. The 1309G→A mutation (D437N, in exon 7) observed in the “hypophosphatemic rickets”-like patient (Case 1 in [8]) is located between the two missense mutations 1121T→G (L374R, in exon 6) and 1603C→T (R535W, in exon 10) and was very close to a splicing mutation C1322-2A→G (in intron 7). The latter three mutations initially reported by Simpson et al. were associated with a generalized hypermineralization in the patients [6], while the former one was associated with “hypophosphatemic rickets” [8]. The secreted form of mouse FAM20C contains 553 amino acid residues (excluding a putative 26-amino acid signal peptide), and its calculated molecular mass is approximately 63 kDa. In our previous study [7], Western immunoblotting analyses of the culture medium from HEK-293 cells transfected with a pMES construct containing full-length mouse FAM20C cDNA demonstrated a single protein band at approximately 65 kDa, consistent with the expected mass of full-length mouse FAM20C. We did not observe any lower molecular weight protein bands that could be recognized by the anti-FAM20C antibodies. Similar results were documented in the analyses of the mouse C3H10T1/2 cells and MC3T3-E1 cells transfected with the FLAG-tagged FAM20C [5]. These observations indicate that mouse FAM20C may not be proteolytically processed into fragments. Taken together, these human and mouse data do not support the contention that different domains or fragments of FAM20C may perform different functions.
In this study, the lentiviral shRNA-mediated knockdown of FAM20C in the human mesenchymal stem cells and human osteoblasts led to a remarkable downregulation of DMP1, along with a significant upregulation of FGF23 (Figure 9B, 9C, 9E, and 9F). The findings in the human cells are consistent with the results in the shRNA-knockdown of FAM20C in mouse MC3T3-E1 cells (Figure 9A and 9D) and with the observations in the Fam20c conditional knockout bone (Figure 8, Table 2); these results indicate that FAM20C is likely to function similarly in humans and mice. Clearly, more studies are warranted to further clarify the discrepancy between the human and mouse data.
As a growth factor, FGF23 principally functions as a phosphaturic hormone via binding to the Klotho/FGF receptor (FGFR) complexes in the kidney [30], [31]. The binding of FGF23 to FGFR accelerates phosphate excretion into the urine, thereby inducing a negative phosphate balance, which helps maintain the serum phosphate levels in the normal range under physiological conditions (Figure 12). Elevation of the FGF23 plasma level is known to lead to renal phosphate-wasting and hypophosphatemia [11]–[15], [28], [32]–[35]. The main sources of FGF23 are the osteoblasts and osteocytes in the skeleton [11], [28], [29], and a number of studies have shown that inactivating mutations in certain molecules expressed by these bone cells increase the plasma level of FGF23, which leads to hereditary hypophosphatemic rickets [10], [11], [13], [14]. Inactivating mutations in the phosphate-regulating gene with homologies to endopeptidases on the X chromosome (PHEX) cause X-linked hypophosphatemic rickets (OMIM 307800) [10], and loss of DMP1 activity results in autosomal-recessive hypophosphatemic rickets (OMIM 241520) [11], [13]. The phenotypic changes in the Fam20c-conditional knockout mice share many similarities (hypomineralization, elevation of FGF23, hypophosphatemia) with those observed in the PHEX- and DMP1-deficient subjects. As in the cases of PHEX- and DMP1-deficiency, FGF23 was overexpressed in the bones of the Fam20c cKO mice (Figure 8K and 8L, Table 2). The overproduction of FGF23 by the bone cells is likely to be responsible for the elevation of this protein in the serum.
Mutations in four genes, FGF23 itself, PHEX, DMP1 and ENPP1, have been reported to remarkably increase the plasma levels of FGF23, leading to hereditary hypophosphatemic rickets [9]–[11], [13]–[15]. Dmp1-, Phex- and Fam20c-deficient mice shared similarities in osteomalacia, hypophosphatemia and the remarkable elevation of FGF23 in the circulation and skeleton. Interestingly, an alteration of Fam20c expression was not observed in the Dmp1 KO mice or Hyp mice, while remarkable downregulation of Dmp1 (but not Phex) was observed in the Fam20c cKO mice. An up-regulation of Dmp1 was observed in MC3T3-E1 cells treated with recombinant FAM20C. On the other hand, a remarkable down-regulation of Dmp1 was seen in human and mouse osteogenic cell lines treated with FAM20C-shRNA. These findings, along with the similarities of skeletal and serum changes between the Fam20c-deficient and Dmp1-deficient mice, raise the question of whether FAM20C regulates DMP1 (Figure 12). Clearly, further studies are warranted to answer this question. Additionally, there are still a pool of patients with hereditary hypophosphatemic rickets whose etiology is unknown [35], [36], and our discovery that the inactivation of Fam20c in mice results in hypophosphatemic rickets necessitates a consideration of screening FAM20C in such patients.
High level FGF23 reduces the expression of renal vitamin D 1α-hydroxylase and increases the expression of the catabolic 25-hydroxyvitamin D 24-hydroxylase, thus leading to decreased levels of 1,25(OH)2D3 in the serum [15], [37], [38]. In the 18-day-old Fam20c cKO mice, the 1,25(OH)2D3 level was significantly lower, while in the 42-day-old cKO mice the 1,25(OH)2D3 level managed to return to the normal range (or a not significantly higher level). Similar shifts in the serum 1,25(OH)2D3 level with aging have been observed in other hypophosphatemic models such as the Fgf23 transgenic mice, Hyp mice and Dmp1-KO mice that have high levels of circulating FGF23 [11], [12], [39].
While elevation of serum FGF23 reduces the expression of the renal 1α-hydroxylase, hypophosphatemia is normally a stimulator for renal 1α-hydroxylase expression to increase circulating 1,25(OH)2D3 [40]; the stimulating effect of hypophosphatemia on the 1α-hydroxylase expression has been well illustrated in the NaPi2a knockout mice (with lower phosphate and lower FGF23 levels in the serum), in which the serum 1,25(OH)2D3 level is elevated due to the increased 1α-hydroxylase expression stimulated by hypophosphatemia [41]. The Fam20c-cKO mice displayed a decreased level of renal 1α-hydroxylase in the presence of hypophosphatemia, indicating that in these mutant mice, the negative modulation of FGF23 on the expression of the 1α-hydroxylase may outweigh the stimulating effect of hypophosphatemia on the 1α-hydroxylase expression.
In comparison with the Fgf23 transgenic mice, Dmp1-KO mice and Hyp mice, a more remarkable upregulation of 24-hydroxylase was observed in the kidney of the Fam20c-cKO mice, which may be due to the fact that Fam20c-cKO mice have a higher serum FGF23 level than the former three [11], [12], [33]. A higher level of 24-hydroxylase in the 18-day-old cKO mice than that in the 42-day-old Fam20c cKO mice (∼30-fold versus ∼7-fold elevation) may be responsible for the significantly lower serum 1,25(OH)2D3 level in the younger animals. In the older Fam20c cKO mice, a significant reduction of FGF23 co-receptor Klotho, along with a relatively lower serum FGF23 level than that in the younger cKO mice, may attenuate the FGF23-elevation effects on circulating 1,25(OH)2D3 and thus may help maintain a relatively normal serum 1,25(OH)2D3 level in the older cKO mice. More likely, the lower serum level of 1,25(OH)2D3 in the younger Fam20c KO mice may have triggered the overproduction of PTH, as in the cases of Fgf23 transgenic mice and Hyp mice. The secondary hyperparathyroidism may play a critical role for reversing the 1,25(OH)2D3 level to the normal range in the older Fam20c cKO mice [42], [43]. In addition, the elevated PTH may synergize with the high level of serum FGF23 to increase renal phosphate excretion by reducing the expression of NaPi2a in the proximal tubules [38], [44]; a significant reduction of NaPi2a was observed in the kidney of the Fam20c cKO mice. In the end, 1,25(OH)2D3 may maintain a relatively normal level in the older Fam20c cKO mice at the expense of a significant phosphorus wasting.
The skeletal defects of the PHEX- and DMP1-deficient subjects are believed to be due to the combined effects of two factors: 1) the intrinsic defects of the PHEX- and DMP1-deficient cells that prevent them from forming and mineralizing ECM properly and 2) hypophosphatemia [11], [33], [45], [46]. The Fam20c-deficient cells responsible for forming the mineralized tissues appeared immature and showed altered expression levels for molecules associated with cell differentiation. While hypophosphatemia in the Fam20c conditional knockout mice can be attributed to the overproduction of FGF23 in the abnormal skeleton, the direct cause of cell differentiation failure may be complicated. As stated above, the defects in the mineralized tissues of the Fam20c conditional knockout mice could be the combined results of cell differentiation failure and hypophosphatemia. Although the way FAM20C regulates cell differentiation has not yet been defined in this study, our data suggest that FAM20C may regulate the differentiation and function of the mineralizing cells by participating in certain signaling pathways.
Several lines of evidence suggested that FAM20C might be associated with the canonical Wnt signaling pathway. The Wnt canonical pathway inhibitors, secreted frizzled related protein 1 (Sfrp1) and Sfrp3, were upregulated in the Fam20c-deficient mice. Accordingly, the downstream target genes of Wnt pathway, leucine-rich repeat-containing G protein-coupled receptor 5 (Lgr5) and lymphoid enhancer-binding factor 1 (Lef1) were significantly downregulated in the Fam20c-deficient bone. Lgr5 is a stem cell marker and a Wnt pathway regulator which has been identified in multiple tissues including bone marrow cells [23], [24]. Lef1 is a Wnt-responsive transcription factor that associates with β-catenin and has been documented to increase osteoblast activity and trabecular bone mass [25]. However, not all of the findings support the postulation that FAM20C is a participant in the Wnt signaling pathway. For example, the level of axis inhibition protein 2 (Axin2), a putative Wnt downstream target gene, was unchanged in the Fam20c-deficient bone. In addition to the molecules in the Wnt signaling pathway, Follistatin (Fst), a potent inhibitor of Activin and the TGF-β pathway, was significantly upregulated in the bones of 3-week and 6-week-old Fam20c-deficient mice. Fst has been reported to inhibit ameloblast and osteoblast differentiation [19], [20], suggesting a possible association between FAM20C and the TGF-β pathway.
Type II and Type X collagen were downregulated in the growth plates of the Fam20c conditional knockout mice. The downregulation of these genes may arise from the intrinsic defects of chondrocytes or occur as a systematic consequence of hypophosphatemia. It has been well documented that hypophosphatemia significantly decreases programmed cell death in growth plates by impairing caspase-mediated apoptosis of hypertrophic chondrocytes [17], [18].
In conclusion, the results in this investigation have demonstrated the crucial role of FAM20C in osteogenesis. Our findings indicate that FAM20C is essential to the differentiation of osteoblasts/osteocytes and is involved in the regulation of phosphate homeostasis via the mediation of FGF23.
All animal procedures were performed in accordance with the NIH Guide for the Care and Use of Laboratory Animals and approved by the Institutional Animal Care and Use Committee of Texas A&M Health Science Center, Baylor College of Dentistry (Dallas, TX, USA).
To generate the Fam20c conditional knockout mice, a 2.2 kb targeting fragment spanning exons 6∼9 of mouse Fam20c was produced by PCR using the genomic DNA of WW6 ES cells as a template (forward primer: 5′-CTCTCGGGTGAGGCTGTAAG-3′; reverse primer: 5′-AGATCTCTTAGGGAAGAGGGGTCAGG-3′). The fragment was subcloned into a floxed BamHI site upstream of an Frt flanked mcl-Neo cassette in the conditional targeting vector pFlox-Frt-Neo [47]. A 2.5 kb 5′ homologous arm was generated by PCR (forward primer: 5′-CTCGAGTGGGTGTGTCAGGAATCGTA-3′; reverse primer: 5′-CTCGAGACCCGAGAGCAACCACATAC-3′), and subcloned into the XhoI site of the targeting vector. A 4.2 kb 3′ homologous arm with EagI sites at both ends was generated by PCR (forward: 5′-TCGGCCGTTGGACATAGGCTCCCAAAG-3′; reverse: 5′-TGTGCAGGATTGAGAACCAG-3′), and subcloned into the NotI site of the targeting vector. Finally, a negative selection PGK-DTA (diphtheria toxin A) cassette was subcloned into the NotI site downstream of the 3′ homologous arm in the targeting vector (Figure 1A). The final targeting construct was linearized with SacII and electroporated into W4 ES cells (Transgenic Mouse Core Facilities, University at Albany, Rensselaer, NY, USA). Clones were picked after G418 selection. Genomic DNA was extracted from the ES cells in duplicate plates, and PCR analyses were performed to screen the targeted clones (5′ screen primers: 5′S-F: 5′-TTTCTGTCCTAGGTAAGGGTGAAG-3′, 5′S-R: 5′-ACTGCTCGATGAAGTTCCTATTCT-3′; 3′ screen primers: 3′S-F: 5′-TGTTCGGATCGAAGTTCCTATACT-3′, 3′S-R: 5′-ACAGCTTCTTGAATTGGGATAAAG-3′) (Figure 1A and 1B). The 5′ and 3′ screening PCR products were sequenced to confirm the correct targeting and the presence of 5′ and 3′ loxP sites.
Two correctly targeted ES cell clones (Clone 286 and 297, Figure 1B) were identified. The neo cassette was removed from the targeted ES clones by transient transfection of pCAGGS-flpE-puro vector (Addgene plasmid 20733) [48]. The removal of the neo cassette was confirmed by PCR (forward: 5′-GCATCTGCAGACCGAGCCCA-3′, reverse: 5′-CCCCCTGTCCTGAGGGCTGA-3′). Random integration of pCAGGS-flpE-puro DNA into ES cell genome was excluded by PCR analyses (forward: 5′-GCATGGCCGAGTTGAGCGGT-3′, reverse: 5′-GGTGACGGTGAAGCCGAGCC-3′). The targeted ES clones recovered from the master plates were injected into the blastocysts of C57BL/6 mice in the Transgenic Core Facility of the University of Texas Southwestern Medical Center at Dallas. Male chimeras were crossbred with C57BL/6 females to produce F1 agouti offspring. The floxed alleles of F1 agouti mice were genotyped by PCR analyses (see below). To generate Fam20c conditional knockout mice, the F1 heterozygous mice (Fam20cflox/+) were first crossbred with Sox2 promoter-driven Cre transgenic mice (Jackson Laboratory) that express the Cre recombinase transgene in the epiblasts at E6.5; this breeding gave rise to Sox2-Cre-Fam20cΔ/+ mice in which exons 6∼9 were removed from one allele of the Fam20c gene. The Sox2-Cre-Fam20cΔ/+ mice were inbred to produce Fam20c conditional knockout (cKO) mice designated as “Sox2-Cre-Fam20cΔ/Δ mice”, in which exons 6∼9 were removed from both alleles of the Fam20c gene.
Genotypes were determined by PCR analyses using genomic DNA extracted from the mouse tails. The floxed allele was distinguished from the wild type (WT) allele by PCR analyses using a mixture of three primers (a: 5′-TCCAGCTTGCTAGGGCTCTGACC-3′, b: 5′-CTATGTCCAACGGCCGCAGCTT-3′, and c: 5′-GTCCTGAGGGCTGACCCAAGACTA-3′) (Figure 1A and 1C). The null allele (i.e., that with exons 6∼9 removed) arising from Cre-loxP recombination events was detected by PCR using primers Rec-F: 5′-GTGGTCTCTGCCGCTGATGTACC-3′ and Rec-R: 5′-TTTGGGAGCCTATGTCCAACGGCC-3′ (Figure 1A and 1C). Genotyping for the Cre transgene was determined by PCR using primers Cre-F: 5′-CCCGCAGAACCTGAAGATG-3′ and Cre-R: 5′-GACCCGGCAAAACAGGTAG-3′.
We also used the 3.6 kb Col 1a1 promoter-Cre transgenic mice (The Jackson Laboratory) to generate “Col1a1-Cre-Fam20cΔ/Δ mice”, in which FAM20C was inactivated in tissues expressing type I collagen. We first crossbred the Fam20cflox/+ mice with 3.6 kb Col 1a1 -Cre mice, and then inbred the offspring of the Col1a1-Cre-Fam20cΔ/+ mice to get Col1a1-Cre-Fam20cΔ/Δ mice.
To generate Fam20c conditional transgenic mice, the coding sequence of mouse Fam20c cDNA was subcloned into a bicistronic pMES vector [49] downstream to a chicken β-actin promoter and upstream to an IRES-EGFP cassette as previously described [7]. A floxed STOP cassette was inserted between the β-actin promoter and the Fam20c sequence to block the transcription of the transgene. The Fam20c transgene can be activated only after the floxed STOP cassette is removed by Cre recombinase [7]. This Fam20c conditional transgenic construct was linearized and injected into the pronuclei of fertilized eggs from the C57BL/6 mice in the Transgenic Core Facility of the University of Texas Southwestern Medical Center at Dallas. Fifteen lines of transgenic mice with the conditional Fam20c transgene were identified by PCR genotyping using primers located in the exogenous EGFP sequences (GFP-forward: 5′-ACGTAAACGGCCACAAGTTC-3′ and GFP reverse: 5′-TGCTCAGGTAGTGGTTGTCG -3′). The mice carrying the conditional transgene were crossbred with the Sox2-Cre transgenic mice to remove the floxed STOP cassette between the chicken β-actin promoter and the Fam20c cDNA, thereby allowing the Fam20c transgene to be transcribed. These conditional transgenic mice (cTg mice) were genotyped using the aforementioned GFP primers and Cre primers. The expression level of the Fam20c transgene in the long bones of each line was evaluated by quantitative real-time PCR. Three lines with the transgene expression levels of 4∼8 folds over that of the WT mice were further analyzed.
To generate mice expressing the transgene in the Fam20c conditional knockout background, we crossbred Sox2-Cre-Fam20cΔ/+ mice with cTg mice expressing the highest level of the trangene to obtain Sox2-Cre–Fam20cΔ/+-cTg mice, which were then inbred to produce mice expressing the transgene in the Fam20c conditional knockout background (designated “Sox2-Cre-Fam20cΔ/Δ-cTg mice”). PCR analyses with primers used in the identification of the Fam20c conditional knockout mice and cTg mice were employed in the genotyping of the Sox2-Cre-Fam20cΔ/Δ-cTg mice.
All mice in this study were fed with Teklad 6% fat mouse/rat diet (Harlan, IN) and some of the chow contents are as follows: Calcium 2.4%, phosphorus 1.5%, vitamin D 3.0 IU/g.
Femurs from the Fam20c conditional knockout mice, Sox2-Cre-Fam20cΔ/Δ-cTg mice and the WT littermates were dissected, and total RNA was extracted using an Rneasy Mini Kit (Qiagen) according to the manufacturer's instructions. The total RNAs were converted into cDNAs using a Reverse Transcription Kit (Qiagen). RT-PCR was performed to examine the lack of Fam20c mRNA in the cKO mice using two sets of primers: Set 1-F: 5′-TGCGGAGATCGCTGCCTTCC-3′, Set 1-R: 5′-GCCACTGTCGTAGGGTGGCG-3′; Set 2-F: 5′-GAGAGCAGGAGACGCCGCCT-3′, and Set 2-R: 5′-CCACCACACTGCTCAGCCCG -3′ (Figure 2A).
One-week-old Sox2-Cre-Fam20cΔ/Δ mice and WT littermates were skinned, eviscerated and fixed in 95% ethanol. Alizarin Red/Alcian Blue staining of the skeletons was performed to visualize the skeleton and the overall mineralization levels, as described previously [50].
The narcotized mice or the dissected jaws and long bones from hind legs were analyzed using X-ray radiography (Faxitron MX-20DC12). Micro-computed tomography (Micro-CT) analyses were performed using a Scanco micro-CT35 imaging system (Scanco Medical) with a medium-resolution scan (7.0 µm slice increment) on the dissected tissues, as previously reported [51]. The images were reconstructed with the EVS Beam software using a global threshold at 240 Hounsfield units.
Tibia and jaw tissues dissected from the mice were fixed with 4% paraformaldehyde in 0.1% diethyl pyrocarbonate (DEPC)-treated PBS solution at 4°C overnight and then were decalcified in 0.1% DEPC-treated 15% EDTA (pH 7.4) at 4°C for 8 days. The tissues were processed for paraffin embedding, and serial 5 µm sections were prepared for histological analyses. H&E staining was performed as previously described [17].
BrdU was administrated to 3-week-old Sox2-Cre-Fam20cΔ/Δ mice and WT littermates at a dosage of 1 ml per 100 g body weight by intraperitoneal (i.p.) injection according to the manufacturer's instructions (Invitrogen). Two hr after the injection, the mice were sacrificed. Tibias were dissected and processed for paraffin embedding, and 5 µm sections were prepared for BrdU detection using a Zymed BrdU staining kit (Invitrogen) following the manufacturer's instructions. Apoptosis in growth plates was examined by TUNEL assay using the ApopTag Plus Fluorescein In Situ Apoptosis Detection Kit (Millipore) according to the manufacturer's instructions. Six serial sections from each of six individual samples of Sox2-Cre-Fam20cΔ/Δ mice and WT littermates were counted, and the data were analyzed statistically.
The IHC experiments were carried out using an ABC kit and a DAB kit (Vector Laboratories) according to the manufacturer's instructions. A polyclonal C-terminal anti-FAM20C antibody was used at a concentration of 1 µg IgG/ml for the IHC experiments, as previously described [7]. A monoclonal FGF23 antibody (Cell Essentials) was used at a dilution of 1∶400 following the manufacturer's instruction. A polyclonal biglycan antibody (LF-159) was kindly provided by Dr. Larry Fisher (NIDCR, National Institutes of Health) [52]. Methyl green was used for counterstaining.
A 380 bp fragment from the region of exons 6–9 of Fam20c cDNA was obtained by PCR using forward primer 5′- CCGAGCATGCCCTGTGTGGG -3′ and reverse primer 5′- TGCAGCACTGATGAAGAGGAGCG -3′. The PCR product was subcloned into the pCRII-TOPO vector (Invitrogen) and then linearized with EcoRV to synthesize the antisense RNA probes using the Sp6 RNA polymerase or with HindIII to synthesize the sense RNA probes using the T7 RNA polymerase. The constructs used to generate RNA probes for DMP1, osteocalcin, collagen type I, collagen type II, and collagen type X were provided by the laboratory of Dr. Jian Q. Feng. The constructs were linearized and labeled with digoxigenin (DIG) using a RNA Labeling Kit (Roche, Indianapolis, IN) as previously described [7]. DIG-labeled RNA probes were detected by an enzyme-linked immunoassay with a specific anti-DIG-AP antibody conjugate (Roche) and an improved substrate (Vector Laboratories), which produces a red color for positive signals, according to the manufacturer's instructions. Methyl green was used for counterstaining.
Tibias dissected from 6-week-old mice were fixed in 4% paraformaldehyde overnight. The specimens were dehydrated through a graded series of ethanol (70–100%) and embedded in methylmethacrylate (MMA) without prior decalcification, as previously described [53]. Ten µm sections were prepared for Goldner staining and double-labeling fluorescent analysis.
Double fluorescence labeling was performed as previously described [54]. Briefly, calcein (5 mg/kg i.p.; Sigma-Aldrich) was administered to the 5-week-old mice, followed by injection of an Alizarin Red label (20 mg/kg i.p.; Sigma-Aldrich) 7 days later. The mice were sacrificed 48 hr after the injection of the second label and the tibias were embedded in MMA; 10 µm sections were then prepared. The unstained sections were viewed under epifluorescent illumination using a Nikon E800 microscope, interfaced with Osteomeasure histomorphometry software (version 4.1, Atlanta, GA). The mean distance between the two fluorescent labels was determined and divided by the number of days between labels to calculate the mineral deposition rate (µm/day).
Ten µm undecalcified sagittal sections from the tibias were stained using Goldner-Masson trichrome assay, as previously described [55]. The cortical bone areas in the midshaft were photographed using a Nikon microscope at 10× with Bioquant OSTEO v.7.20.10 (R&M Biometrics) software. Unmineralized osteoid stains red, and mineralized bone stains green/blue.
For resin-casted osteocyte lacunocanalicular SEM, the surface of the MMA embedded tibia was polished, acid-etched with 37% phosphoric acid for 2–10 s, washed with 5% sodium hypochlorite for 5 min and then coated with gold and palladium and examined by FEI/Philips XL30 Field emission environmental SEM. Backscattered SEM was performed as we previously described [56]
Total RNAs were isolated from the calvaria bones and decapsulated kidneys of 3-week-old mice and cultured cells. The kits for RNA extraction and reverse transcription were the same as in the RT-PCR experiments. Quantitative real-time PCR was performed on a Bio-Rad CFX96 system (Bio-Rad) using SYBR Green Master Mix (Stratagene). The Ct values were normalized to the reference gene 18s rRNA (SABiosciences), and then expressed as fold changes compared with the experimental controls. The primers for human 18s rRNA, mouse 18s rRNA, mouse Fam20c (NM_030565) and human FGF23 (NM_020638) were bought from SABiosciences. All other primers were synthesized by Integrated DNA Technologies (Table S1).
Microarray analyses were performed in the Microarray Core Facility of University of Texas Southwestern Medical Center at Dallas, using total RNA extracted from the calvaria of 3-week-old mice. GeneChip Mouse Genome 430 2.0 Array (Affymetrix) was employed in the microarray analyses, following the manufacturer's instructions. Data analyses were performed using GeneSpring software (Agilent Technologies). The microarray results were uploaded to MAGE-TAB ArrayExpress database (accession number E-MTAB-772).
Serum phosphorus was measured using the phosphomolybdate-ascorbic acid method, as previously described [57]. Serum calcium was measured using a colorimetric calcium kit (Stanbio Laboratory). The serum FGF23 and PTH levels were measured using a full-length FGF23 ELISA kit (Kainos Laboratories) and a mouse intact PTH ELISA kit (Immutopics). Serum 1,25(OH)2D3 was measured using a 1,25 Dihydroxy Vitamin D EIA Kit (Immunodiagnostic Systems). Blood urea nitrogen (BUN) was measured using a BUN Reagent Kit (BQ Kits).
The recombinant mouse FAM20C was expressed by insect cells using a Bac-to-Bac baculovirus expression system (Invitrogen). Briefly, the N-terminal of mouse FAM20C (signal peptide removed) was fused with a baculovirus signal sequence gp67 (envelope glycoprotein), 6xHis (tag), SUMOstar (Small Ubiquitin-like Modifier), and a TEV (Tobacco Etch Virus) cleavage site. The fusion gene was inserted into the pFastDual vector (Invitrogen) downstream of a polyhedrin promoter, and a GFP cDNA was inserted downstream of a P10 promoter serving as a baculovirus indicator. The construct was transformed into DH10Bac E. Coli cells (Invitrogen), in which the fusion gene was introduced into BacMid via homologous recombination. The BacMid was extracted from DH10Bac E. Coli cells and transfected into Sf21 insect cells (Invitrogen) to produce the baculovirus. The insect cells infected with the baculovirus secreted the recombinant FAM20C into the SFX-insect cell culture medium (Hyclone). After two rounds of scale-up, the cell culture medium was collected and subjected to a one-step Ni-NTA purification. Turbo-TEV (Eton Bioscience) was used to release mouse FAM20C from the fusion protein, and reverse Ni-NTA purification was performed to remove Turbo-TEV, His tag, and SUMOstar.
Mouse MC3T3-E1 cells were grown in α-MEM medium (Gibco) supplemented with 10% fetal bovine serum (Hyclone) and antibiotics (Gibco); human Saos-2 cells (ATCC, osteoblasts from human osteosarcoma) were grown in McCoy's 5a medium (Gibco) supplemented with 15% fetal bovine serum; human mesenchymal stem cells (hMSC) (Lonza) were maintained in MSCGM BulletKit medium (Lonza). Cells were treated with recombinant FAM20C or the viruses when they reach 80% confluence. For the gain-of-function analyses, MC3T3-E1 cells were treated with mouse recombinant FAM20C at the concentrations of 200 ng/ml, 400 ng/ml and 800 ng/ml. For the loss-of-function studies, MC3T3-E1, Saos-2 and hMSC cells were infected with the mouse or human shRNA-lentiviruses (all from Santa Cruz) containing a mixture of three target-specific shRNA sequences against the mouse or human FAM20C. A lentivirus expressing the scrambled shRNA (with no specific target in the genome) served as the control virus (Santa Cruz). The infection rate was monitored by infecting these cells with another control virus (Santa Cruz) expressing the GFP indicator. After 1-week selection with 5–8 µg/ml concentrations of puromycin (Santa Cruz), the control virus with GFP showed a nearly 100% infection rate. To induce osteogenic differentiation for the MC3T3 cells and Saos-2 cells, the culture medium was supplemented with 100 µg/ml ascorbic acid, 10 mM β-glycerophosphate and 30 nM dexamethasone; the osteogenic differentiation of human MSC was induced using the Osteogenic BulletKit (Lonza) following the manufacturer's instruction. Total RNA was extracted from the cells at different time points. For the gain-of-function analyses, RNA was extracted after 3 weeks of induction. For the “shRNA knockdown” analyses, RNA was extracted at 3 days after the lentiviral infection and before the osteogenic medium (inducing the osteogenic differentiation) addition, and at 1, 2 and 3 weeks after the start of osteogenic induction. Real-time PCR was performed to evaluate the mRNA levels of the selected genes. The mineral deposition rate was determined by nodule formation and Alizarin red concentration in each well were measured using the Osteogenesis Quantitation kit (Millipore) for MC3T3-E1 cells treated with recombinant FAM20C (the gain-of-function experiments). MC3T3-E1 cells and hMSC cells infected with the FAM20C-shRNA viruses became unhealthy (showing lot of cell death) after 4-week culture. The Saos-2 cells could not survive in the osteogenic medium for longer than 2 weeks. The data collected from the culture of cells at the “unhealthy stages” were discarded.
The data are expressed as the mean ± SD of at least 6 individual determinations in all experiments unless otherwise indicated. We statistically evaluated the data employing ANOVA to test for any differences among the sample groups. When a difference was determined, 2-sample t tests were employed to evaluate all possible pairs of samples.
|
10.1371/journal.ppat.1002501 | Msb2 Shedding Protects Candida albicans against Antimicrobial Peptides | Msb2 is a sensor protein in the plasma membrane of fungi. In the human fungal pathogen C. albicans Msb2 signals via the Cek1 MAP kinase pathway to maintain cell wall integrity and allow filamentous growth. Msb2 doubly epitope-tagged in its large extracellular and small cytoplasmic domain was efficiently cleaved during liquid and surface growth and the extracellular domain was almost quantitatively released into the growth medium. Msb2 cleavage was independent of proteases Sap9, Sap10 and Kex2. Secreted Msb2 was highly O-glycosylated by protein mannosyltransferases including Pmt1 resulting in an apparent molecular mass of >400 kDa. Deletion analyses revealed that the transmembrane region is required for Msb2 function, while the large N-terminal and the small cytoplasmic region function to downregulate Msb2 signaling or, respectively, allow its induction by tunicamycin. Purified extracellular Msb2 domain protected fungal and bacterial cells effectively from antimicrobial peptides (AMPs) histatin-5 and LL-37. AMP inactivation was not due to degradation but depended on the quantity and length of the Msb2 glycofragment. C. albicans msb2 mutants were supersensitive to LL-37 but not histatin-5, suggesting that secreted rather than cell-associated Msb2 determines AMP protection. Thus, in addition to its sensor function Msb2 has a second activity because shedding of its glycofragment generates AMP quorum resistance.
| Microbial pathogens are attacked by antimicrobial peptides (AMPs) produced by the human host. AMPs kill pathogens and recruit immune cells to the site of infection. In defense, the human fungal pathogen Candida albicans continuously cleaves and secretes a glycoprotein fragment of the surface protein Msb2, which protects against AMPs. The results suggest that shed Msb2 allows fungal colonies to persist and avoid inflammatory responses caused by AMPs. Msb2 shedding and its additional role in stabilizing the fungal cell wall may be considered as novel diagnostic tools and targets for antifungal action.
| Crosstalk between pathogens and the human host determines the outcome of microbial colonization and disease [1]. Pathogen-host communication occurs between cells and secreted proteins of both organisms. Surface structures of the important human fungal pathogen Candida albicans bind to dectin receptors on immune cells and trigger responses inhibiting fungal proliferation including the production of antimicrobial peptides (AMPs) and reactive oxygen species (ROS) (for a review, see [2], [3]. In addition, binding to immunoglobulins and complement factors by the fungal pathogen facilitate its phagocytosis and killing (for a review, see [4]). Conversely, C. albicans partially overcomes host defenses by secreting hydrolytic enzymes and proteins that block the complement system (for a review, see [4], [5]). Furthermore, by switching its growth from a yeast to a hyphal growth form C. albicans is able to evade immune cells and to penetrate into host niches less accessible to the immune system.
Survival of fungal pathogens in the human host requires that their cell surfaces are intact. Defects in the cell wall of C. albicans that occur under immune attack or by treatment with antifungals are sensed and activate compensatory activities [6]. Reduced glucan content leads to the activation of the protein kinase C (PKC) pathway that includes the Mkc1 MAPK module, which activates the glucan synthase activity and stimulates the transcription of genes involved in glucan and chitin biosynthesis [7], [8]. In addition, defective N- or O-glycosylation activates the Cek1 MAPK module and recent results indicate that PMT genes encoding protein-O-mannosyltransferases are downstream regulatory targets [9], [10]. Sensing through this pathway is accomplished by the Msb2 and Sho1 cytoplasmic membrane proteins, which signal via the Cdc42 GTPase to Cek1. Intact N-glycosylation is detected by Msb2 and represses PMT1 transcription, while defective N-glycosylation induces Cek1 phosphorylation and de-represses PMT1 transcription [9], [10]. In a different mode of regulation, defective Pmt1-type O-glycosylation is sensed by Msb2, activates Cek1 and induces PMT2 and PMT4 expression. Induction of PMT2/PMT4 genes by inhibition of Pmt1 and damage of β1,3-glucan also requires Msb2 and Cek1 suggesting that cell wall damage is reported to Cek1 via Msb2 [10]. This function of Msb2 is supported by its associated partner membrane protein Sho1 [9]. Defects in either Mkc1 or Cek1 pathways lead to defective hypha formation on some semi-solid media, supersensitivity against antifungals and other stressors and reduce the virulence of C. albicans [9], [11], [12].
Msb2 is a type I membrane protein containing a single transmembrane region that separates a large extracellular from a small cytoplasmic domain; this structure is conserved in several fungal species [13]–[16]. Msb2 in the yeast Saccharomyces cerevisiae has been shown to be continuously cleaved by the Yps1 yapsin protease, releasing the extracellular domain into the growth medium [17]. This property, coupled with the high level of N- and O-glycosylation of the extracellular domain has led to the concept that fungal Msb2 proteins represent functional analogs of the mammalian MUC1/2 signaling mucins, which by proteolytic cleavage generate highly hydrated mucous glycoprotein layers around cells and at the same time confer transcriptional regulation by the cleaved cytoplasmic domain [18]. In fungi, intertwining of Msb2 hydrated glycostructures with cell wall components may be related to the sensing function of Msb2. Cleavage of the ScMsb2 cytoplasmic domain has not been reported and its presence may be required for Cdc42 binding, which is an essential upstream element of the Kss1 MAPK pathway [13]. Here we report that the glycosylated extracellular domain of C. albicans Msb2 is released into the growth medium in considerable amounts and we show that the shed protein has the function to protect against AMPs produced by the host. In humans, the most prominent AMPs exhibiting strong antimicrobial and immunostimulatory activities are the histatins, which are produced by salivary glands and secreted into saliva and the cathelicidins and defensins, which are produced by neutrophils and macrophages (for a review, see [19]–[21]). The human cathelicidin LL-37 occurs on mucosal surfaces at a concentration of 2–5 µg/ml but its concentration rises to 1.5 mg/ml in acute inflammation [22]. Histatin-5 and LL-37 are cationic AMPs that damage the cytoplasmic membranes of C. albicans [23]–[25] and histatin-5 also attacks intracellular targets [26]. The combined findings of this study suggest that shed Msb2 is a glycoprotein that effectively protects C. albicans against killing by AMPs LL-37 and histatin-5, allowing C. albicans to evade immune reactions and to allow its persistence as a commensal.
To immunologically detect Msb2 we constructed a strain producing a variant Msb2 protein carrying an HA-epitope within the large extracellular domain and in addition a V5-epitope in the middle of the short cytoplasmic domain (Figure 1 A). MSB2 was expressed either under the control of the constitutive ACT1 promoter when plasmid pES11a was integrated in the LEU2 locus (strain ESCa3) or by the authentic MSB2 promoter when pES11a was integrated in the partially deleted msb2Δ1 allele of strain FCCa28 (strain ESCa10). The msb2Δ1 allele encoding 406 N-terminal residues of Msb2 was found to be completely non-functional in all phenotypic assays (see below) and it was fully complemented in transformants containing pES11a integrated in both genomic loci; complementation efficiencies were equal between transformants carrying singly HA-tagged or doubly HA-V5-tagged Msb2 versions. Thus, while several msb2Δ1 mutant strains were as supersensitive to caspofungin and tunicamycin as the pmt4 control strain [10] complementation by the epitope-tagged versions of Msb2 restored normal resistance (Figure 1 B). While tunicamycin-supersensitivity indicates that msb2Δ1 mutants require intact N-glycosylation for growth, O-mannosylation by Pmt1 appears not relevant since mutants grew normally in the presence of the Pmt1 inhibitor. The tagged versions of Msb2 were also fully active to reverse the hyphal growth defects of the msb2Δ1 mutants [9] (Figure 1 C).
In addition, we constructed plasmid pES11c, which encodes the HA-tagged Msb2 variant carrying the V5 epitope at its C-terminal end (allele MSB2HA-V5 end). The phenotypic results for pES11a- and pES11c-transformants were identical (data not shown).
Release of a Msb2 subfragment into the growth medium has been observed in S. cerevisiae and other fungi [13]–[16]. When we examined cells and growth medium of C. albicans transformants producing tagged Msb2 by immunoblotting we discovered that the majority of HA-carrying Msb2 was present in the medium and migrated as a diffuse band of >460 kDa (Figure 2 A). No significant difference regarding the amount of immunoreactive protein was detected in strains either transcribing MSB2 from the ACT1 or MSB2 promoters (compare lanes 3 and 5) suggesting that both promoters are of comparable strength. As expected, the tagged ER-membrane protein Pmt1HA was associated only with cells (lane 2). In contrast to HA immunodetection the V5-tagged Msb2 protein was found exclusively in association with cells and not in the medium, similar to the Pmt2V5 control protein (Figure 2 B). The V5-tagged Msb2 protein migrated as a doublet of about 15 kDa and thus corresponded in size to the cytoplasmic domain of Msb2. Thus, it appears that during growth in liquid culture the Msb2 full-length protein is mostly cleaved proteolytically into the large extracellular (HA-tagged) and the small cytoplasmic (V5-tagged) subfragments. Importantly, release of the Msb2HA fragment was almost quantitative during growth in complex YPD growth medium and was not altered significantly in YEPG medium containing galactose as in S. cerevisiae [17] or during hypha formation in YP medium containing 10% serum (data not shown). The released extracellular fragment or Msb2 will now be referred to as Msb2*.
To examine if Msb2* secretion would also occur during growth on a semisolid agar surface we used a double sandwich system consisting of a PVDF membrane used for immunoblotting topped by a membrane filter precluding the passage of cells, which were both placed on YPD agar (Figure 2 C, a). Cells grew on the membrane filter (Figure 2 C, b) and immunoanalysis of the PVDF filter detected HA-proteins only released from cells producing Msb2* (Figure 2 C, c 3, 4) but not from cell producing tagged Pmt1HA protein. This result indicates that the extracellular Msb2 fragment is also detected in surface growth of C. albicans.
Considering the possibility that Msb2 is cleaved immediately upstream of the transmembrane region it was expected that Msb2* has an approximate molecular mass of 131 kDa but the heterogeneity and apparent molecular mass in immunoblotting (Figure 2 A) suggested extensive glycosylation. To estimate its molecular mass more accurately we carried out fractionation of culture fluid containing Msb2* by gel filtration, using a column previously calibrated with standard proteins (Figure 2 D, a, b). Fractions eluted from the column were examined by immunodetection and yielded a major peak from 468–614 kDa (Figure 2 D, c) in agreement with the above immunoblotting results. A minor peak in the void volume, presumably representing aggregated Msb2* of >1000 kDa, was also detected. Since this result suggested that glycosylation contributed equally to the mass of Msb2* as its protein content we attempted to clarify the type of protein glycosylation. Extensive treatment of the growth medium (and of purified Msb2*, see below) with PNGase F did not result in a significant alteration of its apparent molecular mass (data not shown), while β-elimination led to a mass reduction to about 300 kDa (Figure 2 E, a) indicating that Msb2* is significantly O-but not N-glycosylated. On the other hand, complete chemical deglycosylation by trifluoromethanesulfonic acid (TFMS) reduced the mass of Msb2* to about 117–130 kDa (Figure 2 E, b) consistent with the proteolytic cleavage of the Msb2 precursor protein immediately upstream of the transmembrane region (expected molecular mass of unmodified 1291 residue fragment is 130 kDa). It is yet unclear if the different deglycosylation results obtained for β-elimination and TFMS treatments is due to residual O-glycosylation not removable by β-elimination, by residual N-glycosylation, which is not removed by PNGase F or by yet unknown modifications. However, because clear evidence for O-glycosylation of secreted Msb2 was obtained we produced epitope-tagged Msb2 in C. albicans mutants lacking each of the 5 isoforms of protein-O-mannosyltransferases. Immunoanalysis of secreted Msb2* showed faster electrophoretic mobility in the pmt1 mutant, while in the pmt4, pmt5 and pmt6 homozygous mutants no difference to the control strain was detected (Figure 2 E, c). We conclude that Pmt1 is at least partially involved in Msb2 O-glycosylation, although the contribution of Pmt2 (only testable in a heterozygous PMT2/pmt2 strain since it is essential for growth [27]) cannot be excluded. Compensatory upregulation of other Pmt isoforms in a pmt1 mutant [10], [28] may also account for remaining Msb2 O-glycosylation, which showed a very broad mobility distribution corresponding to apparent molecular masses from 240–480 kDa.
It has been reported that in S. cerevisiae the yapsin-type protease Yps1 is responsible for cleavage and secretion of Msb2 [17]. In C. albicans the closest homolog to Yps1 is Sap9 (21.9% identity), while Sap10 is also structurally similar because it is GPI-anchored in the cytoplasmic membrane [29]. When we expressed the tagged MSB2HA-V5 allele in the sap9 mutant (ESCa33), the sap10 mutant (ESCa34) or the sap9 sap10 double mutant (ESCa35) we did not observe any difference in amounts and molecular masses of Msb2* (data not shown). We also observed normal secretion of Msb2 in a mutant (ESCa36) lacking the furin-type and Golgi-resident Kex2 serine endoproteinase, which in S. cerevisiae is required for cleavage and shedding of the Flo11 protein [30]. Furthermore, we repeatedly added high concentrations (15 µg/ml) of the aspartyl protease inhibitor pepstatin, of the metalloprotease inhibitor amastatin (15 µg/ml) or of a commercial mix of inhibitors for serine- and cysteine proteases (complete mini tablets; Roche) to growing cultures of ESCa3 but we did not find any effect on Msb2* release (data not shown). We conclude that the processing mechanism of Msb2 in C. albicans requires an as yet unidentified protease and that Sap9, Sap10 and Kex2 proteases are not involved.
We constructed several C. albicans strains producing deleted Msb2 variants under the control of the ACT1 promoter in a msb2 mutant background and tested Msb2-dependent phenotypes including secretion of Msb2, hypha formation and resistance to caspofungin; furthermore, the ability of variants to activate the Cek1 MAP kinase was examined. The results are summarized in Figure 3 A and presented in Figure 3 B and Figure S1.
Two major deletion variants either lacking 449 residues of the extracellular domain (Msb2-ΔN) or lacking the complete cytoplasmic tail of 103 residues (Msb2-ΔC) were fully able to complement all msb2 mutant phenotypes. In contrast, strains only producing the N-terminal region of Msb2 up to the transmembrane region (variant Msb2-ΔTM-C) or solely the 108 cytoplasmic variant Msb2 tail residues were as defective for Msb2 phenotypes as mutants REP18 carrying a complete deletion of the MSB2 ORF or strain FCCa27 only producing N-terminal residues 1–406 of Msb2 (Msb2-Δ1). Inactivity of the Msb2-ΔTM-C variant was not caused by lack of protein biosynthesis since amounts of Msb2* released into the medium were comparable for all HA-tagged variants (Figure S1). However, with regard to the activation of Cek1 a particular phenotype of these deletion variants was observed. The wild-type strain ESCa3 showed low levels of phosphorylation in stationary phase and phosphorylation was increased during logarithmic growth, which was stimulated further in the presence of tunicamycin [9] (Figure 3 B). In contrast, strain ESCa25 producing the Msb2-ΔN variant activated Cek1 not only in stationary phase but also in the absence of tunicamycin to high levels. In addition, strain ESCa38 carrying the Msb2-ΔC variant was impaired in its ability to activate Cek1 in response to tunicamycin. Strains producing the Msb2-ΔTM-C and the Msb2-tail were completely unable to activate Cek1 phosphorylation. Thus, it appears that the Msb2 N-terminal, transmembrane and cytoplasmic domains region convey different functions in Cek1 phosphorylation.
C. albicans ESCa3 expressing ACT1p-MSB2HA-V5 released considerable amounts of the Msb2* glycoprotein into the complex YPD growth medium, amounting to 76 µg/ml and 150 µg/ml in logarithmic growth (OD600 = 1) and in stationary phase (OD600 = 6). Msb2* was quantitated immunologically by a dot-blot procedure, because its high glycosylation status prevented quantitation by standard methods. We considered that this glycoprotein could contribute to defense against immunological responses of the human host, in particular to the attack by AMPs [20]. To verify this concept we first tested if the presence of Msb2 would contribute to basal levels of AMP resistance of C. albicans. Wild-type strains were significantly more LL-37-resistant than msb2 mutants (Figure 4 A). Sensitivity of a msb2 sho1 double mutant was only slightly increased compared to a msb2Δ1 single mutant and a sho1 single mutant showed wild-type resistance indicating that Msb2 but not Sho1 mediates LL-37 resistance. The increased LL-37 sensitivity of msb2 mutant strains versus a wild-type strain was also correlated with increased fluorescent staining of mutant cells [26], [27] by TAMRA-labelled LL-37 (Figure 4 B). We also observed that in the presence of LL-37 the msb2 mutant tended to aggregate more readily than wild-type cells [31].
We next tested the LL-37 sensitivity of the above series of transformants producing truncated Msb2 variants. Interestingly, while the transformant only synthesizing the C-terminal tail of Msb2 was as sensitive as the msb2Δ1 mutant all other transformants showed wild-type sensitivity (Figure 4 C). Even the transformant producing Msb2 deleted for its transmembrane region and C-tail was not supersensitive, although as described above this Msb2 variant was inactive in complementing msb2 mutant phenotypes (Figure 3). It was concluded that the basal resistance of C. albicans to LL-37 depended on the secreted extracellular domain of Msb2 but its N-terminal domain was not required for this action. Since full-length and N-terminally deleted Msb2* are O-glycosylated to a large part by Pmt1 (Figure 2) transformants were constructed producing doubly tagged Msb2 in C. albicans strains defective in each of the 5 Pmt proteins (a heterozygous strain was used in case of PMT2 because of its essentiality for growth). Among these transformants only the pmt1 mutant was LL-37 supersensitive supporting the notion that Pmt1-directed O-glycosylation of Msb2* is required to provide resistance to LL-37.
In conclusion, these results suggest that the secreted extracellular Msb2* domain is required for LL-37 basal resistance of C. albicans.
Several mechanisms are possible to explain the requirements of Msb2 (and Sho1) for LL-37 resistance and one mechanism is inactivation of LL-37 by the secreted Msb2*. To verify this concept we first purified Msb2* fragment from the growth medium by affinity chromatography using anti-HA antibody and verified that the purified material consisted solely of the heterogeneous >460 kDa protein by silver staining and immunoblotting (Figure 5 A). Next we asked if the purified Msb2* would proteolytically attack cathelicidin LL-37. Msb2* and AMPs were co-incubated and then assayed AMPs on a 18% SDS-PAGE gel (which excludes Msb2*). Msb2* co-incubation did not diminish amounts of LL-37 and no degradation products were observed (Figure 5 B) even if a 22.5% SDS-PAGE gel was used (data not shown). Furthermore, long term incubations (16 h) of Msb2* preparations with substrates of a protease detection kit able to detect a wide variety of protease did not detect any protease activity (data not shown). Therefore, it was concluded that Msb2* preparations had no general proteolytic activity. In additional pre-tests we bound Msb2* (or Msb2-ΔN*) to wells of microtiter dishes and checked if TAMRA-labelled LL-37 would absorb to these wells. Msb2* coating did indeed stimulate binding of LL-37-TAMRA significantly, while preincubation with unlabelled LL-37 reduced subsequent binding (Figure 5 C). This result indicates that LL-37 has a specific binding site on Msb2*.
To test a potential function of Msb2* in AMP protection we set up an AMP activity assay, in which we treated C. albicans for 1.5 h with AMPs in the absence or presence of purified Msb2* and then assessed fungal viability by determination of colony-forming units (CFU). The results show that added Msb2* rescued C. albicans from LL-37 killing, which was obvious for the wild-type strain and even more significant for msb2 and msb2 sho1 mutants; even an E.coli strain was protected against LL-37 by Msb2* (Figure 5 D). Interestingly, even the shortened Msb2*-ΔN fragment secreted and purified from strain ESCa25 was able to provide protection, although a concentration dependence of its activity revealed that it is slightly less active in AMP inactivation compared to the full-length Msb2* protein (Figure 5 E). AMP inactivating activity was also detected by merely using medium (secretome) of a C. albicans wild-type strain (CAF2-1) for co-incubation with LL-37 (Figure 5 F). As expected, medium of the msb2Δ1 strain (FCCa27) had no protective effect, while medium of the pmt1 mutant (SPCa2) had reduced inactivating activity.
These findings demonstrate that the extracellular Msb2 domain has an additional function in C. albicans biology, e. g. in LL-37 defense, which is different from its roles in cell wall integrity and filamentation.
C. albicans is known to be sensitive to low levels of histatin-5 [23]–[26], [32], [33]. We considered the possibility that higher Msb2* levels occurring in the vicinity of C. albicans colonies in the human host could protect against histatin-5 as we had found for LL-37. Although we did not observe a significant higher sensitivity to histatin-5 in msb2 mutants (as for LL-37) we found that added purified Msb2* did indeed protect C. albicans strains significantly against histatin-5 (Figure 6). As expected, HA peptide used for elution of Msb2* from the anti-HA antibody in affinity chromatography did not provide protection. The protective action of Msb2* was not restricted to C. albicans because even an E. coli strain was rescued from histatin-5 killing (Figure 6). Thus, we conclude that protection by the secreted Msb2 glycofragment is not specific for LL-37 but extends to other AMPs including histatin-5 and affects microorganisms other than C. albicans.
A complex interplay of responses and counter-responses characterizes the encounter of microbial pathogens with the human host. Opportunistic pathogens including C. albicans may be commensals, held in check by the immune system and supported by actions of the pathogen that favour a commensal life-style [1], [34]. Conversely, immunological impairment or other conditions can favour propagation of pathogens and result in disease through microbial virulence traits and/or immune hyperstimulation causing autoimmune damage [35] Immune cells detect surface structures of C. albicans including glucan and mannoproteins and trigger IL-17-dependent reactions [2], [3] including the production of AMPs, which kill the pathogen and attract immune cells [19], [20]. The C. albicans protein Msb2 has a dual function to stabilize the fungal cell wall and we show here that it is also required to block an important aspect of the immune response by inactivating AMPs (Figure 7).
Fungal pathogens have a relatively high ability to resist attack by hydrolytic enzymes or small toxic molecules including antifungals in the human host. Cell wall damage is restored or compensated for by signaling pathways that sense the defect and initiate appropriate rescue responses [6]. In C. albicans defects in glucan or chitin are sensed especially by pathways containing the Mkc1 or Hog1 MAP kinases that trigger enhanced glucan or chitin biosynthesis [7], [36]. Defects in protein glycosylation are transmitted mainly via the Cek1 MAP kinase pathway and lead to activation of individual isoforms of protein-O-mannosyltransferases [9], [10]. Blockage of N-glycosylation by tunicamycin depends on Cek1 and upregulates PMT1 transcription, while inhibition of Pmt1-O-glycosylation stimulates transcription of PMT2 and PMT4 genes. Interestingly, we found that the Msb2 membrane sensor protein functioning at the head of the Cek1 pathway is itself a highly glycosylated protein as in other fungal species. Despite the presence of 5 potential acceptor sites no evidence for N-glycosylation of Msb2 was obtained but the secreted Msb2 migrated faster in a pmt1 mutant (not in other homozygous pmt mutants) indicating that Pmt1 is partially responsible for Msb2 O-mannosylation. Residual O-chains in a pmt1 strain were removed by chemical treatment suggesting that they are contributed by the Pmt2 isoform, which is essential for growth [27]. Lack of Pmt1 glycosylation was previously shown to increase phosphorylation of Cek1 and to activate PMT2/4 transcription [9], [10] and we add here that lack of the N-terminal Msb2 glycodomain leads to constitutive Cek1 phosphorylation. Conceptually, lack of Msb2 O-glycosylation could trigger Cek1 phosphorylation but other O-glycosylated proteins interacting with Msb2 could also provide the triggering signal. Signaling by proteins interacting with Msb2 is suggested by the finding that tunicamycin-treatment induces Cek1 phosphorylation, although Msb2 does not appear to be N-glycosylated itself. In S. cerevisiae, however, Msb2 is N-glycosylated and O-mannosylated by the Pmt1, 2 and 4 isoforms; furthermore, activation of the Cek1 homolog Kss1 occurred only in cells lacking Pmt4 and inhibited for N-glycosylation by tunicamycin [37], [38]. Thus, Msb2 glycosylation and resulting MAP kinase activation proceed differently in C. albicans and S. cerevisiae.
The single transmembrane region of Msb2 divides the protein in a large glycosylated extracellular and a small cytoplasmic domain in C. albicans, S. cerevisiae and other fungi. A S. cerevisiae Msb2-GFP fusion has been shown to get efficiently cleaved leading to release of the extracellular domain into the medium [17]. This processing occurs at a yet undefined site and requires the Yps1 yapsin-type protease suggesting that it is directly or indirectly involved in the cleavage. Similarly, using doubly epitope-tagged Msb2 we found that in C. albicans Msb2 is cleaved almost quantitatively, which sheds the extracellular domain into the medium and retains the cytoplasmic domain in the cells. However, in C. albicans the closest homologs of ScYps1, Sap9, Sap10 [29], and serine endoproteinase Kex2 [30] were not required for CaMsb2 processing. Cleavage/release was found to occur both in liquid and on surfaces and the amount of secreted Msb2 depended on the number of growing C. albicans cells. Thus, importantly, the level of released Msb2 is a measure of C. albicans propagation. In agreement, Msb2 peptides were recently identified in the secretome of C. albicans yeast and hyphal cultures; peptides corresponded to the extracellular domain including residue 1290 upstream of the transmembrane region [39].
The relationship between Msb2 structure, processing/secretion and Cek1 phosphorylation was studied using C. albicans strains producing Msb2 variants. A large deletion of 450 N-terminal residues adjacent to the signal sequence (Msb2-ΔN) led to functional Msb2 able to complement defects of the msb2 mutant; this variant differed from the native protein, however, in that the Cek1 MAP kinase was constitutively phosphorylated. In agreement, S. cerevisiae Msb2 deletions of the extracellular domain have been found to hyperactivate the dedicated MAP kinase Kss1 [17]. Different phenotypes were obtained for C-terminal deletions of C. albicans Msb2. While a Msb2 variant deleted for its C-terminal end and the transmembrane region (Msb2-ΔTM-C) was completely inactive, a deletion retaining the transmembrane region (Msb2-ΔC) was fully functional in complementing msb2 phenotypes. Unexpectedly, however, the latter variant did not respond to tunicamycin-treatment by induction of Cek1 phosphorylation, in agreement with results obtained for a similar S. cerevisiae Msb2 variant [38]. We conclude that the transmembrane region of Msb2 is absolutely required for Msb2 functions and furthermore, that tunicamycin-regulated signaling to the Cek1 MAP kinase requires the cytoplasmic domain. Conceivably, the cytoplasmic domain could be directly involved in regulation of Cek1 kinase activity or it could participate in gene regulation as has been reported for signaling mucins and the Notch protein in higher eukaryotes [18], .
In the human host C. albicans contacts surfaces of body cells including immune cells, which may phagocytose the pathogen and elicit a wave of antifungal activities. Resident or induced soluble defense molecules such as immunoglobulins, complement factors and AMPs kill or block the growth of the pathogen. AMPs have a wide range of antiviral, antibacterial and antifungal activities and provide an antimicrobial barrier on mucosal surfaces such as histatins produced and secreted by salivary glands or they are components of the antimicrobial armory of neutrophils that produce cathelicidins (LL-37) and defensins [20]. Furthermore, AMPs act as chemoattractants recruiting leukocytes to sites of infection [19], [21]. C. albicans is known to be sensitive to histatins, LL-37 and defensins, which inhibit fungal growth by cytoplasmic membrane disruption, interference with mitochondrial activity or yet undefined mechanisms [23]–[26]. Furthermore, binding of LL-37 or histatins to cell wall carbohydrates prevents adhesion of C. albicans to host cells and plastic surfaces [31]. It should be noted also that bacterially-produced AMPs such as the lantibiotic nisin secreted by Lactobacillus lactis contribute to the diversity and high concentration of AMPs in the human body [41]. Nevertheless, a myriad of microbial commensals including some opportunistic pathogens persist as cohabitants because they are at least partially AMP-resistant. Several AMP-resistance mechanisms have been reported. Cleavage of AMPs by soluble or membrane-bound proteases has been described for many bacterial species and it has been shown that C. albicans is also able to cleave histatin-5 by the yapsin-type protease Sap9 [42], [43]. Another evasion mechanism known in bacteria is the secretion of AMP-binding proteins that act as decoys deflecting AMPs from their dedicated action at microbial cell surfaces. Examples include the secreted SIC, staphylokinase and FAF proteins by Streptococcus pyogenes, Staphylococcus aureus and the commensal Finegoldia magna, respectively [44]–[46]. Here we describe that an analogous mechanism is relevant also for fungal pathogens since shedding of a large glycosylated fragment of the Msb2 sensor protein renders C. albicans AMP-resistant. Msb2 shedding reached high levels during liquid growth (about 150 µg/ml in stationary phase) and was also observed during surface growth. Purified Msb2 fragment effectively blocked the fungicidal activity of histatin-5 and LL-37 even at a >20 fold molar excess of AMPs suggesting multiple binding sites. Interestingly, a C. albicans msb2 mutant was supersensitive to LL-37 but not to histatin-5 suggesting that the relatively small amount of cell-associated Msb2 suffices to protect against LL-37 but not against histatin-5. This finding agrees with the recent finding that LL-37 but not histatin-5 binds to C. albicans cell-wall carbohydrates [31]. The underlying molecular mechanisms for AMP binding to Msb2* remain to be determined. We found that the Pmt1-type of O-mannosylation is partially required for Msb2 glycosylation, its binding to LL-37 and for LL-37 resistance of wild-type cells, which raises the question if the glycostructures of Msb2* directly or indirectly affect LL-37 binding. Previous work has established the binding of LL-37 to various glycostructures including bacterial lipopolysaccaride [47], bacterial exopolysaccharides [48], human glycosaminoglycans [49] and fungal cell-wall polysaccharides [31]. These glycostructures may provide anionic contact sites for cationic AMPs such as LL-37 and histatin-5, which are enriched for basic amino acids (net charge +6 and, respectively, +12 at physiological pH). Since O-mannosyl side chains of Msb2* do not add net charge (unless they carry as yet undefined modifications) they do not allow ionic interactions with cationic AMPs, although non-ionic interactions cannot be excluded. Possibly, the functional role of O-mannosylation is indirect by providing an extended, bottle-brush conformation of the protein, as it is often observed in highly O-glycosylated protein domains [50]; this conformation could help to expose carboxylate side groups of aspartate and glutamate residues in Msb2* that could interact with basic residues of AMPs. Other C. albicans components including members of the Hog1 MAP kinase pathway are also involved in basal AMP resistance [51]; since Msb2 is not an upstream element in the Hog1 pathway of C. albicans [52] it probably regulates AMP resistance independently of Hog1. In a process that is analogous to functions of Msb2, the Pra1 protein of C. albicans is partially shed and impairs immune responses, in this case by binding of human factor H in solution leading to downregulation of the complement system in the vicinity of fungal cells [53].
We reported previously that in the standard mouse model of systemic infection (tail vein injection) no significant attenuation of virulence was detected for a msb2 mutant [9]. However, the systemic infection model may not appropriately reflect growth of C. albicans in the form of biofilms or foci of infection within organs, which are expected to be surrounded by a diffusion cloud of shed Msb2 at high levels that cause quorum resistance depending on fungal cell numbers. Shedding of Msb2 may also be important for C. albicans commensal growth, e. g. survival in the gut, where it is confronted with AMPs of other microbial commensals such as nisin produced by Lactobacillus [41]. On the other hand, shed Msb2 is able to provide cross-protection for other species as we have shown for protection of E. coli against LL-37 and histatin-5. Therefore, we propose that novel models for virulence and commensalism are needed to test the biological relevance of Msb2 and its shedding. Shed Msb2 may be of diagnostic value since its levels reflect fungal growth in the human host. Shed Msb2 is highly soluble and proteolytically stable because of its extensive glycosyl modifications and its presence in body fluids may be indicative of hidden localized fungal infections.
C. albicans strains are listed in Table 1. In C. albicans strain REP18 the MSB2 ORF of both alleles is completely removed [9]; this msb2 mutant allele is referred to as msb2Δ0. Strain FCCa27/28 contains partially deleted alleles designated msb2Δ1 (encoding the 406 N-terminal residues of Msb2), which were constructed using the URA-blaster method. A 3.8 kb genomic fragment encompassing MSB2 was PCR-amplified using primers IPF6003-NotI and IPF6003-SacII and cloned into pUK21 (NotI, SacII). The large BamHI-KpnI fragment of the resulting plasmid was ligated to the hisG-URA3-hisG blaster cassette of p5921 to generate pUK-6003.ko.Urab. The NotI-SacII disruption cassette of this plasmid was used according to the standard URA blaster protocol to partially delete both MSB2 alleles in C. albicans CAI4 generating FCCa27 (Ura+) and FCCa28 (Ura−). Strain FCCa28 allows integration of MSB2 expression vectors in the MSB2 locus by transformation with HpaI-cleaved plasmid and ectopically in LEU2 after digestion with EcoRV, which place MSB2 alleles under transcriptional control of the MSB2 and ACT1 promoter, respectively. The disruption was verified by colony PCR using primers IPF6003-3verif/ i-p2-Ura3ver and by Southern blottings (data not shown). E. coli strain DH5αF′ was used for plasmid constructions and for AMP protection experiments.
Strains were grown on/in standard YPD or SD media. Pmt1-inhibitor OGT2599 was resuspended in DMSO to prepare a stock solution of 10 mM [54]. Standard drop dilution tests (10 fold dilutions to 10−5) were used to determine sensitivity to inhibitors. Hyphal formation was induced by growth at 37°C on YPM medium containing 2% mannitol as sole carbon source or in liquid YP medium containing 10% serum [27].
Relevant restriction site used for the construction of MSB2 variant alleles are shown in Figure 1A. A MSB2 allele encoding heme agglutinin (HA)-tagged Msb2 was constructed by first PCR-amplifying the 5′-end of the MSB2 coding region using primers Msb2-ATG-XhoI and IPF6003-3′ (all oligonucleotides are listed in Table S1). The PCR fragment contained a novel XhoI site upstream of the ATG and extended to bp position 3227 of the ORF, 50 bp downstream of the PstI site. The XhoI-PstI subclone in pUC21 was mutagenized using the Quikchange kit (Stratagene) and primers HA-hin and HA-her were used to insert the sequence encoding a single HA epitope (11 amino acids) 1500 bp downstream of the ATG start codon sequence. The 3′-end of the MSB2 ORF was then amplified by genomic PCR using primers Msb2-int2 und Msb2-Stopp-XhoI-NotI, which generated a fragment containing a MSB2 sequence from 61 bp upstream of the PstI site to the XhoI site downstream of the stop codon sequence that was generated in the PCR reaction. This 3′ PCR fragment was mixed with the above 5′ XhoI-PstI fragment and the full-length modified MSB2 allele was generated by overlap PCR using the flanking primers Msb2-ATG-XhoI und Msb2-Stopp-XhoI-NotI. The resulting XhoI fragment was cloned downstream of the ACT1 promoter in C. albicans expression vector pDS1044-1 to generate plasmid pES10.
To insert the V5 epitope-encoding sequence into MSB2 a 1037-bp region from upstream of the PstI site to the middle of cytoplasmic domain sequence was PCR amplified using pES10 as template and primers PCR1 Hin und PCR1 Mitte Her, the latter primer added V5 sequences to the PCR product. In addition, a second PCR fragment (712 bp) was generated by PCR using primers PCR2 Mitte Hin (containing the V5 sequence) und PCR2 Her (downstream of the ApaI site in the 3′-UTR). Because both fragments contained the V5 sequence an overlap PCR using flanking primers PCR1 Hin und PCR2 Her generated a 1695 bp PCR fragment that was cut with NheI and ApaI and then inserted into pES10 to replace the corresponding unmodified fragment. The resulting expression plasmid encoding the MSB2HA-V5 allele was designated pES11a. In a similar approach, an expression vector encoding a Msb2 variant carrying the V5 epitope at the C-terminal end of Msb2 was constructed using primers PCR1 Hin, PCR1 Ende Her, PCR2 Ende Hin and PCR2 Her; the resulting plasmid was designated pES11c (MSB2HA-V5 end).
Expression vectors encoding Msb2 variants were constructed by primer-directed mutagenesis of plasmid pES11a, using the Quikchange kit (Stratagene). Plasmid pES14 encoding Msb2-ΔN lacking residues 33–481 of Msb2 was constructed using primers Cla1 Del1 next1/-2, plasmid ES16 encoding the Msb2-ΔC variant lacking the cytoplasmic tail of Msb2 was constructed using oligonucleotides MSB2 Stopp nach TM Hin/-Her and plasmid ES17 encoding the Msb2-ΔTM-C variant lacking transmembrane region and cytoplasmic tail was constructed using oligonucleotides MSB2 Stopp vor TM Hin/-Her. Plasmid ES15 encoding the Msb2-tail variant was constructed by PCR-amplification of sequences encoding the cytoplasmic tail by primers C-Tail vor/-rück and inserting it into downstream of the PCK1 promoter in plasmid pBI-1. Plasmids were integrated into the LEU2 locus of strain FCCa28 as described above.
Strains were grown in 50 ml YPD or SD medium at 30°C to OD600 = 6–10 and cells were harvested by centrifugation. Cells were washed with water and resuspended in lysis buffer (50 mM HEPES/pH 7.5; 150 mM NaCl; 5 mM EDTA; 1% Triton X-100) containing protease inhibitors (Complete, Mini, Roche). Cells were broken by shaking with glass beads at 4°C for 2×10 min on a vibrax (Janke & Kunkel, 2200 rpm) or with a FastPrep homogenizer (MP Biochemicals). Cell debris and glass beads were separated from the crude cell extract by centrifugation. For immunoblottings proteins were separated by SDS-PAGE (8%, 18% or 4–20% acrylamide) and blotted to PVDF membranes. Protein standards used were the PageRuler set (Fermentas; 11–170 kDa) or the HiMark set (Invitrogen; 31–460 kDa) of proteins. Membranes were probed using rat anti-HA monoclonal antibody (1∶2000; Roche) or mouse monoclonal anti-V5 antibody (1∶2000; Serotec) and visualized using peroxidase-coupled goat anti-rat or anti-mouse antibodies (1∶10000; Thermo) and the SuperSignal West Dura chemiluminescent substrate (Pierce).
Gel filtration chromatography was done on a Superdex 200 10/300 GL column (GE healthcare) equilibrated with SD medium. Elution characteristics were established using a set of standard proteins (Sigma) containing carboanhydrase (23 kDa), BSA (66 kDa), ADH (150 kDa), β-amylase (200 kDa), apoferritin (434 kDa) and thyroglobulin (669 kDa); the void volume (V0) was determined using Blue dextran (2000 kDa). Protein elution volumes (Ve) were monitored at 280 nm and fractions were collected by an ÄKTA prime plus (GE Healthcare) at a flow speed of 0.4 ml/min. To determine the molecular mass of secreted Msb2, strain ESCa3 (Msb2HA-V5) was grown in SD medium to OD600 = 10. Cells were removed by centrifugation and 500 µl of the medium was degassed, sterile-filtered and applied to the Superdex column. 200 µl fractions were collected and 20 µl per fraction were tested for the presence of Msb2HA by immunoblotting. The approximate molecular mass of Msb2HA was calculated from the standard protein graph using the equation y = 62258e−3,695x (x: Ve/Vo; y: molecular mass).
Deglycosylation reactions using PNGase F and α-mannosidase (jack bean) were carried out according to the instructions of the manufacturers (Roche; Sigma). To remove O-glycosylation the GlycoProfile β-elimination kit (Sigma) was used, either without or with pretreatment of the sample at 80°C. 200 µl of the ESCa3 growth medium was acetone-precipitated and resuspended in the same volume of water. 40 µl of the reagent mixture was added and the sample was incubated over night at 4°C. The sample was neutralized with HCl and 20 µl were analyzed by immunoblotting. The GlycoProfile IV kit (Sigma) was used to remove all forms of protein glycosylation by trifluoromethanesulfonic acid (TFMS). 1.5 ml of the growth medium of strain ESCa3 was lyophilized and 150 µl of TFMS was added and the proteins incubated at 4°C for 25 min. 4 µl of 0.2% bromophenol blue was added and neutralization by precooled pyridine (added drop-wise) was monitored by the yellowish coloring. This latter step was carried out in a bath of dry ice in ethanol. Reagents in the samples were removed by dialysis against PBS using Slide-A-Lyzer cassettes (Thermo).
The secreted Msb2HA domain was purified by affinity chromatography from cultures grown in SD medium containing 2% casamino acids to an OD600 = 10 using a column (1 ml) containing agarose beads covalently coupled to 3.5 mg of monoclonal anti-HA high affinity antibody (Roche). The column equilibrated with buffer (20 mM Tris/HCl, pH 7.5; 0.1 M NaCl; 0.1 mM EDTA) and 50–400 ml of the culture medium containing Msb2HA were loaded and the column was washed with 20 bed volumes of wash buffer (20 mM TrisHCl/pH 7.5; 0.1 M NaCl; 0.1 mM EDTA; 0,05% Tween 20). The Msb2HA protein was eluted twice by 1 ml (1 mg) of HA peptide (Roche) in Tris-buffered saline.
Proteins on SDS-PAGE gels were routinely visualized by Coomassie blue or silver staining and protein concentrations were determined by the Bradford assay using a commercial assay kit (BioRad). Because of the high glycosylation status of Msb2* its concentration could not be determined reliably by any of these methods. Therefore, we developed a dot blot procedure, in which known molar concentrations of HA peptide were compared to Msb2* (or Msb2-ΔN*) signals resulting from reaction with the anti-HA antibody. Dilutions of a HA peptide solution (Roche) were spotted on an activated PDVF membrane and a dilution series of the sample containing unknown amounts of Msb2* was spotted alongside. The membrane was processed as for immunoblottings and the resulting signals were recorded using a Fujifilm LAS400 mini image analyzer and evaluated with the Fujifilm Multi Gauge program. The standard curve derived from the HA peptide were used to calculate molar amounts of the Msb2* sample.
Msb2* samples were assayed for protease contamination using the Protease Detection Kit (Jena Bioscience) that detects a wide variety of proteases, including serine proteases, cysteine proteases and acid proteases. Substrate solution (50 µl) and incubation buffer (50 µl) were mixed with 100 µl (50 µg) of Msb2* in TBS and incubated at 37°C for 16 h. 120 µl precipitation reagent was added and samples were incubated at 37°C for 30 min. Tubes were centrifuged at 12.000× g for 5 min and 50 µl of the supernatant was transferred to a flat bottom 96 well plate, 150 µl assay buffer was added and absorbance at 492 nm was measured using a plate spectrophotometer (Biotek).
Strains were grown over night to stationary phase in YPD medium and diluted into YPD medium to an OD600 = 0.1. Cells were grown to OD600 = 0.8 at 37°C and incubated further for 1 h in the presence (+) or absence (−) of tunicamycin (2 µg/ml). Immunoblots were prepared as described previously verifying equal loading by Ponceau red staining of the membranes [9]. Blots were probed with anti-phospho-p44/42 MAP kinase (Cell Signaling Technology) to detect phosphorylated Cek1 protein and ScHog1 polyclonal antibody (Santa Cruz Biotechnology) was used to detect all forms of Hog1 [9].
Over night cultures of C. albicans and E. coli DH5αF′ were diluted and grown in YPD at 30°C to an OD600 = 0.3. Cells were harvested by centrifugation and washed with and resuspended in PBS. Triplicate assays containing 5 µl cell suspension and 0–10 µg LL-37 (Sigma) or histatin-5 (AnaSpec Inc.) in a total volume 25 µl were incubated 1.5 h at 37°C, diluted 500 fold and plated on YPD. Colony forming units were determined after 2 d of growth at 30°C. The action of LL-37 on cells was visualized by fluorescence microscopy using LL-37-TAMRA (Innovagen).
To assay binding of LL-37 to Msb2* a microtiter plate assay was used. 10 µg Msb2* or Msb2-ΔN* in 200 µl PBS were allowed to bind wells of a 96 well flat bottom polystyrene plate over night at 4°C. The wells were washed three times with PBST (PBS containing 0.05% Tween 20). Then 200 µl of blocking buffer (5% w/v nonfat dry milk in PBST) was added for 2 hours at room temperature. Wells were washed three times and incubated with 5 µg LL-37 5-TAMRA for one hour. After washing three times, the fluorescence was measured on a Tecan infinite 200 plate reader (excitation 560 nm, emission wavelength 590 nm). In a competition experiment, following Msb2* binding, 3 µg LL-37 was added to wells and incubated for one hour before cells were washed and LL-37-TAMRA was added.
|
10.1371/journal.pntd.0007317 | The gendered impact of Buruli ulcer on the household production of health and social support networks: Why decentralization favors women | Buruli ulcer [BU] is a chronic and debilitating neglected tropical skin disease caused by Mycobacterium ulcerans. The treatment of moderate to severe BU affects the well-being of entire households and places a strain on both gender relations within households and social relations with kin asked for various types of support. In this paper, we employ the conceptual lenses provided by the Household Production of Health approach to understanding the impact of illness on the household as a unit of analysis, gender studies, and social support related research to better understand BU health care decision making and the psychosocial experience of BU hospitalization.
An ethnography attentive to circumstance and the nested contexts within which stakeholders respond to BU was conducted employing semi-structured interviews, illness narratives, and case studies. An iterative process of data collection with preliminary analyses and reflection shaped subsequent interviews. Interviews were conducted with 45 women in households having a member afflicted with BU in two communes of Benin with high prevalence rates for BU. The first commune [ZE] has a well-established decentralized BU treatment program and a well-functioning referral network linked to the Allada reference hospital specializing in the care of BU and other chronic ulcers. The second commune [Ouinhi] is one of the last regions of the country to introduce a decentralized BU treatment program. A maximum variation purposeful sample was selected to identify information-rich health care decision cases for in-depth study.
Study results demonstrated that although men are the primary decision makers for healthcare decisions outside the home, women are largely responsible for arranging care for the afflicted in hospital in addition to managing their own households. A woman’s agency and ability to influence the decision-making process is largely based on whatever social support and substitute labor she can mobilize from her own network of kin relations. When support wanes, women are placed in a vulnerable position and often end up destitute. Decentralized BU treatment is preferred because it enables a woman to remain in her own household as a patient or caretaker of an ill family member while engaging in child care and petty revenue earing activities. Remaining in the hospital (a liminal space) as either patient or caretaker also renders a woman vulnerable to rumor and innuendo about sexual liaisons and constitutes a form of social risk. Social risk in some cases eclipses the physical risk of the disease in what we would describe as a hierarchy of risks.
This study illustrates the importance of decentralized treatment programs for NTDs such as BU. Such programs enable patients to remain in their homes while being treated, and do not displace women responsible for the welfare of the entire household. When women are displaced the well-being of the entire household is placed in jeopardy.
| In this gender-focused study of the neglected tropical disease Buruli ulcer (BU) in Benin, West Africa, we document how seeking care for BU is influenced by broad-based concerns about the household production of health and the availability of resources women can mobilize from their social support networks. Women and girls shoulder a disproportionate share of the burdens incurred by BU treatment and prefer decentralized treatment from local health stations to free hospital care. Long term and often-indeterminate residence in hospital threatens the integrity of households and results in marital stress, economic vulnerability, school and vocational training dropout, and loss of essential income-generating activities. The case study of BU clearly demonstrates the necessity of recognizing the household, and not just the patient, as a unit of analysis in public health and the need to consider the ripple effect of serious illness beyond the household to one’s social network. We draw attention to the fact that while men are the decision makers about health care in patrilineal Beninese society, a women’s agency in influencing decision making is tied to her accumulation of social capital, capital that is taxed by long term medical treatment weakening her safety net in the future.
| Much has been written about the health care seeking process in low and middle income countries (LMICs) and the predisposing, enabling, and service-related factors that contribute to health care decision making for different types of health conditions and diseases[1, 2]. Studies have also addressed how cultural perceptions and past interactions with practitioners affect present and future health care actions in a pluralistic health care arena. What has been underrated is how households cope with the direct, indirect, and opportunity costs of health care and the impact of illness on not just the afflicted, but other household members and the members of one’s broader social network[3–6]. A more complete understanding of health care decision making demands greater attention to the household production of health, gender relations, the mobilization of therapy management groups, and the ripple effect of illness on social support networks.
Adopting a household production of health (HHPH) approach to decision making[7, 8] situates health care within the full range of activities undertaken to achieve well-being for the household as a unit of analysis. Well-being extends beyond physical health to considerations of social relations, moral identity, and psychological health. Notably, this approach considers a household’s selective investment of time and limited resources, the trade-offs it makes when addressing pressing needs and real-world contingencies, and the opportunity costs of different courses of action.
Anthropologists have drawn an important distinction between the household as a structural and a functional unit[9]. An HHPH approach favors a functional, task oriented definition of the household that privileges the processual study of how health is produced, promoted, maintained, and protected by household members defined less by cohabitation (structural criteria) and more by routine participation (functional criteria) in health/well-being related activities. Households can include kin [and fictive kin] who are working or living elsewhere, but contribute to the household in some way, especially at times of urgent need, and who derive part of their identity by an affiliation to the household.
Social scientists studying the household as a unit of analysis are well aware that relations within households are both competitive and cooperative at different times, that the social status of members is not equal, that status changes over time according to varying criteria (e.g., age, work, financial contribution, marital status), and that intrahousehold negotiation between men and women over the use of resources takes place in subtle ways [10, 11]. Times of sickness in households with scarce resources are often occasions when tensions run high, especially when health care decisions implicitly or explicitly (dis) favor particular household members or courses of action. Decisions often take place in the context of ambiguity, do not reflect consensus, and are contingent.
Gender has been recognized as an important factor in studies of the HHPH, health care decision-making, and the allocation of scarce resources in times of sickness [12–14]. However, good case studies that illustrate different ways in which gender roles and relations within a household are affected by illness in LMICs, especially longstanding and chronic illness, are rare. Needed is research that examines differing demands treatment places on men and women during different points in an illness treatment trajectory. Given that women typically attend to the ill, special attention needs to be focused on economic, social, and affective challenges to women tasked with being caretakers for both children and the ill, and the ramifications of health care decisions.
A third dimension of health care decision-making addressed in the literature is social support. Of particular importance is the mobilization of therapy management groups (TMGs) from within one’s larger support networks. TMGs are the constellation of individuals who take charge of various aspects of therapy management with or on behalf of the afflicted[15,16]. They are composed of all members of one’s social support network having an impact on any aspect of health care decision-making, care seeking, and support. Members may include kin, friends, community health workers, health staff, and traditional healers. In short, the TMG is composed of everyone who weighs in or contributes to health care in some way. TMG address many “works of illness” from decision making and economic assistance to substitute labor and psychosocial support [17].
To date, few studies have addressed gender and temporal dimensions of TMG mobilization. Men and women have different social support networks and resources to draw upon. We know far too little about who each turns to, for what, and with what expectations. We also know little about how the composition of TMGs change over time and the degree to which levels of support are responsive to competing demands on member’s time, resources, and other social obligations. Missing in the therapy management literature is adequate consideration of the temporal dimension of TMGs mobilized to respond to longstanding and chronic disease. Also missing are studies that address reasons for TMG failure and patient abandonment.
In this paper, we employ the conceptual lenses provided by HHPH, gender studies, and social support related research to better understand health care decision making for Buruli ulcer (BU), a neglected tropical skin disease endemic in West Africa.
Buruli ulcer is a chronic, debilitating disease caused by Mycobacterium ulcerans[18]. It usually manifests through non-ulcerated lesions such as nodules, plaques, or edema that may evolve into massive skin ulcerations, joint and bone deterioration if left untreated[19]. Most cases of BU are found in West Africa and Benin is one of the endemic countries[20].
Fifty percent of those afflicted with BU are adults and 50% children. Most of those afflicted experience lesions on their limbs, although lesions may appear any place on the body[21]. The disease is non-contagious, the route of BU transmission unknown, and its incubation period poorly understood[22, 23]. The poorly understood transmission of the disease and the fact that scattered households, not clusters of households, are typically affected has reinforced local speculation about BU related wounds being possible signs of supernatural contact or witchcraft.
Up until 13 years ago, the management of BU required surgical removal of all sites of infection. In 2004, antibiotic treatment was found effective at early stages of the disease (category I: lesions < 5cm in diameter; and category II: lesions between 5 and 15cm in diameter)[19]. At present, the management of BU has three main components. Antibiotic treatment is based on daily oral rifampicin (10 mg/kg) and streptomycin (15 mg/Kg) injection for 56 days, which allows lesions whose diameter is less than 10 cm to heal without surgery[24, 25]. Effective outpatient antibiotic treatment at early stages reduces wound dressings and avoids skin grafts, which are needed for large ulcerations. More advanced cases often require long-term hospital treatment of indeterminate duration and physical therapy to prevent disability, amputation, and functional limitations after care[26]. Treatment for BU is provided free in most West African countries either at hospitals (centralized in-patient care) or at local health stations (decentralized outpatient treatment care).
Studies of BU in West Africa have found that biomedical treatment for BU is often delayed for reasons linked to perceptions of causality, fear of surgery and amputation, and the logistics and costs of seeking “free care,”[27]. With respect to cost, several studies [4, 28–30], have drawn attention to indirect and opportunity costs of “free medical care” to households. BU provides an excellent opportunity to address limitations in the health care seeking literature highlighted above. More specifically, it provides an opportunity to more closely examine both how households and social networks are affected by hospital based medical treatment of indefinite duration, and risks to female patients and patient caretakers.
The study took place in Benin West Africa. Benin is bordered by Togo to the west, Nigeria to the east, and Burkina Faso and Niger to the north. The country is highly dependent on subsistence farming, regional trade, cotton as a cash crop, and remittances from seasonal migrant work largely to Nigeria. Over twenty different sociocultural groups inhabit Benin, the vast majority of which are patrilineal, meaning that children are part of their father’s lineage. Women typically maintain close ties with their own female kin. While both men and women contribute to household economics, women are largely responsible for providing resources for routine household needs. Women generally do so through the cultivation and sale of agricultural products as well as petty trade. Microfinance schemes for women are available in some, but not all regions of Benin.
Benin is one of the most endemic countries for BU in West Africa [20]. Benin is divided into four regions and twelve departments subdivided into 77 communes. The National Control Program for BU in Benin supports four reference centers (CDTUB) located in Allada (Atlantic region), Lalo (Couffo region), Pobè (Ouémé region), and Zagnanando (Zou region). Each referral center supports a number of peripheral health centers that provide decentralized case management [31]. The mission of peripheral centers, which are state run health stations, is to provide accessible care for simple cases of BU (category 1 and 2). Reference centers, like the Catholic mission hospital of Zagnanado, are in charge severe cases. Field sites chosen for this study were located in two regions with high BU prevalence rates [31]: the Atlantic region (Zè commune) and Zou region (Ouinhi commune). Decentralized management of BU patients is well established in the Atlantic, Ouémé and Couffo regions of Benin. In these regions, most cases of mild to moderate BU (Category I and II) are treated at health stations staffed by nurses. More serious (category II and III) cases are referred to reference hospitals. Decentralized treatment of BU has only recently been introduced in the Zou region[31]. Up until 2016 when a pilot decentralization project was initiated in Ouinhi commune, BU patients in the region were served almost entirely by a Catholic mission hospital renowned for surgery-based treatment for all cases (category I, II, III) of BU [31]. Zou region has only begun the process of adopting decentralized BU treatment. In Ouinhi commune, only one of four health stations are presently treating BU cases. As noted in an earlier publication [31] this health station become very popular and is receiving patients who had previously refused to be treated in hospital.
A circumstantial ethnography[32] was conducted employing semi-structured interviews, illness narratives, and case studies. A circumstantial ethnography focuses on how nested sets of actors influenced by differing life circumstances respond to a focal phenomenon, in this case the treatment of BU. The ethnography was attentive to the experiences of patients and caretakers as well as responses of therapy management group members responding to requests for support. The study design allows for an iterative process of data collection with preliminary analyses and reflection shaping subsequent interviews. Case studies were collected using a “life history” approach, which focuses on the interviewee, and their storytelling to understand how perspectives and discourses are constructed [33]. In the present study, the focus was on the experiences of women deliberating and reflecting on BU treatment decisions, institutional care, household survival issues, and social support relationships. A narrative approach was chosen in which the focus is on people’s evaluations of their own life experiences [34].
A maximum variation purposeful sample[35] was selected to identify information-rich experiences for in-depth study. Interviews were conducted with 45 women who were either afflicted with BU themselves, caretakers for a family member with the disease, or the decision maker for whom in a household should accompany a patient to the hospital. One man who had uncharacteristically taken on the task of managing his son’s BU treatment was also interviewed. The sample included women whose husbands resided for most of the year in their homes and women whose husbands were migrant workers, married and widowed women, and women faced with managing moderately severe and more advanced cases of BU in the hospital (centralized treatment) and by daily visits to a health station (decentralized treatment). Informants were identified with the help of community health volunteers and clinic staff in community, clinic, and hospital settings in both Ze and Ouinhi communes. Hospital patients included both residents from the region in which the hospital was located, and patients and caretakers traveling to the hospital from outside the region. Once community health workers and health care providers identified people afflicted with BU, they were contacted and asked to participate in qualitative interviews about their illness experience.
The principles of thematic narrative analysis were followed [36]. After re-reading interview transcripts, the findings of interviews and narratives were discussed by team members, and coded for both focal and emergent themes. Focal themes included predisposing, enabling, and service related factors influencing BU treatment decisions, gender relations, social support, choice of patient-caretakers, and patient abandonment. Emergent themes include rumor and social risk, quality of childcare, and impact of treatment on children’s schooling. After an extensive consideration of the data obtained along, short vignettes and interview extracts in line with the study’s objectives were chosen as exemplars for use in this publication. Vignettes chosen illustrate the backstage of treatment decision-making and care management along with the complexities and contradictions revealed by a study of real-life circumstances[37]. Themes introduced in the results section provide answers to core research questions posed as a heuristic [38].
Ethical approval was obtained from Benin’s National Ethical Committee of Health Research before the start of the research (IRB00006860 N° 148 /MS/DC/SGM/DFRS/CNPERS/SA). Informed consent procedures already in place at Allada hospital were strictly adhered to over the course of the project. All patients and staff interviewed were assured that interviews would be kept confidential. The use of oral consent was approved by the ethical review board because many study participants were illiterate. When a participant was under 18 years of age, both the child/adolescent and his/her caretaker were informed about the nature and aim of study before being asked to give oral consent.
The results of our ethnographic study are presented as responses to seven key questions posed as a heuristic: 1) Who makes decisions about when and where to receive BU treatment; 2) What core HHPH concerns influence health care and patient caretaker decisions; 3) How do women influence BU related decision making; 4) What is the ripple effect of BU beyond the household; 5)What leads to abandonment during hospitalization; 6) Why are widows and their children a group at risk; and 7) Why do households, and particularly women, prefer decentralized treatment for BU. Data presented in response to the first six questions enable a nuanced response to the seventh question and provide valuable insights into community response to “free” BU treatment. Case vignettes are provided as a means to give a human face to core issues being highlighted and brief reference to Pan-African themes are noted to situate research results in a larger context.
In Sub-Saharan Africa, gender roles and social norms of seniority and power strongly influence how health care decisions are made [39–43]. In our research sites, most ethnic groups are patrilineal. Health care decisions that entail treatment outside of the home are made by husbands or senior members of their kin network. This is true even if a husband is employed as a migrant worker and absent from home much of the year. In our sample, 21 women acted as heads of their household during all or much of the year. Only two women reported making a BU related health care decision on their own. Women followed the health care advice of a husband or his kin regardless of whether they offered any financial support for BU treatment. Our informants noted that if a woman did not seek approval from her husband or senior members of his family, she left herself open to social censure. In some cases, however, a husband and his kin abandoned a sick child, an issue we will address shortly.
It has been widely reported in studies of health care seeking in West Africa that enabling factors are as important as predisposing factors (such as perceived cause) in determining when and what kind of health care is sought [6, 27]. Our study corroborated this finding. The enabling factors most commonly referenced in interviews about hospital-based BU care were the indirect costs of “free treatment” such as transportation costs, food and incidental costs (soap, mobile phone credit, etc.), and the opportunity cost of lost labor. Decision makers (husbands, elder kin) took stock of available sources of substitute labor within the household as well as a wife’s social capital, her ability to mobilize support from her own kinship network.
A mother’s absence from home on a daily basis to obtain outpatient treatment for BU or her need to remain at a hospital as either a patient or caretaker was deemed feasible only when essential household duties were taken on by somebody else. Women’s labor demands varied by season and household composition and encompassed agricultural labor, cooking, securing water and firewood, and childcare. Daughters were generally turned to first to take on a mother’s responsibilities in the household or to serve as a caretaker for a hospitalized family member. When a mother did not feel it was safe to leave small children at home to be cared for by an older child, or her labor in the fields was required for household survival, a daughter was commonly sent to care for a sibling in the hospital. This often interfered with her own schooling or apprenticeship activities. If, on the other hand, a mother was the patient, a daughter was sometimes asked to take charge of household duties in her absence. The following cases illustrate the complexity of patient caretaker deliberations as an important factor in health care decisions, and the role children play as patient caretakers given household production of health concerns.
Madeleine (daughter, patient), aged 11, was admitted to Allada hospital for treatment of BU after initially receiving decentralized treatment at a health station near her village. Her mother suffers from poor health, making it difficult for her to manage the household and tend to the fields. As a result, her husband took on a co-wife, who has three children of her own. Madeleine’s mother was afraid to accompany Madeleine to hospital and leave her other four children at home under her co-wife’s charge as she suspected they might be mistreated. Madeleine’s mother received assistance from her own mother and two sisters when she took Madeleine for decentralized care at a local health station a few kilometers away. However, when the child’s wounds did not heal, they were reluctant to offer long-term support for Madeleine if she was hospitalized. Madeleine’s father decided that the best option was to send Madeleine to the hospital along with her 8-year-old sister, Reine. Reine was taken out of school to care for Madeleine. Madeleine’s two older brothers were not asked to be a caretaker as this was seen as women’s work. Both Madeleine and Reine wished to continue their education, but their mother recognized that this was unlikely if long term BU treatment was required. In effect, Reine’s future was sacrificed to attend to her sister.
Clemency (mother of three, patient) needed to be hospitalized for an advanced case of BU, but she had no adult family member able to provide support. Her own mother was deceased and her two sisters were working in Nigeria. It was decided that Clemency’s teenage daughter would remain at home to tend to the household and that her younger, five-year old daughter would serve as her caretaker in the hospital. Her husband agreed to supply necessary resources during treatment. Clemency entered the hospital with her five-year old daughter and her 18-month-old son. Clemency required several surgeries and was confined to bed and a wheel chair. Her five-year-old daughter performed all tasks necessary for their survival in the hospital including going to the market, cooking, washing clothes, taking care of her baby brother, and making sure her mother took her medicine on time. Clemency’s daughter was helped by other caretakers and nurses in the hospital who spoke of her with great admiration. One often saw Clemency’s daughter going about her business with her younger brother on her back. Her mother described her daughter as a gift from God. However, she worried about her future, especially her schooling. She noted: “I do not know when I will finish with this treatment, no one tells me. If I can finish in a few months then my daughter will be able to go to school and can catch up. But, if I have to remain in the hospital longer, what will happen to her? While she is very intelligent, it will be hard for her to succeed in school.” As in the case of Reine, the future of a young patient caretaker was placed in jeopardy as an opportunity cost of treating a sibling afflicted with BU in hospital.
Two other household production of health issues emerged in BU illness narratives that are rarely discussed in the health care seeking literature. The first is a mother’s concern about the quality of childcare in her absence. This psychosocial concern sometimes eclipsed concerns about a child’s physical condition. The following case illustrates the importance of the quality of childcare in health care decision-making. The case involves a decision to decline free hospital treatment for a child afflicted with BU.
Prisca (mother, caretaker) is the sole resource-provider for her household. Her husband works, but most of the money he earns is spent on sodabi palm wine. One of Prisca’s children, an 11-year-old daughter, suffers from advanced (category II) BU, which requires hospitalization and possibly surgery. At first, Prisca administered home treatment to her daughter. When her lesions grew in size, Prisca asked permission from her husband to seek outpatient treatment for her daughter from the district health center 4 KM away. This proved challenging as Prisca still had to find the means to support the household on a daily basis through petty trade. After two months of treatment at the health center, her daughter’s condition was still serious and health staff referred her to Allada hospital, where she could receive free treatment.
At first Prisca refused to take her daughter to the Allada hospital even though she was concerned about the size of her lesions. Health staff and a doctor from Allada visited Prisca and attempted to change her mind, but she did not agree, stating she had no one to look after her other young children. She did not feel secure leaving her children in the hands of her husband. She stated, “Seeking care at the district health center is possible because it does not prevent me from going about my business and ensuring the well-being of everyone. Leaving the house for who knows how long, that is simply not possible.”
A few days later, a social worker from the hospital returned to Prisca’s house and offered to look after her daughter while in the hospital if no family member could accompany her. Prisca spoke to her husband, who agreed to allow their daughter to go to the hospital as long as significant cost was not involved. After two days of treatment at the hospital, however, Prisca returned and took her daughter home. When interviewed as to why she did so, the mother stated that her heart would not allow her to leave her daughter in the hands of an unknown woman. She went on to note: “I prefer that my daughter continue with the bandaging at the district health center even if this is not the best treatment. Some infirmity may result, but it is better than the total destruction of all members of my house.” She then when on to state: “When it comes to sickness only a mother can comfort and care for a child properly. In the hands of someone my daughter does not know, she is likely to suffer. How can I have a quiet heart at home worrying about her?”
A second notable concern that we identified as having a big effect on health care seeking and patient caretaker decision making was social risk (risk to reputation and to present and future social relationships) [44, 45]. When a woman leaves the confines of her village either to visit a health post some distance away or to reside in a hospital, she risks becoming the subject of rumors about sexual indiscretion. Such rumors question a woman’s moral identity and a husband’s masculinity and cause strife between husbands and wives. We found this to be a common reason a mother took a child with her when visiting a health post or when residing in a hospital. However, we found that even when a woman brought young children with her to hospital, she was still subject to rumor. Fear of rumor was a constant worry for some women, adding to the stress of social isolation and trying to survive with minimal resources. The following case illustrates how an apparently stable marriage was destroyed by rumor and innuendo:
Ruth (mother, caretaker), was given permission by her husband to care for their five-year-old daughter while she was being treated for BU in hospital. Ruth also brought her infant son to the hospital, as she was still breastfeeding. Ruth received regular visits from her husband, who was very attentive to her needs and those of their children. However, during one visit to the hospital he became quite agitated. Late in the evening, he awoke to see someone enter the ward, approach the bed of a young female patient, hold her hand and kiss her before departing. This event shocked her husband and he began to suspect his own wife’s fidelity. He began to see the hospital as a site of moral dissolution where patients and caretakers engaged in extramarital behavior. Without evidence of any wrongdoing on the part of his wife, he took the extraordinary measure of abandoning his wife and small children. When interviewed, he remained resolute, exclaiming, “These doctors, they may bring healing, but they destroy homes!”
Fear of rumor influenced who was chosen to be a patient caretaker in hospital. Daughters who had not yet reached puberty were preferred. The hospital was seen as a liminal space and time in the hospital to be quite boring. Several informants noted that “people” suspect that any young woman with limited resources will engage in sexual relations if outside the watchful eye of community members. We recorded cases where a daughter as young as 15 was sent to the hospital as a caretaker only to be returned home when rumors about sexual relations emerged. The following is an example:
Florent (son, patient) aged 18, was admitted to the Allada center for BU treatment. Florent is the third of seven children. His father lives and works as a brick maker in Nigeria with one of Florent’s brothers. Florent’s older sister, an apprentice seamstress, was asked to leave her apprenticeship to be his caretaker at hospital. Florent’s father suggested this course of action given that there were young children at home that needed their mother’s care.
During Florent’s hospitalization, his mother heard a rumor that Florent’s sister charged with his care was becoming romantically involved with men at the hospital. Fearing that her daughter’s reputation might be spoiled or that she might become pregnant, Florent’s mother sent her daughter back to her apprenticeship and replaced her as Florent’s care provider. This necessitated bringing four of her children with her: her two-year-old daughter and three children who had been attending elementary school.
Taking care of young children in the hospital wards is not easy for Florent’s mother. Her husband supports her, but sends money irregularly and what is sent is not enough to meet their needs. Because she can no longer work in the fields or engage in petty commerce in her village, Florent’s mother tries to make money any way she can while in the hospital by washing clothes, cleaning, and running errands for staff and other patients.
As has been noted elsewhere in sub-Saharan Africa[46], although women do not have the same kind of authority as men, it would be misleading to present them as having no impact on health care decision making. Most women we interviewed asserted that although men have the final say in decisions about health care, women’s input and counsel influence decisions. Women typically accepted their husband’s initial health care decision, even if they did not agree with it. However, they often encouraged husbands to reconsider decisions based on shifts in disease trajectory as well as the availability of different types of material and social support. And in a few cases, they took matters into their own hands when they felt they were being abandoned by a husband and his kin.
Also, as in other parts of Africa [6, 47], we found that women’s agency in health care decision-making was largely based on two things: the resources she has at hand, and her ability to mobilize resources from kin in the form of material goods, labor, and childcare. The best way a woman could influence BU-related health care decisions was by working out how a treatment option she favored could take place with only minimal disturbance to essential household production activities. Having a daughter, as noted in the cases of Madeleine and Clémency, was an asset. If one’s own daughters were old enough to serve as the caretaker of a sick family member, or to remain home and assume household responsibilities, a mother had some flexibility. However, when a woman did not have a daughter to assist her, she was compelled to approach kin and ask them for support and to play a more active role in therapy management.
Based on our data, most of a woman’s requests for assistance were to her own mother and sisters, followed by friends and neighbors. Asking members of her husband’s family for assistance was only a last resort. The following case illustrates kin coming to the aid of a sick relative wanting to be treated in hospital for BU and in need of a caretaker. In this instance, a niece was removed from vocational training to care for her aunt and as a result experienced biographical disruption, an interruption and destabilization of the life trajectory of the caretaker [48].
Gisèle (caretaker), aged 23, has been the caretaker for her aunt in Allada hospital for the last 19 months. Her aunt’s wounds form BU are quite serious and her treatment is likely to go on for some time. Prior to coming to the hospital, Gisèle was an apprentice seamstress attending a vocational training course. She planned to open up her own small tailoring shop soon after graduation. Gisèle was asked by her mother to take leave from her tailoring course to care for her aunt while in hospital. Her aunt had assisted their family in the past and she had no daughters of her own to ask for help.
When interviewed, the first thing Gisèle said was that she had never imagined how much her life would change when assuming a caretaker role in the hospital. She did not resent taking care of her aunt, but was sad about her fate stating that her “heart was in a vice.” She noted “I agreed to stay with my aunt because she is like a mother to me. She has always helped my family. But, by being here I have lost many things. I have no income-generating activities here. I have lost both financially and professionally. I was at the end of my apprenticeship and I was working to raise money necessary to obtain my diploma. My classmates have already graduated and they are now employed, but I am here. I worry about losing my tailoring skills, and I worry how I will raise money for my graduation. I try to find small jobs in the hospital, but whatever money I make is spent on food. My boyfriend has also become distant. He came here once and saw a male nurse teasing me, and he now suspects that I found a ‘doctor.’ I call him and he does not pick up the phone.”
As noted in the case of Gisèle, BU hospitalization does not just affect members of one’s immediate household; it also affects one’s broader social support network. In short, asking for and receiving support from kin in times of illness creates a ripple effect. For women living on the margin and having multiple work responsibilities of their own, assisting kin (and fictive kin) out of friendship or obligation is an effective means of reaffirming and strengthening reciprocal exchange relations. Volunteering to take children afflicted with BU to health stations for outpatient care, watching children when a mother is away from home, and lending money or supplying food were all found to be means of solidifying social bonds between women. This form of “bonding social capital” [49–51] provides women with a safety net associated with norms of social reciprocity and cooperation for mutual benefit.
On the other hand, we found that when requests for time or resources exceeded the capacity of kin to provide, social bonds were weakened. The same was true when a mother felt the amount of resources or care provided to her children by kin was inadequate. Requests for long term support often caused conflict within the households of kin. In some cases, there just were not enough material resources to share, and in other cases the opportunity costs of attending to somebody else’s children reduced the time a woman had available to generate revenue needed to support her own household. In short, social capital was a contingent and conditional resource dependent on the presence of resources [52].
Some women interviewed belonged to microfinance schemes and they had to repay loans in order to maintain the integrity of the group. The ripple effect of BU affected the entire group when members were unable to live up to their financial obligations due to the indirect and opportunity costs of BU treatment. The following case illustrates how BU affected one woman’s livelihood and microfinance group membership. Her predicament affected not just her present, but her chances of recuperating economically in the future.
Juliette [mother, caretaker] is a food vendor and a palm oil processor who is a member of a local micro-finance group. She contributes to the group monthly to pay off loans she has taken for her business. When her daughter, aged eight, was diagnosed with BU, Juliette took her to the hospital for treatment and resided with her for the next year. Remaining in the hospital disrupted her ability to pay back loans and this affected the entire microfinance group. Even though group members understood that she was caring for a sick child, they pressured her to find money. Her inability to repay loans compromised both the financial standing of the group and her future ability to borrow money. She noted:
“I have so much worry and stress now. What should I do to pay off my debts? The whole village knows that I owe money. I tried to arrange my business affairs from here. I entrusted my aunt with the sale of my goods in order to allow me to pay my debt each month. However, she mismanaged my business. What can I do now? … (she cries). I feel my reputation is now destroyed and I am resented. Women in the microfinance group will not welcome me back into the group. Without a loan, how can I reestablish my business?”
As noted by Ribera et al. in Cameroon [53], abandonment is an extreme household coping strategy initiated during catastrophic or protracted illness to avoid plunging a household into a “spiral of impoverishment.” Abandonment was a major concern voiced by our informants. Those residing in hospital as well as those contemplating going to hospital worried they might be abandoned if they remained in hospital beyond the length of time their household could provide for their basic needs. Wives under treatment worried that a husband might find it necessary to take a co-wife to maintain the house in her absence, and husbands afflicted with BU worried that wives might find other men to take care of them in their absence. We documented cases of both scenarios. Hospital administrators in Allada were especially concerned with child abandonment and noted cases where caretakers suddenly just disappeared. They pointed to several abandoned children now residing on the hospital grounds post treatment because they had no place to go. The presence of these children at the hospital was a constant reminder to others of what can happen when household resources are stretched too thin.
When interviewed, women who were abandoned displayed considerable psychological distress related to failed expectations of support in keeping with cultural values based on reciprocity. A common narrative emphasized how much a woman had sacrificed in the past to support other family members in times of need. The following are two examples:
Honon (mother, caretaker) is a widow with six children, the youngest of whom are twins. After the death of her husband, his family encouraged her to remarry one of his younger brothers, a proposal that Honon refused for undisclosed reasons. As a result, her husband’s family abandoned her and offered no support for her children. Honon was forced to return to her own family. She and three of her children went to live with her paternal uncle. The other three children were entrusted to other family members in a foster care arrangement (vidomègon) common in West Africa wherein children receive care in return for labor. One of her young twins developed BU. Honon’s mother, sisters, and uncle encouraged her to try various types of home remedies for the child. When the child’s wounds became more serious and required hospital treatment, her family members were unwilling to offer support either for Honon to care for the child while in hospital or to care for her other two children in her absence. Her kin felt that the burden of either action would place the household in jeopardy.
Honon was pressured to return the ill child to her deceased husband’s household. This suggestion was quite unsettling to Honon for two reasons. First, her deceased husband’s family had taken no responsibility for his children up to this point in time. She felt that if the child was received, they would be neglected. Second, she strongly suspected that someone in her deceased husband’s family had sent bad luck to the child resulting in wounds that would not heal.
Honon stated that she felt abandoned by both her own family and the household of her husband. Against the advice of her own family, she opted to go to Allada hospital and care for her sick daughter. She brought two of her other children along with her as there was no one willing to care for them at home. Honon received basic food rations from the hospital and otherwise survived by taking on small jobs when she could find them. She was very bitter about her abandonment. In her own words “Before leaving for Allada my mother promised to come visit me during our stay. It has been 18 months and neither she nor any of my sisters or brothers have visited or contacted me. When my sister was sick and hospitalized, I was the one who had been at her bedside. I was there for so many others in my family in their time of need. For me, no one is offering assistance or showing love. It is as if I am without parents. If it was not for the generosity of the hospital to whom I owe everything, I would be destitute.”
Conforte (female adult patient), aged 38, is Togolese and traveled to nearby Benin in search of treatment for an advance case of BU. Her older sister resides in Benin and informed her about Allada hospital. Conforte traveled to Allada and her sister provided her support during the first months of her treatment. However, as the months passed her sister’s resources dwindled and she began to tire of her sister’s illness. Then one day, Conforte noted with bitterness, her big sister stopped coming to visit and would not return her calls. Conforte never heard from her sister again and survived on charity offered to her by her church and hospital staff.
Conforte noted, “When I came here to seek treatment, I gave my big sister all my savings and belongings to hold for me while in the hospital. However, she abandoned me in the most difficult moments of my life. She is the one who asked me to come to Benin. I helped her so much when she faced illness in the past, she felt obliged to support me. It is true that she really helped me during the first months of my hospitalization, but over time, she regretted encouraging me to come here for treatment. She abandoned me. It is nasty, no? When our parents died, I was the one who helped all members of my family until I became ill. Now, where is their support for me in return? Today, strangers help me. In the past, I have helped others outside my family as a good Christian. Perhaps it was the help I gave to others, that led others to help me now.”
Widows are in a structurally vulnerable position in Beninese society whether or not they agree to a levirate marriage to a brother of a deceased husband who has other wives. In many cases, when a woman with young children is widowed or divorced, she raises her children in the house of her own kin until they are old enough to be sent to the household of their deceased husband. However, as we noted in the case of Honon, if a child becomes ill and is a burden to the household, a widow may be pressured to send the sick child to her deceased husband’s household to bear the costs and responsibility of treatment. If her deceased husband’s family does not offer support, and she opts to bring a sick child to the hospital she may be encouraged to place her other children in foster care as she will no longer be able to provide for them. The following case illustrates the predicament in which many widows find themselves.
Aline (mother, caretaker) is a former BU patient herself. When her husband died seven years ago due to an accident, she found herself having to care for their six children on her own. One of her daughters developed BU. When it became clear that she would have to be hospitalized, her deceased husband’s family remained silent about what should be done, and did not offer financial support for treatment. Aline’s mother’s household is very poor and was unable to offer her support. In order to admit her daughter to the hospital, Aline was compelled to place four of her children in foster care in the households of distant kin. This was an act of desperation. Aline was upset by the decision, but felt she had no other option.
Aline brought her ill daughter and a young son to Allada hospital. Because of their dire financial situation, the hospital offered Aline’s daughter basic daily food rations. To otherwise survive, Aline’s mother (like Honon) is constantly looking for work to support herself. Every day, she sells dumplings to schoolchildren at a nearby school, and gets a small payment in return from the dumpling-makers. The hospital is not happy having caretakers like Aline leave the hospital grounds to engage in petty business, but this is the only way she is able to survive.
Our research revealed a strong preference for decentralized BU treatment. Six reasons were identified from interviews that asked women about the advantages of decentralized care. First and foremost, decentralized care does not disrupt a household’s daily routine by removing a mother from her household. Decentralized treatment, for all but very severe cases of BU, still enables a woman to do chores and watch her children as well as engage in entrepreneurial activities essential for household survival. Second, decentralized treatment avoids the many indirect and opportunity costs associated with hospitalization. Third, it allows children to stay in school while being treated, and it reduces the need to remove children from school and apprenticeships to serve as patient caretakers. Fourth, a mother does not have to worry about the quality of sibling care and foster care in her absence. Fifth, she also does not have to worry about pernicious rumors undermining her own or a daughter’s reputation. Sixth, remaining at home during treatment is less stressful. A woman worries less about abandonment.
Fathers who take responsibility for the treatment of a child with BU also favored decentralized care. Although not the focus of our research, we encountered one father who, having refused to send his son to the hospital for BU treatment, agreed to take him for decentralized care. The case illustrates both why he favored decentralized care and why he accepted responsibility for taking his child to a health station. He acted in accord with kinship norms and obligations, and cared for a child from a co-wife he was not currently living with instead of asking a wife home maintaining his home to attend to the health needs of a child by a different marriage.
Djalil (father caretaker—outpatient) lives with his first wife and two children in a village in Ouinhi commune. His first wife was infertile and his two sons are children by a second wife, a Nigerian shopkeeper whom he cohabits with while working in Nigeria for some months each year. When his son aged 8 was diagnosed with BU by heath staff attached to the mission hospital he refused to send the boy to the hospital to be treated. He was afraid the boy would have to undergo surgery and remain in an unfamiliar environment for a long time. He was not comfortable asking his first wife to remain with the boy in the hospital. She was not his mother and was involved in petty trade activities that both helped support his household and allowed her to offer some level of support to her mother. It was not possible for him to remain at the hospital, as he too had to work to sustain his household. Djalil also did not want the boy to lose a year or more of school. The boy continued to go to school because his wounds were not painful and his ulcer was not noticeable if hidden under clothing. Djalil initially planned to send the boy to his grandmother’s house in a nearby village during school holidays to be treated by a traditional healer.
As a result of a mass BU outreach education program, Djalil learned about the availability of decentralized treatment at a health station a few kilometers from his house. He consulted the nurse at the health station, who assured him that he could treat his son’s wounds with medicine and bandaging if he adhered to a treatment that required daily visits to the health station for some months. Djalil agreed and engaged himself in agricultural activities at home instead of returning to Nigeria to work. He strictly adhered to the decentralized care offered by the nurse for eight months until his son had fully recovered. In his words, “I was greatly relieved to receive treatment here from the major (nurse) as this type of treatment is not available in the place of his mother in Nigeria. I brought my son to the health station every day for five months. Then, for two months, we visited the station every three days. Finally, in the eight month, I brought him every four days. My son remained at home, surrounded by family, and was able to complete his school year as well.”
Beyond research on BU, this study contributes to the fields of health service research, household and gender studies, and the study of conditional kinship obligations in contexts of poverty where “the capacity to care and the decisions about who undertakes care work are shaped by other considerations: resources and assets that facilitate the work and costs of care; expectations and commitments….; alternative obligations and responsibilities to other house holders; and affective ties to others that do not always follow prescribed kinship ideals” [54]. Geographical accessibility is an important enabling factor influencing healthcare seeking behavior, but as we note in this paper, cultural accessibility and social acceptability also contribute in significant ways. Understanding these factors helps us better understand why decentralized care for BU is preferred, and why centralized care, even when offered for free, is rejected by some households.
Ten important lessons from this study may be highlighted. First, BU related treatment decisions are seen to be to be largely pragmatic when one considers the well-being of the household as the primary unit of analysis and not individuals afflicted with the disease. This is not to say that ill individuals are neglected by their households, but that basic HHPH needs eclipse individual medical needs. Second, practical logic supersedes biomedical reasoning when hospital-based treatment requires a mother to be displaced from her household. Removing a mother places the household at risk unless adequate substitute labor can be secured. Even then, the opportunity cost of a woman not being able to engage in resource generating activities constitutes a threat to her household. It also diminishes her ability to support others with whom she shares bonds of reciprocal exchange and mutual assistance.
A third lesson pertains to social capital as essential for survival among those living on the margin. Women in Benin constantly balance the need to invest in activities that provide material goods for their own household’s well-being with the need to convert time and material resources into social capital through helping kin in times of need. In the case of BU, offering support to kin is a means of accruing bonding social capital. However, the support a woman can offer is contingent on her carrying capacity (how many people she supports) and a husband’s ability and willingness to make contributions to the household. In Benin, male contributions to the household are unpredictable if a husband is working outside his community for months at a time.
A fourth lesson relates to health care decision-making and women’s agency. In the cases presented, it was largely men and their kin who made initial decisions about where BU treatment would be received. However, it was generally up to women to mobilize support from within their own kin networks to enable therapy management. Women’s agency and ability to influence decision making over time came from the resources at their command and the resources they could mobilize.
A fifth, related lesson is that social support is conditional. There are limits to the amount and duration of support kin can provide. In cases of BU treated in hospital, resource sharing and the constitution of TMG membership shifted over time. The amount of time needed to heal BU related wounds is difficult to predict, and hospital staff are generally not very forthcoming when it comes to communicating to a patient about how their healing process is progressing [17]. As a result, hospital patients are not sure what to tell family members when they inquire. Advanced BU patients often have to remain in hospital far longer than family members and caretakers imagined.
The inability to predict how long BU treatment might take in hospital settings placed strains on social relations. Women often expected more support than they received based on their assistance to others in the past. This was clearly seen in cases of diminishing support and abandonment. Diminishing support caused women considerable distress not just about the present moment, but the likely future. When their social capital was expended, they had no safety net.
A sixth lesson concerns the many risks hospital-based BU treatment poses for women in what might be thought of as a hierarchy of risks that encompasses physical, economic, social, and supernatural risks [45,55]. Three primary and multiple secondary risks face women. The three primary risks identified in this study are: risks to household economic survival associated with their displacement, risks to non-afflicted children left in the care of others, and the social risk to the reputation of a mother or daughter when they are compelled to reside in the liminal space of a hospital.
A seventh lesson is that whenever possible, decentralized outpatient treatment of BU is favored over hospital care. Decentralized treatment avoids the indirect and opportunity costs of “free” hospital care, allows an afflicted child and one’s other children to stay in school, and enables agricultural labor and income-generating activities to continue to the extent possible. It also reduces the risk of pernicious rumor, especially when community leaders visibly support community-near outpatient care. Visiting local health stations further avoids the serious problem of social isolation commonly felt by patients and caretakers in hospital, minimizes marital stress, and reduces fear of abandonment [29].
An eighth and related lessons is that decisions to go for decentralized BU care are not solely based on the accessibility of points of care. It is well documented that access to health facilities is a major factor affecting health care seeking [56]. It is certainly the case, that visits to traditional healers for chronic wounds are in part due to their close proximity. However, our research suggests that even when a health station offering decentralized treatment was equidistant to a BU reference hospital, people still preferred going to the health station for all but the most serious cases of BU. Data from a pilot project in Ouinhi commune confirm this observation.
In a recent publication[31] we reported on a pilot study that introduced decentralized treatment in Ouinhi commune. Notably, when decentralized care was introduced by a nurse with special training in chronic wound management, many households that had refused centralized BU treatment agreed to outpatient treatment at a local health station once they became aware of the service. In the case of Djalil this was due to convenience and not wanting to take his son out of school. In the cases of women we interviewed who had made a similar choice; it also had to do with diminishing social risk following an outreach program. Following the introduction of mass BU outreach education programs attended by chiefs, healers, and health staff, women traveling daily to get decentralized treatment were not subject to rumor. While the hospital was deemed a liminal space where behavior was suspect, the health station was seen as a valued community-near institution.
A ninth lesson entails decisions made about which family member should serve as a patient caretaker when hospitalization is required. This is one of the first articles to draw attention to young children serving as patient caretakers in West Africa for both their siblings and their mothers [57,58]. Children are chosen to be caretakers for two reasons. First, they enable a mother to remain at home to take care of other children and household affairs. Second, prepubescent girls are chosen because they are less likely to be the object of rumors about sexual liaisons in the hospital.
This brings us to the tenth and final lesson, the need for in-depth qualitative research, beyond surveys, as a means to more fully understand predisposing, enabling, and service related factors influencing health care and patient caretaker decision-making. Two examples may be highlighted: social risk and mothers’ concern about quality of childcare. As already noted, women recognized the danger of being treated in hospital posed by rumor. Adult women under treatment for BU, or functioning as patient caretakers, almost universally brought young children with them. They did so both to care for these children and to buffer them against pernicious rumor. Notably, during semi-structured interviews, childcare issues were commonly mentioned as a reason for sending young daughters to serve as caretakers, but concerns about rumors were not mentioned. These concerns only emerged during illness narratives, which may be the reason that the importance of social risk, moral identity, and rumor have been little discussed in studies that have relied on surveys and structured interviews that have not specifically queried this subject.
A second example involves substitute childcare as a significant factor in women’s decision making about hospital treatment. Mothers carefully consider the quality of care a daughter, grandmother, sister, co-wife, or foster parent might provide their children. Notably, their assessment is not just based on whether the person can provide food and shelter. Emotional support is also quite important. It was only during the collection of illness narratives that quality of childcare emerged as a primary health concern. In cases where good quality of care was not deemed available, mothers either opted not to seek hospital care for a child afflicted with BU at the expense of the well-being of their siblings, or took her children with her to hospital. Taking multiple children to a hospital has led to all kinds of logistical issues for hospital administrators.
A qualitative study utilizing a purposeful sample is designed to identify the range of factors effecting phenomena: in this case decision making about BU treatment and the experience of patients and caretakers in hospital. The study was not designed to measure which factors are most responsible for treatment delay, drop out, or no show after BU identification. This will require a quantitative study, which measures the order of magnitude of factors identified in this study.
In this paper, we have used an ethnographic study of BU to make a case for focusing on the household as a unit of analysis when studying NTD related health care decision-making and its sequela in LMICs. Toward this end, we have found three conceptual lenses to be particularly useful: HHPH, gender relations, and TMG mobilization. Use of these lenses broadens our understanding of factors influencing treatment choice and patient caretaker selection when hospitalization is required. They also provide us with a nuanced account of the ripple effects of longstanding illnesses like BU beyond the household, and an appreciation that social support from kin is contingent and conditional.
In the areas of Benin studied, like many other regions of West Africa, men are the primary decision makers for healthcare decisions outside the home. A woman’s agency and ability to influence the decision-making process is largely based on whatever social support and substitute labor she can mobilize from her own network of kin relations. The brief BU case vignettes presented in this paper speak to the importance of bonding social capital for women in times of illness, describe ways in which this capital is accrued through reciprocal assistance, and draw attention to strains on social relationships when the duration of support needed is longer than expected and/or indeterminate. The three lenses further help us identify groups at risk for treatment delay, non-adherence, and patient abandonment at hospital due to structural vulnerability.
In conclusion, we argue that public health programs for diseases requiring long-term treatment, like BU, need to take into consideration household survival and gender relations and not just the medical needs of the afflicted. We concur with Grietens et al.[29] who argue that public health programs need to recognize the folly of designing programs that save the patient at the cost of compromising the integrity of the household and the health and well-being of other household members, especially children.
|
10.1371/journal.ppat.1006483 | Checkpoints of apicomplexan cell division identified in Toxoplasma gondii | The unusual cell cycles of Apicomplexa parasites are remarkably flexible with the ability to complete cytokinesis and karyokinesis coordinately or postpone cytokinesis for several rounds of chromosome replication, and are well recognized. Despite this surprising biology, the molecular machinery required to achieve this flexibility is largely unknown. In this study, we provide comprehensive experimental evidence that apicomplexan parasites utilize multiple Cdk-related kinases (Crks) to coordinate cell division. We determined that Toxoplasma gondii encodes seven atypical P-, H-, Y- and L- type cyclins and ten Crks to regulate cellular processes. We generated and analyzed conditional tet-OFF mutants for seven TgCrks and four TgCyclins that are expressed in the tachyzoite stage. These experiments demonstrated that TgCrk1, TgCrk2, TgCrk4 and TgCrk6, were required or essential for tachyzoite growth revealing a remarkable number of Crk factors that are necessary for parasite replication. G1 phase arrest resulted from the loss of cytoplasmic TgCrk2 that interacted with a P-type cyclin demonstrating that an atypical mechanism controls half the T. gondii cell cycle. We showed that T. gondii employs at least three TgCrks to complete mitosis. Novel kinases, TgCrk6 and TgCrk4 were required for spindle function and centrosome duplication, respectively, while TgCrk1 and its partner TgCycL were essential for daughter bud assembly. Intriguingly, mitotic kinases TgCrk4 and TgCrk6 did not interact with any cyclin tested and were instead dynamically expressed during mitosis indicating they may not require a cyclin timing mechanism. Altogether, our findings demonstrate that apicomplexan parasites utilize distinctive and complex mechanisms to coordinate their novel replicative cycles.
| Apicomplexan parasites are unicellular eukaryotes that replicate in unusual ways different from their multicellular hosts. From a single infection, different apicomplexans can produce as few as two or up to many hundreds of progeny. How these flexible division cycles are regulated is poorly understood. In the current study we have defined the major mechanisms controlling the growth of the Toxoplasma gondii acute pathogenic stage called the tachyzoite. We show that T. gondii tachyzoites require not only multiple protein kinases to coordinate chromosome replication and the assembly of new daughter parasites, but also each kinase has unique responsibilities. By contrast, the mammalian cell that T. gondii infects requires far fewer kinase regulators to complete cell division, which suggests that these parasites have unique vulnerabilities. The increased complexity in parasite cell cycle controls likely evolved from the need to adapt to different hosts and the need to construct the specialized invasion apparatus in order to invade those hosts.
| Obligate intracellular parasites of the phylum Apicomplexa are responsible for many important diseases in humans and animals, including malaria, toxoplasmosis and cryptosporidiosis. Severity of the disease is tightly linked to parasite burden, and currently, the most successful therapies block parasite proliferation. Apicomplexan parasites use flexible mechanisms to replicate that are different than those of their hosts. In endodyogeny each round of duplication is completed with assembly of two internal daughters [1]. Alternatively, multiple buds can be formed inside of the mother or emerge from its surface in the processes called, endopolygeny and schizogony, respectively [2]. Toxoplasma gondii undergoes endodyogeny in intermediate host stages, but replicates by endopolygeny in the definitive feline host. Fundamental differences between division modes are embedded in the features of the apicomplexan cell cycle comprised of two chromosome cycles [2, 3]. During nuclear cycles, chromosomes are replicated and segregated without budding, while each round of chromosome replication in the budding cycle leads to production of the daughter parasites. Ultimately, the number of nuclear cycles determines the scale of the parasite progeny. We recently showed that the ability to switch between chromosome cycles is partially linked to the unique bipartite structure of the T. gondii centrosome [3]. Weakening or separation of the outer centrosomal core that controls budding favors the nuclear cycle, while the strong association of the outer core with the inner core promotes cytokinesis and the budding cycle [3].
Endodyogeny of T. gondii tachyzoites represents one the simplest modes of replication and important cell cycle transition points where potential checkpoints may operate have been defined [2, 4–19]. The G1 phase of T. gondii endodyogeny comprises half of the division cycle, and like other eukaryotes, canonical housekeeping tasks preparing for S phase commitment are performed in the apicomplexan G1 period [4, 9, 13, 15, 18, 19]. Evidence also indicates that cell cycle exit to form dormant developmental stages as well as drug-induced dormancy is controlled by mechanisms acting at the transition from M/C into early G1 [4, 20, 21]. Duplication of the centrosome marks the transition from G1 to S phase [3, 6, 15, 17], and we have defined some of the components of this critical transition in T. gondii that should be targets of a G1/S checkpoint mechanism [17]. A peculiar feature of apicomplexan replication is the short (or absent) G2 phase [1, 2] that is thought to be marked by natural S phase populations that possess partially duplicated genomes (1.7–1.8N DNA) [1, 13]. Resolving the molecular basis for this important transition should help solve the mystery of these unusual DNA distributions. During the apicomplexan mitosis numerous specialized structures are replicated, built or converted in precise order to produce healthy infectious daughters (for review see [2, 22]). Our previous work and the studies of others have established that duplication of the bipartite centrosome is coordinated with division of the centrocone that holds the mitotic spindle [3, 6, 10, 23], and also with the replication and segregation of the bundled centromeres [23, 24], which in turn, is synchronized with assembly of the basal and apical complexes of the future daughter [25–28]. The temporal-spatial coordination of these overlapping processes likely requires similarly complex regulatory machinery, for which the molecular basis is still largely unknown.
In eukaryotes, cell cycle progression is governed by the activity of cyclin-dependent kinases (Cdks) and their regulatory cofactors, cyclins [29, 30]; dynamic expression of the latter provides clockwise control of Cdk function. Cdk4/6-cyclin D complexes support the progression of G1 phase, and Cdk2 complexes with cyclins E, A and B govern the progression and fidelity of DNA replication in S phase and chromosome segregation in mitosis [31]. Cdks functions were originally thought to be restricted to cell cycle regulation, however, today we understand that activated Cdk-cyclin complexes are master regulators of such major biological processes as transcription, RNA processing, translation and development [30]. Extrapolating current models of cell cycle checkpoints that involve Cdk-cyclins to eukaryotes in general is challenging, as there are many branches of the eukaryotic tree where cell division is quite unusual and the molecular controls are likely to be different [29]. This includes the large group of obligate intracellular parasites from the phylum Apicomplexa. Mining the initial genomes of important disease causing apicomplexans has revealed major differences [1, 32–34] characterized by the reduction of components and also the complete absence of the key regulatory elements, including canonical cyclins [35], major cell cycle Cdks [33, 34] and their immediate downstream effectors [1]. The lack of conserved cell cycle factors of higher eukaryotes indicates there are significant changes in the cell cycle molecular machinery of these ancient protozoa. Here we describe the first comprehensive study of Cdk-related kinases (Crks) and cyclins in T. gondii. Using genetic approaches, we have analyzed the function of seven TgCrks and four TgCyclins and uncovered major cell cycle TgCrks controlling the replication of the tachyzoite stage of T. gondii. Our results demonstrate that, unlike the traditional eukaryotic cell cycle, the intricate division of apicomplexan parasites is regulated by multiple essential Crks acting independently at several critical transitions and in unusual spatial contexts.
To determine the core Cdk-cyclin complexes that regulate division in T. gondii, we systematically searched the parasite genome and identified ten genes encoding a kinase domain that included a cyclin-binding sequence (C-helix) (S1 Fig) [30]. Following standard convention, we named these factors Cdk-related kinases (Crks) until cyclin-dependent activation of the kinase can be established [36, 37]. T. gondii Crks (TgCrks) were a diverse group of proteins ranging from 34 to 212kDa (S1A Fig) that also varied in mRNA abundance (S1B Fig). T. gondii transcriptome data (ToxoDB) indicated that eight TgCrks were expressed in tachyzoites and/or bradyzoites, and mRNA profiles for two kinases, TgCrk2-L1 and TgCrk5-L1 indicated they were restricted to the definitive or environmental life cycle stages; merozoite and/or sporozoite (ToxoDB). The profiles of TgCrk4, TgCrk5 and TgCrk6 mRNAs were cyclical in tachyzoites (S1B Fig) [4], which was confirmed at the protein level for TgCrk4HA and TgCrk6HA by endogenous epitope tagging (S3 Fig). The dynamic cell cycle regulation of the three TgCrk factors differs significantly from the typical constitutive expression of Cdks of studied model eukaryotes [38, 39].
Phylogenetic analysis of TgCrks defined the general Cdk families present in the Apicomplexa phylum [34]. Putative Crks from T. gondii, P. falciparum [33, 34], Theileria annulata and Cryptosporidium parvum were compared to the ancestral free-living unicellular eukaryote Chromera velia, and the extensively studied Cdks of human cells. The analysis sorted ten TgCrks into eight general phylogenetic clades with five groups restricted to the superphylum Alveolata (Fig 1, pink shade): TgCrk5 and TgCrk5-L1 were apicomplexan adaptations. Kinases in three other clades were shared with a recognizable higher eukaryotic counterpart that included several kinases known to regulate cell cycle and gene expression. Specifically, TgCrk1 grouped with the Cdk11 family kinases that regulate mRNA synthesis and maturation (Fig 1, green shade), while TgCrk7 was similar to the Cdk-activating kinase (CAK), HmCdk7 (Fig 1, yellow shade). Lastly, TgCrk2 clustered with the family of eukaryotic cell cycle regulators, in particular, with neuronal HmCdk5 (Fig 1, light blue shade).
To define the function of TgCrks, we constructed tet-OFF conditional knockdown mutants by replacing the native promoter with a tetracycline-regulatable promoter in the Tati-RHΔku80 strain [24, 40]. Each kinase was concurrently tagged with a 3xHA-epitope fused to the N-terminus (S1D Fig, diagram). We successfully generated tet-OFF mutants for seven TgCrks demonstrating that promoter replacement and N-terminal HA-tagging were remarkably tolerated in the tachyzoites (Fig 2 and S1F Fig). The only other TgCrk factor expressed in tachyzoites is TgCrk5, which is the subject of another project and was not studied here (Naumov and White, personal communication). Immunofluorescence assays (IFA) of HATgCrks determined that most of these kinases were predominantly nuclear in tachyzoites (Fig 2, -ATc conditions) with the exception of HATgCrk2, which was expressed throughout the cell, and HATgCrk4, which was exclusively localized to the cytoplasm. Despite sharing the same tet-OFF promoter, individual HATgCrks showed a wide range of protein abundance that closely matched the expression level of the factors regulated by their native promoters, indicating there are major post-transcriptional mechanisms controlling TgCrk levels in tachyzoites (S1E Fig).
The HATgCrk proteins were all successfully down regulated by a 24 h incubation with 1μg/ml anhydrotetracycline (ATc) (Fig 2, IFA and Western blot analysis). Exploiting the ATc-induced conditional knockdown, we determined by plaque growth assay (Fig 2) that four TgCrks were essential, two were required for tachyzoite growth and only TgCrk8 was dispensable. Quantitative growth rates (24 h) for the TgCrk4 and TgCrk6 tet-OFF mutants with or without ATc treatment confirmed that TgCrk4 was required and TgCrk6 was essential for tachyzoite growth (S1F Fig). IFA analysis determined that ATc-induced growth arrest of TgCrk3 and TgCrk7 tet-OFF mutants was cell cycle independent (Fig 2 and S2B Fig) consistent with the potential role in regulating general cellular processes required around the entire cell cycle. Recent studies of TgCrk7, and its P. falciparum ortholog Pfmrk, implicated a role in transcriptional regulation [41, 42] and PfCRK3 kinase complexes were associated with chromatin-dependent regulation of gene expression [37]. Based on parasite morphology, the loss of TgCrk2 appeared to block tachyzoites in the G1 phase and knockdown of TgCrk1, TgCrk4 or TgCrk6 resulted in extensive mitotic and cytoskeletal defects, consistent with growth arrest in the S/M/C half of the cell cycle (S2A Fig). These four kinases were selected for further characterization in this study.
In general cyclins are poorly conserved, although the presence of a cyclin box and one or more destruction motifs permits these genes to be identified by genome mining. Utilizing this approach we identified seven novel cyclin factors in the T. gondii genome (S4A Fig) [35]. Only cyclins related to P-, H-, L- and Y- types were found, while no canonical A-, B-, D-, and E- types, that are vital to higher eukaryotic cell division, were identified [43]. Based on the extensive transcriptome data (see ToxoDB), five of the seven TgCyclins appeared to be expressed in tachyzoites (log2 RMA value higher then 6, S4B Fig). To determine expression and localization of TgCyclins in tachyzoites, we epitope tagged the C-terminus of TgCycH, TgCycL, TgCycY with a 3xHA-epitope by genetic knock-in. IFA analysis showed that TgCycHHA, TgCycLHA and TgCycYHA, were moderately expressed and localized to the nucleus in tachyzoites (Fig 3A), and remarkably, TgCycYHA was the only oscillating cyclin with peak expression in the G1 phase (Fig 3B). To visualize the lower abundant cyclin, TgPHO80, we engineered transgenic parasites ectopically expressing TgPHO80 that was N-terminally tagged with a 3xmyc epitope fused to a FKBP destabilization domain (DDmyc), which permits conditional expression using the small molecule Shield 1 [15, 44]. After 3 h stabilization with Shield 1 (100nM), DDmycTgPHO80 was found in large cytoplasmic speckles (Fig 3A), which was the only cyclin exclusively localized to the tachyzoite cytoplasm.
Utilizing a similar knockdown approach as was used for the TgCrks (S1D Fig), we successfully converted by genetic knock-in four TgCyclin genes to tet-OFF mutant alleles: TgPHO80, TgCycH, TgCycL and TgCycY (Fig 3C, Western Blot analysis). ATc-knockdown of TgPHO80 slowed the rate of replication (S4C Fig) and increased the fraction of parasites with a single centrosome (76.5% ± 9.8 vacuoles after 48 h with 1μg/ml ATc compared to 49% ± 1 vacuoles in -ATc conditions) indicating the G1 phase was lengthened by the loss of TgPHO80. The ATc-induced depletion of nuclear cyclins, TgCycL and TgCycH, caused tachyzoite growth arrest. TgCycL deficiency led to mitotic death (S2A Fig) and knockdown of TgCycH resulted in a quick non-specific growth arrest (Fig 3C and S2B Fig). The conditional knockdown of TgCycY with ATc demonstrated this cyclin was not essential for tachyzoite growth (Fig 3C and growth quantification in S4C Fig).
The conditional knockdown of the tet-OFF TgCrk2 mutant appeared to cause a cell cycle arrest in G1, which we confirmed by analyzing centrosome duplication. As expected, the majority (~80%) of TgCrk2-depleted parasites (+ATc) possessed a single centrosome (Fig 4A) indicative of G1 phase arrest. Since canonical G1 cyclins (e.g. D- and E- type) [29] were absent from the T. gondii genome, it was of particular interest to identify the T. gondii cyclin that binds TgCrk2. We genetically engineered a series of dual epitope-tagged strains by first knocking-in 3xHA epitope tag into the TgCrk2 gene followed by stable ectopic expression of four different TgCyclins that were epitope tagged with 3xmyc (see S1 Table for list of transgenic strains and Material and methods for production details). TgCrk2 protein complexes were affinity purified with antibody against the epitope tag and then probed with alternative antibody to define the interacting cyclin. We detected a weak interaction between TgCrk2 and TgCycH (Fig 4B) confirming previous yeast two-hybrid screens [45], however, TgCrk2HA formed the most abundant complexes with cyclin TgPHO80 (Fig 4B), which corroborates the G1 phenotype we observed for the TgPHO80 tet-OFF mutant following ATc-knockdown (Fig 3C and S4C Fig).
Conditional loss of TgCrk1 in the tet-OFF mutant resulted in severe chromosome mis-segregation and accumulation of deformed zoites (Fig 2). To more precisely define the TgCrk1-loss phenotype, we focused on the early mitotic steps just prior to daughter bud formation. The MORN1 protein associates with two compartments that are duplicated in mitosis, the spindle compartment (centrocone) and basal complex referred to here as the MORN-ring (Fig 5A, image a) [27, 28]. Utilizing the MORN1 marker, we determined that in parasites lacking TgCrk1 (+ATc for 16 h) the development of the daughter MORN1 rings was defective (Fig 5A and 5B, compare image a to c, e). Deficiency of the basal complex became more evident when newly produced alveolar sacs (Fig 5A and 5B, image f) accumulated near fragmented MORN-rings in the late stages of mitosis (Fig 5A and 5B image e). In addition, the normal coordination of mitosis with cytokinesis failed and the IMC compartment appeared as an unstructured mass (Fig 5A and 5B, image h). In addition, TgCrk1 deficiency affected the structural integrity of the tachyzoite apical end (Fig 5C). Following ATc treatment the robust cone-shaped apical cap (ISP1 staining) [46] became a weak rod-like or lopsided structure that caps a deformed IMC1-positive mass (Fig 5C). Despite severe defects in mitosis caused by the loss of TgCrk1, the duplication and segregation of centromeres, centrosomes and the plastid (S5A Fig) and nuclear division were evident (Fig 5A, image i). Altogether these results indicate that TgCrk1 primarily regulates cytokinesis but not karyokinesis.
Constitutively expressed TgCrk1Ty (S5B Fig) and TgCycLHA are similarly localized and knockdown phenotypes of TgCycL and TgCrk1 tet-OFF mutant parasites were also similar (Figs 2 and 3 and S2A Fig), indicating these two proteins may be paired in T. gondii. Indeed, co-IP experiments using the dual epitope-tagging approach employed with TgCrk2 above, confirmed preferential interaction of TgCrk1 with TgCycL (Fig 5D). Furthermore, IFA of the dual-tagged strain co-expressing TgCrk1HA and TgCycLmyc revealed tight co-localization of these factors in a unique sub-nuclear compartment (Fig 5E) that was independent of the nucleolus (TgNF3, [47]), centromere-compartment (TgCenH3, [23]), centrocone (MORN1, [26]), or nascent particles containing the RNA polymerase complex (TgRPB4, [48]) (S5C Fig).
Conditional knockdown of TgCrk4 and TgCrk6 tet-OFF mutants also caused severe mitotic defects (Fig 2 and S2A Fig) indicating that similar to TgCrk1, these Apicomplexa-specific kinases function in the second half of the tachyzoite cell cycle. IFA analysis of endogenously tagged TgCrk6HA and TgCrk4HA determined that these proteins have different subcellular localization (Fig 2). TgCrk4HA was distributed in large cytoplasmic aggregates with accumulation in the apical perinuclear region (S3C Fig), while TgCrk6HA extended its nuclear localization to the centromeric region (S3D Fig, CenH3 and Centrin1 staining) [23].
To build clues to TgCrk6 function, we performed detailed IFA analysis following short term ATc treatment (16 h) using the MORN1 (mitotic structures) and IMC1 (cytoskeleton) cell cycle markers. The normal assembly of the daughter scaffold is initiated in late S phase followed by duplication/separation of the centrocone (Fig 6A, -ATc) [2, 3]. Cytokinesis progressed in TgCrk6-deficient parasites, yet the centrocone spindle compartment was not properly duplicated as evidenced by the single MORN1-positive dot positioned between two growing daughter buds (Fig 6A, +ATc). Other evidence supported TgCrk6 function in karyokinetic processes. ATc-downregulation of TgCrk6 disrupted the usual dynamics of kinetochores visualized by co-staining of the kinetochore complex component, TgNdc80myc, and acetylated Tubulin A that labels active sites of the microtubule assembly including spindle and internal daughters (Fig 6B) [24]. In normal parasites, the TgNdc80myc signal largely disappeared at mid-bud development (Fig 6B, -ATc image e). By contrast, TgCrk6 deficient parasites (+ATc) retained single assembled kinetochores well into the budding process (Fig 6B, image g). Longer term ATc incubations (>24 h) amplified the loss of coordination between cytokinesis and karyokinesis leading to the catastrophic phenotype of severe DNA mis-segregation and assembly of buds lacking DNA shown in Fig 2. These results combined with the observed accumulation of the nuclear TgCrk6HA in the centromeric region during peak expression in S/M phase (S3D Fig) supported a key role for TgCrk6 in T. gondii spindle regulation. Similar deficiency to split a spindle was recently described in the knockdown mutant of the distantly related P. falciparum CRK4 (Fig 1) [49].
Similar to TgCrk6, knockdown of TgCrk4 caused defects in mitosis (Fig 2), although TgCrk4 differed from TgCrk6 in being localized to the cytoplasm rather than the nucleus (S3B versus S3D Fig). This difference led us to examine the role of TgCrk4 in regulating the cytoplasmic components of the mitotic machinery. Asexual stages of T. gondii divide by enclosed mitosis (as do most apicomplexans) that coordinates attachment of nuclear centromeres to kinetochores/spindle and to a unique centrosome containing two independent functioning core structures [3, 22]. Consistent with a role in controlling mitosis through cytoplasmic structures, down regulation of TgCrk4 with 1μg/ml ATc led to defective duplication of both centrosomal cores (Centrin1/outer core and CEP250myc/inner core), but did not affect centromere duplication/segregation (CenH3 marker) or nuclear division (Fig 7A). Interesting, plastid segregation, which is controlled by the centrosome [50], was also defective in parasites lacking TgCrk4 (Fig 7A, TgAtrx1 marker). Although TgCrk4-deficient parasites showed abnormal centrosome replication (under and over reduplication) (Fig 7A and 7B), we did not observe uncoupling of the centrosome cores (Fig 7C) as we have documented in some temperature sensitive mutants [3]. Moreover, we found that centrosome re-duplication occurred around assembled kinetochores (Fig 7D) despite the disruption in normal centrosome stoichiometry. The proper ratios of centrosome to kinetochore observed in a regular tachyzoite mitosis are established (Fig 7D, -ATc; 2:1 images a, b; 2:2 images c, d; 2:0 images e, f). By contrast, down regulation of TgCrk4 led to abnormal stoichiometry of 4 centrosomes to 2 assembled kinetochores (Fig 7D, +ATc, images g, h). Further examination revealed that only one of the reduplicated centrosomes remained associated with the nucleus (Fig 7C, inset), which may explain why the loss of TgCrk4 did not lead to unregulated karyokinesis. Altogether, our results support the role of TgCrk4 kinase in the regulation of centrosome duplication and segregation during mitosis within the context of other essential mitotic regulatory controls such as TgCrk6 above.
Interestingly, TgCrk4 and TgCrk6 were cyclically expressed during tachyzoite replication (S3 Fig) with the peak expression in S/M phase consistent with functions in regulating mitotic processes. In higher eukaryotes, mitotic Cdk activity is typically controlled by an oscillating cyclin partner, while the Cdk protein is constitutive [29, 38, 39]. To identify the cyclin partners for TgCrk4 and TgCrk6, we performed co-IPs from dual tagged strains expressing TgCrk4HA or TgCrk6HA and four different ectopically expressed TgCyclins (See Material and methods). No detectable interaction between the TgCyclins tested and TgCrk4 and TgCrk6 was observed (S4D Fig) suggesting T. gondii may have become dependent on direct mechanisms of dynamic expression to regulate mitotic TgCrk4 and TgCrk6 factors leading to the loss of a periodic activating cyclin partner.
The molecular basis of cell cycle regulation in eukaryotes has been mainly shaped by studies in one branch of eukaryotes, Unikonta that includes the clades animalia, fungi and amoebas [29]. This is a significant deficiency because the replication biology of eukaryotes from the Bikonta branch, comprised of the three supergroups, the Excavata, SAR (Stramenopiles, Alveolates and Rhizaria) and Archaeplastida, is quite extraordinary, if also beyond our reach experimentally [51–53]. From this point of view, our genetic analysis of T. gondii Cdk-related kinases and cyclins provides much needed insight into the cell cycle regulation of an ancient protozoan from the SAR supergroup. One of the core findings of this study is the surprising complexity and unusual regulation of cell cycle controls that are essential for Apicomplexa cell division. The results of multiple gene knockouts in higher eukaryotes reveals that a single active Cdk (Cdk1/2 family) is sufficient to sustain basic chromosome segregation in the somatic cells of both multicellular and unicellular eukaryotes [29, 54]. By contrast, this study and a second project in progress (TgCrk5 studies, Naumov and White, personal communication) have established that T. gondii requires five Crks to successfully regulate the peculiar parasite cell cycle called endodyogeny (Fig 8). Two Crks regulate centrosome duplication (TgCrk4) and organization of the daughter bud cytoskeleton (TgCrk1) during interwoven S, M and C phases. T. gondii has also evolved independent controls for the restriction or START checkpoint in G1 (TgCrk2), the DNA licensing checkpoint in S phase (TgCrk5, Naumov and White, personal communication) and the spindle assembly checkpoint (TgCrk6) acting at the metaphase to anaphase transition in mitosis (Fig 8). A Recently published study of P. falciparum CRK4 confirms that cell cycle checkpoints are regulated by the related kinases across Apicomplexa phylum [49]. Similar to TgCrk6, nuclear PfCRK4 was dynamically expressed in S/M phase (late trophozoite to schizont during blood stage) and upon downregulation parasites lost the ability to split the spindle and, consequently properly segregate chromosomes (Fig 2b, f and g in [49]). The high number of the putative cell cycle checkpoints was unexpected and this favors a model of much tighter cell cycle regulation in the Apicomplexa than previously thought [1]. In fact, the lack of reversible and abundant catastrophic phenotypes observed in T. gondii cell cycle mutants [9], which was often interpreted as a lack of cell cycle controls, is likely a consequence of the complexity of this system. We propose that apicomplexan parasites evolved separate Crks for individual cell cycle stages to facilitate switches between flexible division modes [2, 3]. During the chromosome cycle of schizogony or endopolygeny, the G1 phase is completed only once and is uncoupled from the multiple rounds of the S/M phase, which are, in turn, uncoupled from the budding process until the very last unified S/M/C phase. Therefore, evolution of multiple Crks offers independent control of the segments, permitting modular regulation of the complex cell cycles, yet, leaving an open question of the master regulator(s) of the switch.
Analyzing the T. gondii cell cycle we noticed multiple parallels in topology of the cell cycle regulation between apicomplexans and a few studied Bikonta models, particularly, plants. First, similar to plants, mitosis in apicomplexans is regulated by clade-specific Cdks (Fig 1) [49, 55]. Second, mitotic TgCrk4, TgCrk5 (Naumov and White, personal communication), TgCrk6 and PfCRK4 [49] are dynamically expressed (protein and mRNA), which is a distinctive feature of the CDKB family kinases in plants [55, 56]. Third, similar to most Archaeplastida members, apicomplexan parasites do not possess or encode a highly diverged Cdc25 phosphatase ortholog (EupathDB) [56, 57], which primary function is to activate mitotic Cdks. Interestingly, based on the functional parallels between Arabidopsis thaliana CDKB1;1 and Drosophila melanogaster Cdc25, plant biologists proposed that the Cdc25-controlled onset of mitosis may have been evolutionary replaced by plant-specific B-type Cdk pathway [57]. Whether the similar scenario had taken place in Apicomplexa evolution will require further studies, involving broader analysis of the Bikonta organisms. Unfortunately, the current limitation of the bikont studies also does not permit us to make a definitive conclusion of whether the plant-like features of the Apicomplexan cell cycle were inherited at the time of the Unikonta and Bikonta diversion or were the result of a secondary symbiosis of the Chromoalveolata and red algae [58].
The majority of characterized cell cycle cyclin-Cdk complexes are composed of constitutive and dynamic subunits with the cyclin generally the oscillating partner [38]. Consistent with the concept that functional topology is more important than the conservation of individual parts [59], we found that T. gondii cyclins differed from their high eukaryotic counterparts in being pre-dominantly constitutively expressed. Intriguingly, a single oscillating cyclin, TgCycY, was not essential for tachyzoite division nor did it interact with the seven TgCrks we analyzed (S4D Fig) and, therefore, a possible Cdk-independent role will need to be explored in future studies [60]. In fact, only two cell cycle complexes, TgCrk2-TgPHO80 and TgCrk1-TgCycL, were detected in which both subunits seem to be constitutively expressed in tachyzoites, suggesting mechanisms other then cyclin-binding that regulate cell cycle activity of TgCrk1 and TgCrk2. On the contrary, mitotic TgCrk4 and TgCrk6 are rare examples where Cdk-related kinases are dynamically expressed and not found in the complex with TgCyclins (S4D Fig). Given that the role of cyclin is to provide a temporal context to Cdk function, it is possible that T. gondii no longer needs cycling partners for TgCrk4 and TgCrk6. There is reason to speculate that constitutively expressing mitotic kinases might not be ideal given the complexity of the apicomplexan mitosis that is associated with the extensive de novo biosynthesis of the motility and invasion apparatus of daughter parasites [4]. Delivering the master conductors "just-in-time" would avoid accidentally triggering the cascade of mechanisms that unfold during mitosis and cytokinesis before the parasite is ready to egress. This hypothesis, however, cannot explain the absence of cyclin interaction with the constitutively expressed TgCrk3 and TgCrk8 (S4D Fig). Interestingly, our study revealed that the only T. gondii Crks that have orthologs in other eukaryotes (Fig 1, TgCrk1, TgCrk2 and TgCrk7) interacted with conventional cyclins that were also expressed in other eukaryotes (TgCycL, TgPHO80 and TgCycH). Since all the novel TgCrks were orphan, it is tempting to suggest that a non-cyclin factor may have co-evolved that acts as an oscillating component in the complexes with novel TgCrks. Alternatively, these unusual TgCrks may function without a cyclin partner. Future unbiased approaches will be needed to sort out these possibilities, and identify if there are other protein co-factors that have replaced the traditional cyclin partner.
Another core finding was the strong molecular support for the remarkable physical partitioning of Apicomplexan cell division functions. Similar to other apicomplexans, T. gondii divides by enclosed mitosis where a set of tethered structures localized in the nucleus or in the cytoplasm must be constructed/deconstructed (kinetochores, spindle microtubules and striated fibers), duplicated/segregated (bipartite centrosome and centrocone) or assembled (buds) in a timely manner to produce infectious progeny [2, 3, 8]. Moreover, in our previous study [3], we demonstrated that T. gondii has divided the regulatory responsibility for karyokinesis and cytokinesis between two unique centrosome cores that have fixed orientation to nuclear and cytoplasmic biosynthetic events. Results of our study here largely support this nuclear/cytoplasmic organization, which was likely needed to overcome limitations of enclosed mitosis. T. gondii has evolved the nuclear TgCrk6 mechanism to control events requiring the intranuclear spindle, while cytoplasmic TgCrk4 regulates centrosome duplication and associated plastid segregation. While proper assembly of the nuclear spindle and cytoplasmic daughter bud were essential processes, reduplication of the centrosome in TgCrk4-deficient parasites only partially affected tachyzoite survival. We believe that apicomplexans may have relaxed the control of centrosome reduplication in order to more easily adapt cell division to the scale needed in different hosts [3, 22]. It should be noted that dissolution of the nuclear membrane during open mitosis in higher eukaryotes is a key difference with apicomplexan cell division that may permit extensive re-arrangements of the nuclear (e. g. kinetochore, spindle) and cytoplasmic (e. g. centrosome) structures by a single Cdk.
A possible exception to the nuclear/cytoplasmic functional organization is the assembly of the cytoplasmic daughter buds that was unexpectedly controlled by the nuclear TgCrk1-TgCycL complex. Initiation of the daughter buds near the centrocone, spindle pole, which continues to grow with the progression of mitosis, is a specialized event occurring in the budding cycle of Apicomplexa [2, 25]. How does the nuclear TgCrk1 regulate assembly of a cytoplasmic structure? Recent studies of eukaryotic splicing kinase Cdk11 discovered an unexpected role for this kinase in mitotic progression [61]. It has been shown that activity of Cdk11 is required to regulate sister chromatid cohesion [62]. Since knockdown of T. gondii TgCrk1 did not affect DNA segregation, it is possible that the role of TgCrk1 was re-adapted to regulate splicing of mRNAs whose products will be required to control assembly of the daughter buds. Our hypothesis is supported by the fact that many components of the cytoskeleton are delivered “just-in-time” during cell cycle progression [4]. In future studies, the analysis of transcriptome changes caused by TgCrk1 deficiency will help determine whether this kinase operates primarily as a regulator of mRNA expression.
Transmission stages are formed at the end of each apicomplexan life cycle, that are, in most species, specialized G1/G0 states. For example, mature bradyzoites and sporozoites of T. gondii remain growth arrested until appropriate external signals from the same or new host trigger recrudescence or de-differentiation, respectively, resulting in re-entry into G1 phase of the asexual proliferative cycle (Fig 8). In higher eukaryotes, Cdk4/6-Cyclin D complexes are responsible for the cell fate decision to divide or differentiate (restriction checkpoint) [63, 64], which are factors not present in the Apicomplexa. Here we showed that T. gondii parasites have replaced the canonical G1 machinery of higher eukaryotes with a novel complex of TgCrk2 (Cdk5 family) and TgPHO80 cyclin to regulate progression through the tachyzoite G1 phase. Recent studies in kinetoplastids and another apicomplexan, Plasmodium berghei, determined that a related G1 Crk and a P-type cyclin regulate developmental stages in these protozoans, which is similar to our discoveries of TgCrk2 function in T. gondii (Fig 8) [35, 65, 66]. It is also worth noting that T. gondii possesses paralogs of TgCrk2 and TgPHO80 cyclin that are developmentally regulated (ToxoDB, S1 and S4 Figs) opening the possibility that the TgCrk2-TgPHO80 pathway has diverse functions in development and could also be involved in the regulation of drug-induced dormancy.
In conclusion, the systematic approach we have used to analyze the cell cycle machinery in T. gondii has opened the way into learning how cell division is regulated in apicomplexan parasites. Our study has also evoked important issues that still need to be addressed. For example, greater numbers of TgCrks requires greater coordination; so is there a master regulator after all? Can the complexity of mitosis coupled to cytokinesis in T. gondii division explain a rise of multiple mitotic Crks? What are the mechanisms responsible for periodic Crks, and how are these mechanisms controlled? Does the lessons learned here translate to the exotic mitotic mechanisms in related alveolates [51, 52]? Clearly, there is still much to be done to understand the molecular basis of apicomplexan cell division, fortunately, we now have important new genetic tools to go forward.
No human subjects or animal research were used in this study.
T. gondii strains RHΔhxgprt [67], Tati-RHΔku80 [40], and RHΔku80Δhxgprt [68] were cultured in human foreskin fibroblasts (HFF, gift Dr. David Roos, University of Pennsylvania) according to published protocols [69]. Viability of transgenic strains was measured in plaque assays as previously described [15]. Monolayers of HFF cells were infected with 150–200 parasites per 35mm dish and individual plaques formed after 6 days were stained with crystal violate and counted. To determine division rates, parasites were counted by IFA using α-IMC1 (surface; kindly provided by Dr. Gary Ward, University of Vermont) antibody and DAPI (nucleus) in 50 randomly selected vacuoles in three biological replicates after 24 hours growth. Statistical significance was calculated using an unpaired T-test and Bonferroni correction (Prism6).
Transgenic strains and primers created in the study are listed in S1 Table.
Monolayers of HFF cells were grown on coverslips and infected with parasites under indicated conditions. Cells were fixed in 4% PFA, permeabilized with 0.25% Triton X-100, blocked in 1%BSA and incubated sequentially with primary and secondary antibody [16]. The following primary antibodies were used: mouse monoclonal α-ISP1 (clone 7E8) [46] and α-Atrx1 (clone 11G8) (kindly provided by Dr. Peter Bradley, UCLA) [75], α-TgCenH3 [23] (kindly provided by Dr. Boris Striepen, University of Georgia, Athens), α-acetylated alpha Tubulin (Abcam), rat monoclonal α-HA (clone 3F10, Roche Applied Sciences), rabbit polyclonal α-myc (Cell Signaling Technology), α-Human Centrin 2 [17], α-MORN1 (kindly provided by Dr. Marc-Jan Gubbels, Boston College) [27] and α-IMC1 (kindly provided by Dr. Gary Ward, University of Vermont). Alexa-conjugated secondary antibodies of the different emission wavelengths (Molecular Probes, Thermo Fisher Scientific) were used at a dilution of 1:1000. Stained parasites on the coverslips were mounted with Aqua-mount (Lerner Laboratories), dried overnight at 4°C, and viewed on a Zeiss Axiovert Microscope equipped with 100x objective. Images were collected and processed using Zeiss Zen software and were further processed in Adobe Photoshop CC using linear adjustment when needed.
Transgenic parasites co-expressing epitope-tagged TgCrks and TgCyclins were grown for 30–32 hours at 37°C. When expression of the factor was conditional, 100nM Shield1 was added for the last 3 hours of incubation. Parasites (3x108) were collected, washed in PBS and lysed in 1xPBS with 0.5% NP-40, 400mM NaCl, protease and phosphatase inhibitors (Thermo Fisher Scientific) on ice for 30 min. Total protein extract obtained by centrifugation at 21,000xg, 10 min, 4°C was divided and incubated with α-HA or α-myc magnetic beads (MBL International) for 1 h at room temperature. Isolated protein complexes on the beads were washed three times with the lysis buffer and eluted by heating in the Leamlli sample buffer at 65°C, 10 min. Protein extract before and after pull-down, and purified protein complexes were analyzed by Western blotting.
To prepare samples of the total extracts, parasites were purified by filtering through 3μm polycarbonate filters (EMD Millipore), washed in PBS, re-suspended with Leammli loading dye and lysed at 65°C for 10 min. To analyze individual fractions after immunoprecipitation, an aliquot of the fraction was mixed with Leammli loading dye and heated for 10 min at 65°C. After separation on SDS-PAGE gels, proteins were transferred onto a nitrocellulose membrane and probed with monoclonal α-HA (3F10, Roche Applied Sciences), α-myc (Cell Signaling Technology) and α-Tubulin A (12G10, kindly provided by Dr. Jacek Gaertig, University of Georgia) antibodies. After incubation with secondary HRP-conjugated antibodies, proteins were visualized by enhanced chemiluminescence detection (PerkinElmer).
Because of the lack of cell cycle gene annotation in apicomplexans, we searched the T. gondii genome (toxoDB) for protein kinases with cyclin binding C-helix and proteins containing a cyclin box showing similarity to mammalian Cdks and cyclins, respectively. Then, to reduce complexity of the analysis we first identified Cdk classes preserved in apicomplexans by comparing Cdk-related kinases of T. gondii and Cdks of human cells (ncbi.org). The evolutionary history of T. gondii Crks was inferred by using the Maximum Likelihood method based on the Whelan And Goldman model [76] and were conducted in MEGA7 [77]. The analysis involved 54 amino acid sequences of the putative Cdk-related kinases from T. gondii (Tg, ToxoDB), P. falciparum (Pf, PlasmoDB), T. annulata (TA, PiroplasmaDB), C. parvum (Cgd, CryptoDB), C. velia (Cvel, CryptoDB) and pre-selected Cdks from human cells (Hs, ncbi.org) that are listed in S1 Table. All positions containing gaps and missing data were eliminated. There were a total of 158 positions in the final dataset. The bootstrap consensus tree inferred from 100 replicates.
|
10.1371/journal.pcbi.1005506 | When do correlations increase with firing rates in recurrent networks? | A central question in neuroscience is to understand how noisy firing patterns are used to transmit information. Because neural spiking is noisy, spiking patterns are often quantified via pairwise correlations, or the probability that two cells will spike coincidentally, above and beyond their baseline firing rate. One observation frequently made in experiments, is that correlations can increase systematically with firing rate. Theoretical studies have determined that stimulus-dependent correlations that increase with firing rate can have beneficial effects on information coding; however, we still have an incomplete understanding of what circuit mechanisms do, or do not, produce this correlation-firing rate relationship. Here, we studied the relationship between pairwise correlations and firing rates in recurrently coupled excitatory-inhibitory spiking networks with conductance-based synapses. We found that with stronger excitatory coupling, a positive relationship emerged between pairwise correlations and firing rates. To explain these findings, we used linear response theory to predict the full correlation matrix and to decompose correlations in terms of graph motifs. We then used this decomposition to explain why covariation of correlations with firing rate—a relationship previously explained in feedforward networks driven by correlated input—emerges in some recurrent networks but not in others. Furthermore, when correlations covary with firing rate, this relationship is reflected in low-rank structure in the correlation matrix.
| A central question in neuroscience is to understand how noisy firing patterns are used to transmit information. We quantify spiking patterns by using pairwise correlations, or the probability that two cells will spike coincidentally, above and beyond their baseline firing rate. One observation frequently made in experiments is that correlations can increase systematically with firing rate. Recent studies of a type of output cell in mouse retina found this relationship; furthermore, they determined that the increase of correlation with firing rate helped the cells encode information, provided the correlations were stimulus-dependent. Several theoretical studies have explored this basic structure, and found that it is generally beneficial to modulate correlations in this way. However—aside from mouse retinal cells referenced here—we do not yet have many examples of real neural circuits that show this correlation-firing rate pattern, so we do not know what common features (or mechanisms) might occur between them.
In this study, we address this question via a computational model. We set up a computational model with features representative of a generic cortical network, to see whether correlations would increase with firing rate. To produce different firing patterns, we varied excitatory coupling. We found that with stronger excitatory coupling, there was a positive relationship between pairwise correlations and firing rates. We used a network linear response theory to show why correlations could increase with firing rates in some networks, but not in others; this could be explained by how cells responded to fluctuations in inhibitory conductances.
| One prominent goal of modern theoretical neuroscience is to understand how the features of cortical neural networks lead to modulation of spiking statistics [1–3]. This understanding is essential to the larger question of how sensory information is encoded and transmitted, because such statistics are known to impact population coding [4–8]. Both experimental and theoretical inquiries are complicated by the fact that neurons are widely known to have heterogeneous attributes [9–14].
One family of statistics that is implicated in nearly all population coding studies is trial-to-trial variability (and co-variability) in spike counts; there is now a rich history of studying how these statistics arise, and how they effect coding of stimuli [15–19]. Recent work by numerous authors has demonstrated that the information content of spiking neural activity depends on spike count correlations and its relationship (if any) with stimulus tuning [15, 17, 19–21]. Since a population of sensory neurons might change their firing rates in different ways to stimuli, uncovering the general mechanisms for when spiking correlations increases with firing rate (or when they do not) is important in the context of neural coding. Thus, we study this question in a general recurrent neural network model.
One observation that has been made in some, but not all, experimental studies is that pairwise correlations increase with firing rates. This relationship has been observed in vitro [22] and in several visual areas: area MT [23], V4 [24], V1 [25, 26], and notably, in ON-OFF directionally sensitive retinal ganglion cells [21, 27]. The retinal studies involved cells with a clearly identified function, and therefore allowed study of the coding consequences of this correlation/firing rate relationship. Both studies found that the stimulus-dependent correlation structure observed compared favorably to a structure in which stimulus-independent correlations were matched to their (stimulus-)averaged levels. This finding reflects a general principle articulated in other studies [17, 19], that stimulus-dependent correlations are beneficial when they serve to spread the neural response in a direction orthogonal to the signal space.
While many studies have illustrated the connection between stimulus-dependent correlation structure and coding, these have (until recently: see [21, 25, 27]) largely taken the correlation structure as given, leaving open the question of how exactly a network might produce the hypothesized correlation structure [6, 7] (see also the theoretical calculations in [21, 27]). Theoretical studies of the mechanisms that contribute to correlation distributions have largely analyzed homogeneous networks (i.e. cells are identical, aside from E/I identity) [2, 3, 28, 29], which does not allow an exploration of a correlation/firing rate relationship. Thus, how correlation coefficients can vary across a population of heterogeneously-tuned neurons is not yet well understood despite its possible implications for coding.
In this paper we investigated the relationship between correlations and firing rates in conductance-based leaky integrate-and-fire (LIF) neural network models, consisting of excitatory (E) and inhibitory (I) cells that are recurrently and randomly coupled. We introduced neural heterogeneity by allowing thresholds to vary across the population, which induced a wide range of firing rates, and explored different firing regimes by varying the strength of recurrent excitation. We found that with relatively strong excitation, pairwise correlations increased with firing rate.
In theoretical studies, this correlation-firing rate trend has been explained in feed-forward networks driven by common input [22, 30, 31]. Here we investigated whether the correlation/firing relationship in recurrent networks can be explained by this theory, but where the source of input correlations is internally generated; i.e., from overlapping projections within the recurrent network. We first adapted a network linear response theory, to decompose predicted correlations into contributions from different graph motifs, which are subgraphs which form the building blocks of complex networks [28, 32, 33].
We found that in all networks studied here, second-order motifs—and specifically inhibitory common input—were the dominant contributor to overall pairwise correlations. This allowed us to generalize theory from [22], and describe pairwise correlations in terms of a single-cell susceptibility function. Surprisingly, we found that correlations from inhibitory common input could either increase or decrease with firing rate, depending on how cells responded to fluctuations in inhibitory conductances.
We further show that a correlation-firing rate relationship has an important consequence for heterogeneous networks; it can shape low-dimensional structure in the correlation matrix. Low-dimensional structure—often modeled with a low-rank approximation to the correlation matrix—is important because it can be used to improve estimation [34] and even to reconstruct full correlation matrices from incomplete data [35–37]; such structure has been observed in experimental data [25, 38–41] but its origin is not always known. We demonstrate in our networks that when correlation co-varies with firing rate, the (E-E) correlation matrix could be accurately modeled with a low-rank approximation, and the low-rank projection in this approximation was strongly associated with firing rate. Thus we demonstrate that low-rank structure can result from recurrent activity modulated by single-cell characteristics, as well as from a global input or a top-down signal [38].
We studied asynchronous recurrent networks of leaky integrate-and-fire model neurons, and varied the strength of excitation to get different firing behaviors. We found that the covariation of correlations with firing rates—a phenomenon observed in feed-forward networks—occurs here in one firing regime, but not the other. We then found that this could be explained in terms of how single cells responded to fluctuations in inhibitory conductance. Finally, we show that when correlations covary with firing rates, the correlation matrix admits a low-rank approximation.
We performed Monte Carlo simulations of recurrent, randomly connected E/I networks, as described in Methods: Neuron model and network setup. To connect to previous literature on asynchronous spiking, we compared networks with and without single-cell variability—referred to as heterogeneous and homogeneous respectively. Heterogeneity was introduced by allowing cell threshold to vary, which induced a corresponding range of firing rates (see Methods: Neuron model and network setup for details). We first chose parameters so that the networks exhibited the classical asynchronous irregular (Asyn) regime, in which each neuron has irregular Poisson-like spiking, correlations are low, and the population power spectra are flat [42]. In Fig 1A we show raster plots from both the heterogeneous and homogeneous networks, in this regime. The heterogeneous network shows a gradient in its raster plot, because cells are ordered by decreasing firing rate. The population power spectra were flat, for both E and I cells and in both homogeneous and heterogeneous networks (Fig 1C).
When we increased excitation (by increasing both WEE and WIE, where WXY is the conductance strength from type Y to X; see Table 1 for parameter values), we observed occasional bursts of activity. However, the bursts do not occur at regular intervals and do not involve the entire population (we found excitatory bursts involved at most 25% of the population). The network is still moderately inhibition-dominated and neurons are spiking irregularly; example raster plots are shown in Fig 1B. The population power spectra (Fig 1D) are no longer flat (compare to the asynchronous regime, Fig 1C); they show local maxima around 8 Hz, but it is not a pronounced peak. We will refer to this as the strong asynchronous (SA) regime [43].
In both Fig 1C and 1D, we note that—despite the apparent differences in the distribution of spikes across the network, evident in the raster plots—both the autocorrelation functions (Fig 1C and 1D, insets) and the power spectra from the heterogeneous and homogeneous networks are very similar. Thus, we have a fair comparison to examine the role of heterogeneity, independent of other characteristics of the network.
The distribution of both excitatory and inhibitory firing rates are extremely narrow in the homogeneous network, but broad in the heterogeneous network (Fig 1E). This is expected, as each excitatory (inhibitory) cell in the homogenous network has the same uncoupled firing rate; because the number of synaptic inputs is likewise fixed, population variability in synaptic input is limited. The heterogeneous networks have a range of firing rates, which allows us to investigate the possibility of a relationship between (variable) firing rate and pairwise correlations. Population-averaged firing rates were very similar between the heterogeneous and homogeneous networks: in the asynchronous regime ⟪νE⟫ = 10.6 Hz (heterogeneous) and ⟪νE⟫ = 10.1 Hz (homogeneous), while ⟪νI⟫ = 44.3 Hz (heterogeneous) and ⟪νI⟫ = 43.5 Hz (homogeneous). In both regimes Fano factors ranged between 0.9 and 1.1, consistent with Poisson-like spiking (more statistics are given in S1 and S3 Tables).
We next sought a possible relationship between pairwise correlations—quantified via the Pearson’s correlation coefficient for spike counts, ρ i j ≡ Cov T ( n i , n j ) / Var T ( n i ) Var T ( n j )—and single-cell firing rates. Such relationships have been found in feed-forward networks [22, 30, 31], and impact information transfer when considered in concert with stimulus selectivity (i.e. signal correlations) [7, 8, 15, 19]. In heterogeneous networks, the large range of firing rates—equivalently the large range of operating points—admits the possibility that cells at different operating points may differ in their ability to transfer correlations.
To investigate this we plotted pairwise correlations for each distinct excitatory pair ρij, versus the geometric mean of the firing rates ν i ν j, in both regimes (asynchronous and strong asynchronous), for a range of time scales (blue stars in Fig 2). We focus here on excitatory-excitatory (E-E) pairs, because excitatory synaptic connections provide the predominant means of propagating cortical sensory information to higher layers. Our results show a striking difference between the two spiking regimes; while there is no clear relationship with firing rate in the asynchronous regime (Fig 2, top row), the strong asynchronous regime shows a distinct positive trend with firing rate (Fig 2, bottom row). We can quantify a hypothesized relationship between ν and ρ with linear regression, and indeed find that geometric mean firing rate explains a substantial part of the variability of correlations in the strong asynchronous regime obtained from the Monte Carlo simulations, with R2 values (i.e. percentage of variability explained) of 0.41, 0.37, and 0.34 for time windows of T = 5, 50, and 100 ms respectively (in contrast, R2 values for the asynchronous network are below 0.005).
In recurrent networks, the response of each cell is shaped by both direct and indirect connections through the network. We used the linear response theory described in Methods: Linear Response Theory and Methods: Computing statistics from linear response theory to predict the full correlation matrix CT at various time scales, including the limit of long time scales: C ˜ ( 0 ) = lim T → ∞ 1 T C T. We found that this theory successfully captured E-E correlations, both the full distribution of values and coefficients of individual cell pairs (details, including figures, can be found in: Supporting Information: S1 Text).
We then plotted the predicted correlation, C ˜ i j / C ˜ i i C ˜ j j, vs. geometric mean firing rate ν i ν j (magenta circles in Fig 2). The predicted correlations captured the same positive relationship observed in Monte Carlo results, with R2 values of 0.47, 0.4, and 0.36.
Why does a correlation/firing rate relationship emerge in one spiking regime, but not the other? In feed-forward networks, a positive correlation/firing rate relationship results from transferring common input through fluctuation-driven, asynchronously-firing cells [22, 30]. In contrast, the amount of shared input into two cells in a recurrent network is determined by both direct and indirect connections through the network. To separate the impact of different network pathways, we decomposed the linear response-predicted correlations at long time scales (i.e. C ˜ ( 0 ) = lim T → ∞ 1 T C T) into normalized contributions from n-th order motifs, as described in Methods: Quantifying the role of motifs in networks. Common input from a divergent connection, for example, results from the 2nd-order motif K*C0K. In Fig 3, we plot the summed contributions up to sixth order—i.e. R ˜ i j k, for k = 1, 2, …6—versus geometric mean firing rate, ν i ν j. The total normalized correlation, C ˜ i j / C ˜ i i C ˜ j j, is shown as well. In all cases, we plot long time scale correlations ω = 0; each distinct E-E pair is represented.
In the asynchronous regime (top panel of Fig 3A), first-order contributions (R ˜ 1) separate into three distinct “curves”, reflecting a 1-1 relationship with firing rate conditioned on first-order connectivity (no connection between i and j; one connection between i and j; bidirectional connection between i and j). Second-order contributions are overall positive while third-order contributions are overall negative (consistent with [28]); neither appear to have a relationship with firing rate. Second-order contributions are conspicuously dominant; fifth and sixth order terms are near zero.
This qualitative picture changes when we consider the strong asynchronous regime (bottom panel of Fig 3A). First-order contributions follow a similar pattern as in the asynchronous regime, and second-order contributions are likewise positive. However, third-order contributions are positive, and in the heterogeneous network they have a distinctly positive relationship with firing rate (top panel). Thus, in the asynchronous regime, negative third-order contributions partially cancel with positive second-order contributions; in the strong asynchronous regime, first, second, and third-order motifs reinforce each other, contributing to an overall positive relationship with firing rate (black dots).
Despite these differences, second-order contributions are the major determinant of total correlation in both regimes. In Fig 3B we plot the same data (R ˜ i j k) vs. total correlation, rather than geometric mean firing rate. In the asynchronous regime, second-order contributions cluster near the unity line, suggesting they are strongly predictive of total correlation. To quantify this intuition we computed the fraction of variance explained (R2) by performing a linear regression of total normalized correlation (C ˜ i j / C ˜ i i C ˜ j j) against contributions of each order (Fig 3C); in the asynchronous regime R2 values for R ˜ i j 1, R ˜ i j 2, and R ˜ i j 3 were 0.004, 0.969, and 0.0002, respectively. R2 values for higher orders were likewise small: for R ˜ i j 4, R ˜ i j 5, and R ˜ i j 6 they were 0.047, 0.034, and 0.074.
This statistic was more ambiguous in the strong asynchronous regime, where R2 values for R ˜ i j 1, R ˜ i j 2, and R ˜ i j 3 were 0.595, 0.474, and 0.509 respectively. However, note that R ˜ i j 1 and R ˜ i j 3 were positive for all cell pairs; R ˜ i j 2 and total correlation were negative for less than 0.3% of cell pairs. Thus, we considered how each motif contributed to the total correlation by taking the ratio of each contribution to the total, averaged over all cell pairs (Fig 3D). By this measure, second-order contributions were largest; fraction explained for R ˜ i j 1, R ˜ i j 2, and R ˜ i j 3 were 0.239, 0.601, and 0.420, respectively. Note that this measure cannot be used for the asynchronous (Asyn) regime because of the negative values of R ˜ i j k.
Taken together, this evidence points to a distinguished role for second-order motifs (R ˜ i j 2) in determining total correlation. In the asynchronous regime in particular, R ˜ i j 2 is a near-perfect predictor of total correlation.
We next analyzed contributions from specific second-order motifs in Fig 4. There are four distinct second-order motifs that can correlate two E cells. There are two types of chains, from K2C0 and C0 (K*)2. An E → E → E chain tends to positively correlate, while an E → I → E chain will negatively correlate; these are shown as blue and green respectively. There are two types of common input, from KC0(K*); they correspond to common input from E and I cells, i.e. E ← E → E and E ← I → E. They both lead to positive correlations and are shown as red and magenta respectively.
In the asynchronous regime (left panel of Fig 4A), the dominant contributions are I common input (magenta) and negative (E → I → E) chains (green); correlating chains (blue) and excitatory common input (red) are barely visible, as they are clustered near zero. In the strong asynchronous case (right panel), blue and red dots are now visible and show a clear 1-1 trend with firing rate. In both regimes, inhibitory common input appears to be the dominant second-order motif. In Fig 4B we plot the contribution from different second-order motifs vs. the total contribution from second-order motifs, R ˜ i j 2 (rather than geometric mean firing rate, ν i ν j). In both panels, the inhibitory common input (magenta) clusters around the unity line, showing it is the best predictor of the total second-order contribution. In Fig 4C we quantify this observation by reporting fraction of variance explained (R2) from linear regressions: the R2 value for inhibitory common input exceeds 0.8 in both networks, while the R2 values for all other motifs types are less than 0.1.
In conclusion, decomposition of pairwise correlations into graph motifs has shown us two important things: first, while third-order motifs probably contribute to the positive correlation/firing rate relationship observed in the SA regime, second-order motifs still dominate in both regimes. Second, inhibitory common input is the most important second-order motif in both regimes (Fig 4B).
In feedforward networks—i.e., in the absence of a path between two cells—correlations in outputs (i.e. spike trains) must arise from correlations in inputs; for example, through shared or common inputs. We have found that inhibitory common input is the dominant contributor to pairwise correlations in both the asynchronous and strong asynchronous regimes; we now turn our attention to modeling this term (inhibitory common input) specifically.
Previous work that analyzed the relationship between the long-time correlation and firing rate in feedforward networks [22, 30] quantified a susceptibility function that measures the ratio between output and input correlations:
S ≈ ρ c . (1)
If both cells receive a large (but equal) number of uncorrelated inputs, c would be the fraction of inputs that are common to both i and j.
In the networks examined here, each cell had a fixed in-degree for both excitatory and inhibitory cells; however, for any given pair of cells i and j, the number of E and I inputs that synapsed onto both cells will vary from pair to pair. Thus, we next considered the possibility that our (negative) finding in the asynchronous network could be explained by accounting for variable cij.
We focus on inhibitory common input, which is the dominant second-order contribution in the asynchronous network (Fig 4). We segregated pairs by whether they had 0, 1, 2, etc.. common inhibitory inputs; we then use this number as a proxy for c (recall that each excitatory cell had exactly 7 inhibitory inputs, so that this number divided by 7 approximates the common input fraction; two common inputs imply c ≈ 0.28 for example). We plot the results for the asynchronous network in Fig 5A, top panel (data for each distinct value of c is presented by color). As we might expect, correlation increases as c increases. However, for a fixed c, there is not an apparent relationship between firing rate and correlation; if anything, there appears to be a slight decrease. Correlation also increases with c in the strong asynchronous network (Fig 5A, bottom panel); however, here we also see a modest increase with geometric mean firing rate ν i ν j.
Previous theoretical work [22, 30] identified an increase in susceptibility with firing rates in current-driven neurons; we next considered the possibility that this fails to hold for conductance-driven neurons. As described in Methods: Quantifying correlation susceptibility, we estimated correlation susceptibility for each pair of neurons, by using the susceptibility function for each neuron to conductance fluctuations (computed as part of the linear response theory), divided by a measure of the long-timescale spike count variance:
S i j ⟨ g I ⟩ = A ˜ ⟨ g I ⟩ , i ( 0 ) A ˜ ⟨ g I ⟩ , j ( 0 ) C ˜ i i ( 0 ) C ˜ j j ( 0 ) (2)
We plotted the results for both networks in Fig 5B; while susceptibility increases with firing rate in the strong asynchronous network (except for the largest firing rates), it actually decreases with firing rate in the asynchronous network.
We can contrast with the estimated susceptibility to current fluctuations (i.e. Aμ,i, with μi, τeff,i, and σeff,i as in Eq 29) which we also computed for the same set of cell pairs, shown in Fig 5C.
S i j μ = A ˜ μ , i ( 0 ) A ˜ μ , j ( 0 ) C ˜ i i ( 0 ) C ˜ j j ( 0 ) (3)
Here, we see that S i j μ increases with firing rates, in both networks.
We next sought to understand how susceptibility depends on neural parameters; that is, we define the single-cell susceptibility
S i ⟨ g I ⟩ ≡ A ˜ ⟨ g I ⟩ , i ( 0 ) ν i (4)
where
ν i = f ⟨ g I , i ⟩ , σ I , i , ⟨ g E , i ⟩ , σ E , i , σ i , θ i (5) A ˜ ⟨ g I ⟩ , i ( 0 ) = ∂ f ∂ x 1 ⟨ g I , i ⟩ , σ I , i , ⟨ g E , i ⟩ , σ E , i , σ i , θ i . (6)
(“∂ ∂ x 1” indicates that derivative is taken with respect to the first argument, 〈gI, i〉). We have also used the asynchronous spiking assumption, that C ˜ i i ( 0 ) ≈ ν i (compare with Eq 3).
This quantity is shown in Fig 6, where it is plotted vs. firing rate νi (blue stars). Note that this is a negative quantity; since the susceptibility for a neuron pair S i j 〈 g I 〉 = S i 〈 g I 〉 S j 〈 g I 〉 is the product (and therefore positive), an increase in S i 〈 g I 〉 will result in a decrease in S i j 〈 g I 〉 and vice versa.
In principle, the firing rate function (Eq 5)—and therefore susceptibility—can depend on all six parameters defining the cell: our next step was to reduce the dimensionality of the problem. We first looked for any possible relationship between single-cell firing rates and cell parameters (see S1 Text: Approximating single-cell susceptibility in a heterogeneous network, S6 and S7 Figs): in both networks, only threshold θi had an obvious relationship with firing rate. Among the remaining parameters, the mean inhibitory conductance 〈gI〉 had the greatest relative range of values in the asynchronous network (S6A Fig). Therefore, we hypothesized that we could accurately capture S i 〈 g I 〉, by approximating it as a function of the two parameters θi and 〈gI〉.
We reevaluated the firing rate function, where σI,i, 〈gE,i〉, σE,i and σi have been replaced by their average values: i.e.
S ^ i ⟨ g I ⟩ ≡ 1 F ( ⟨ g I , i ⟩ , θ i ) ∂ F ∂ x 1 ⟨ g I , i ⟩ , θ i (7)
where
F ( ⟨ g I , i ⟩ , θ i ) ≡ f ⟨ g I , i ⟩ , ⟨ σ I , i ⟩ p , ⟨ ⟨ g E , i ⟩ ⟩ p , ⟨ σ E , i ⟩ p , ⟨ σ i ⟩ p , θ i (8)
and 〈 ⋅ 〉p denotes the population average. The results are also illustrated in Fig 6 (red triangles). In the asynchronous regime (Fig 6A), the results are remarkably close to the original quantities, indicating that using average parameter values has little effect; in the strong asynchronous regime (Fig 6B) the difference is larger, but the points appear to occupy the same “cloud”. However, we can now visualize the susceptibility as a function of only two parameters, and we do so in Fig 7 by evaluating S ^ i 〈 g I 〉 on a (θ, 〈gI〉) grid; the points corresponding to the actual excitatory cells in our network are illustrated in red. In both the asynchronous and strong asynchronous regimes, the red stars form a scattered cloud around the average value 〈〈gI,i〉〉p, with no obvious relationship with θi.
This fact motivated a further simplification,
S ^ ^ i ⟨ g I ⟩ ≡ 1 F ( ⟨ ⟨ g I , i ⟩ ⟩ p , θ i ) ∂ F ∂ x 1 ⟨ ⟨ g I , i ⟩ ⟩ p , θ i (9)
i.e., we replaced 〈gI,i〉 with its population average, 〈〈gI,i〉〉p, in essence approximating a one-dimensional “path” that the cells take through parameter space. The results are shown in Fig 6 (gold squares) and, as we should expect, allow us to discern a functional relationship with firing rate νi; importantly, it appears to capture the average behavior of the actual susceptibility values S i 〈 g I 〉. Here, we can see clearly that in the asynchronous regime, correlations should actually decrease with firing rate, for νi > 5 Hz. In the strong asynchronous regime, correlations will increase with firing rate, saturating around 10–15 Hz.
Finally, recall that our actual network sampled a relatively small part of the (θ, 〈gI〉) plane. This may be attributed to the fact that we generated firing rate diversity (and therefore heterogeneity), by modulating cell excitability through the cell threshold θi. How might our results have changed, if we had generated firing rate diversity through some other mechanism? In both regimes, we can increase firing rates by either decreasing 〈gI,i〉, or by decreasing θ (see S7 Fig). To explore this, we computed susceptibility values along another curve in the (θ, 〈gI〉) plane; specifically, we held θ fixed and instead varied 〈gI〉 (illustrated with black squares on Fig 7); i.e.
S ^ θ = 1 ⟨ g I ⟩ ( ⟨ g I ⟩ ) ≡ 1 G ( ⟨ g I ⟩ , θ ) ∂ G ∂ x 1 ⟨ g I ⟩ , θ | θ = 1 (10)
where
G ( ⟨ g I ⟩ , θ ) = f ⟨ g I ⟩ , ⟨ σ I , i ⟩ p , ⟨ ⟨ g E , i ⟩ ⟩ p , ⟨ σ E , i ⟩ p , ⟨ σ i ⟩ p , θ
Results are shown in Fig 6 (purple diamonds) and show a strikingly different relationship with firing rate; in the asynchronous regime, correlations should increase with firing rate for ν < 15 Hz; in the strong asynchronous regime correlations will increase with firing rate, saturating near 20 Hz.
To summarize the previous two subsections, we first defined a single-cell susceptibility function (Eq 4), which captures a linear approximation to the cell’s response to input. This quantity relies on an underlying firing rate, which is a function of all parameters that define single-cell dynamics; in this case, six. Each cell occupies a point in this six-dimensional parameter space. We found that in each network studied here, the occupied points approximately lie along a one-dimensional path through this parameter space, along which we could visualize the susceptibility. Finally, we considered the consequences of taking other paths through this parameter space: these paths can be interpreted as generating firing rate heterogeneity using other network mechanisms.
Previous work has identified low-dimensional structure in neural correlation matrices [25, 38–41]; its origin is not always known [3]. We next hypothesized that the positive correlation-firing rate relationship we observed in the strong asynchronous regime, might be reflected in low-dimensional structure in the correlation matrix. For simplicity, suppose that correlations were really represented by a function of firing rate (as in [22]): i.e. ρij = cS(νi)S(νj). Then we could represent the off-diagonal part of the correlation matrix as CT = cSST, where S is a length N vector such that Si = S(νi); that is, CT would be a rank-one matrix.
We followed the procedure outlined in Methods: Low-rank approximation to the correlation matrix to approximate each correlation matrix, CT, as the sum of a diagonal matrix and low-rank matrix:
C T ≈ C T diag + R 1 = λ I + ( σ 1 - λ ) u 1 u 1 T (11)
where λ is given in closed form by the eigenvalues of CT:
λ = λ 1 - ∑ j > 1 ( λ 1 - λ j ) 2 ∑ j > 1 λ 1 - λ j (12)
and σ1, u1 are the first singular value and singular vector of CT.
In Fig 8, we show the results from heterogeneous networks in both the asynchronous (top panel in each subfigure) and strong asynchronous (bottom panel in each subfigure) regimes. We first show CT − λI, where λ is given by Eq 12, in Fig 8A. Cells are ordered by (decreasing) firing rate. While no pattern is visible in the asynchronous state (top panel), the strong asynchronous state (bottom panel) shows larger values in the upper left corner, suggesting that correlation increases with firing rate. This is even more visible in the rank one approximation, ( σ 1 - λ ) u 1 u 1 T, shown in Fig 8B.
We now use C T diag + R 1 to approximate CT, and compare the results, cell pair-by-cell pair (Fig 8C). In the asynchronous network, the approximated correlations take on a narrow range (between 0 and 0.01, compared to between −0.015 and 0.03 for the measured coefficients) and do not show an obvious positive relationship. In the strong asynchronous regime, the range is more accurate (between 0.02 and 0.1, vs. 0.01 and 0.15 for the measured coefficients) and the points cluster around the unity line.
In Fig 8D, we plot the weight of each cell in the first singular vector, (u1)j vs. the firing rate νj. We can clearly see a positive relationship in the strong asynchronous regime (bottom panel), suggesting that the positive relationship between correlation and firing rate is related to the success of the low-rank approximation.
We simulated heterogeneous, asynchronous networks of leaky integrate-and-fire model neurons in order to investigate a possible relationship between firing rates and pairwise correlations in recurrent networks. We found that correlations can either increase or decrease with firing rates; this could be attributed to differences in how cells responded to fluctuations in inhibitory conductances. When correlations did increase with firing rates, this relationship was reflected in low-dimensional structure in the correlation matrix.
This study offers an example of a practical consequence of the difference between treating synaptic inputs as conductances rather than currents; while most synaptic currents are more accurately modeled as conductances, current-based formulations are often used for analytical and computational simplicity. Although it is known that neural models responding to currents vs. conductances differ in their response dynamics [44–46], this approach is supported by findings that steady-state firing rates are qualitatively similar in both settings (e.g. [47]). Here, we found that refined features of the steady-state firing rate surface will govern susceptibility to common input in asynchronous networks; two “cuts” through this surface may yield divergent behavior with respect to correlation susceptibility, despite yielding similar firing rates.
In other words, the relationship between pairwise correlations and firing rate will depend on the means through which firing rate diversity is achieved. In our study, we created firing rate diversity by regulating cell excitability; if we had instead varied mean inhibitory input (by varying the number of inhibitory connections) or a background excitatory current (which would model diversity in stimulus tuning from feed-forward inputs), we would likely have seen a different pattern. Finally, the recurrent network will also shape the path the cells follow through the “firing rate surface”; to generalize [22] to recurrent networks, we need to identify both how firing rates are produced and how they are shaped by the recurrent network.
Thus far, we have not directly connected the presence or absence of a positive correlation-firing rate relationship to other firing statistics (such as being in the asynchronous vs. strong asynchronous state, for example). We believe this will be a challenging question to answer; indeed, we showed earlier (in Results: Susceptibility depends on the mechanism underlying firing rate diversity) that we can construct a network with asynchronous firing, but where correlations do increase with firing rate. Therefore, our goal with this paper is to present a detailed procedure for analyzing how correlations will vary with firing rates in recurrent networks, along with a few illustrative (and nonintuitive) examples.
Finally, while the networks considered in this paper had fixed in-degrees (rather than Erdős-Rényi), this is not necessary. We have reproduced these results in Erdős-Rényi networks, in which parameters are identical to those chosen here, except that each network connection was chosen independently with a probability that depended only on E/I identity. In S9 Fig, we show correlations vs. geometric firing rates, for all excitatory pairs (as in Fig 2), confirming that correlations increased with firing rates in the strong asynchronous network, but not in the asynchronous network. Furthermore, we hypothesize that any difference between fixed in-degree and Erdős-Rényi networks will become less rather than more important with increasing network size, as the variance in synaptic inputs decreases.
Low-dimensional structure has been a common finding in many large-scale neural recordings [25, 38–41]; while the origin is not always known, it is often interpreted as arising from a global input or top-down signal. This is an interpretation that arises naturally from the technique of factor analysis, in which one seeks to explain a data vector as the sum of a random vector and the linear combination of some number of latent factors [48] (for Gaussian random variables, each latent factor can literally be interpreted as a global input with a distinct pattern of projection onto the observed variables). In our network, we found that a single latent factor was effective at capturing correlations in the strong asynchronous regime; however, this latent factor did not reflect common input (there was no global external input into the network) but rather modulation from single-cell characteristics. Thus, we identify a novel mechanism that can contribute to low-dimensional structure in neural recordings.
The networks studied here were not encoding a stimulus; correlations were generated by recurrent activity, given that each neuron had a baseline firing rate in the absence of recurrent input. However, we can readily connect this network to a stimulus coding task, in order to understand how the correlation-firing rate relationship can impact coding.
Consider a population of cells that is responsible for encoding a single scalar stimulus θ, such as movement direction or orientation of a visual stimulus, and that each cell has roughly a bell-shaped tuning curve. Furthermore, we model an incoming stimulus by modulating a stimulus-dependent background current Ii,(θ); i.e., cells which prefer the current stimulus have a higher level of current, and thus a higher firing rate, than cells which prefer an orthogonal or opposite stimulus. The network we studied here would model the response to a single stimulus θ0; that is, the firing rate diversity we observe is present because some cells are strongly tuned to the current stimulus, while others are not.
We could extend this model, by resetting background currents to model a complete set of stimuli {θ1, θ2, …θn − 1}. For each stimulus θj, correlations would show the rough firing rate dependence displayed in the strong asynchronous network, resulting in a stimulus-dependent correlation structure in which pairwise correlations vary like geometric mean firing rate. This is the structure analyzed in [21, 27]: the authors found that such a stimulus-dependent correlation code enhances information, when compared to a stimulus-independent code with the same average correlation level. Intuitively, the mean population response lives on the surface of a (hyper-)sphere in neural response space; the population encodes location on this surface. Positive correlations between similarly tuned cells produce response distributions that are stretched in the radial direction, “orthogonal” to this sphere, and thus have a minimal impact on the encoded variable.
Moreover, the mechanism that produced stimulus-dependent correlations in [21, 27] was similar to that shown here (see also [25]); common input modulated by stimulus-dependent gain factors. Here, we demonstrated how these stimulus-dependent gain factors might arise (or not) in a recurrent network. If excitation is tuned to put the network in the strong asynchronous regime, then the (stimulus-dependent) correlation structure that results will be favorable to coding. If excitation is tuned to put the network in the asynchronous regime, then correlations are overall low and not stimulus-dependent (although, given that average correlations are not matched, we do not here compare information contained within the two networks).
This work has, necessarily, focused only on a subset of network attributes that might affect firing statistics. One important feature is the frequency of higher-order graph motifs; experiments have shown that specific motifs will occur more frequently, than would be expected in an Erdős-Rényi network with fixed single-cell connection probability [49]. Theoretical work has found that in networks of integrate-and-fire neurons, an overabundance of divergent and chain motifs will lead to enhanced correlation [33] (this finding does depend on the dynamical regime; different motifs impact correlations in networks of coupled oscillators [32]). In [33], the authors use the assumption of homogeneous single-cell characteristics to find parsimonious and instructive formulae for the average correlation, and give a roadmap for how this might be generalized to heterogeneous networks. We look forward to considering the combined effect of single-cell and network heterogeneity in future work.
Another source of cell-to-cell heterogeneity is how cells respond to stimuli, as emphasized in the previous discussion [17, 20, 21, 27, 50] (see [19] for a review). Here, we did not consider a specific sensory system with tuning but rather focus on the general question of how the distribution of correlation values arise in recurrent networks. Given the previous discussion, one next step will be to investigate how correlations covary with firing rates, when cell-to-cell heterogeneity is produced by stimulus tuning in a structured network responding to a single variable (such as direction or orientation).
Finally, for numerical tractability our simulations here were performed in relatively small networks. While high average correlations have been measured in experiments [51], theoretical models of asynchronous networks have found that correlations must go to zero as the system becomes large (N → ∞) [2]. However, recent work has found that this does not have to be true, as long as spatial structure is introduced into the network [52]. We anticipate that this may carry over to other forms of heterogeneity, such as single-cell variability, and that therefore the effect we observe here persists for larger networks. We look forward to reporting on this in future work.
We considered randomly connected networks of excitatory and inhibitory neurons. Each cell was a linear integrate-and-fire model with second-order alpha-conductances, i.e. membrane voltage νi was modeled with a stochastic differential equation, as long as it remained beneath a threshold θi:
τ m d ν i d t = - ν i - g E , i ( t ) ( ν i - E E ) - g I , i ( t ) ( ν i - E I ) + σ i τ m ξ i ( t ) , (13)
When νi reaches θi, it is reset to 0 following a refractory period:
ν i ( t + τ ref ) → 0 , ν i ( t ) ≥ θ i (14)
Each neuron was driven by a Gaussian, white background noise, with magnitude σi depending only on the cell type; that is, 〈ξi(t)〉 = 0 and 〈ξi(t)ξi(t + s)〉 = δ(s). The membrane time constant, τm, and excitatory and inhibitory synaptic reversal potentials, E E and E I, are the same for every cell in the network.
Each cell responded to synaptic input through conductance terms, gE,i and gI,i, which are each governed by a pair of differential equations:
τ d , X d g X , i d t = - g X , i + g X , i ( 1 ) (15) τ r , X d g X , i ( 1 ) d t = - g X , i ( 1 ) + τ r , X α X W Y X N Y X ∑ j ∈ X , j → i ∑ k δ ( t - t j , k ) (16)
where Y = {E, I} denotes the type of cell i and X = {E, I} denotes the type of the source neuron j. Each spike is modeled as a delta-function that impacts the auxiliary variable g X , i ( 1 ); here tj,k is the k-th spike of cell j. The rise and decay time constants τr,X and τd,X and pulse amplitude αX depend only on the type of the source neuron; i.e. they are otherwise the same across the population. The parameter WYX denotes the strength of X → Y synaptic connections, which are (once given the type of source and target neurons) identical across the population. The “raw” synaptic weight (listed in Table 1) is divided by NYX, the total number of X → Y connections received by each Y-type cell.
We chose connections to be homogeneous and relatively dense, consistent with the local architecture of cortex. Connection probabilities ranged from 20%–40%, consistent with experimentally measured values [53–55]. For our baseline network state, we then chose synaptic weights so the network is moderately inhibition-dominated (αEWIE < αIWII and αEWEE < αIWEI); that is both E and I cells receive more inhibition than excitation) and shows noisy spiking consistent with the classical asynchronous state. Each neuron receives a fixed number of incoming connections, the identities of which are chosen randomly. (The specific cell ID numbers differ in the different simulations shown below.) For most of the networks we discuss here N = 100 with the 80/20 ratio typical of cortex (i.e nE = 80, nI = 20). Each excitatory cell receives NEE = 32 (40%) excitatory and NEI = 7 (35%) inhibitory connections; each inhibitory cell receives NIE = 16 (20%) and NII = 8 (40%) inhibitory projections.
In heterogeneous networks, the threshold θi varied across the population. For both excitatory and inhibitory neurons, the thresholds θi were chosen from a log-normal distribution between 0.7 and 1.4 (where the rest potential, Vr = 0). To be precise, log θi was chosen from a (truncated) normal distribution with mean - s θ 2 / 2 and standard deviation sθ. With this choice, θi has mean 1 and variance: e s θ 2 - 1. Thus we can view sθ as a measure of the level of threshold heterogeneity.
Throughout this paper, we set sθ = 0.2, which results in a wide range of firing rates compared to the homogeneous case. This was the only source of cell-to-cell heterogeneity; all other parameters were identical across the population, conditioned on neuron type (values listed in Table 2). In homogeneous networks, the threshold was the same across the population: θi = 1.
Monte Carlo simulations were performed using the stochastic forward- Euler method (Euler-Maruyama), with a time step much smaller than any time scale in the system (Δt = 0.01 ms). Each network was simulated for one second of simulation time, after an equilibration time. Then, a large number of realizations of this interval (nR = 105) were simulated. Spike counts were retained in each 1 ms window (for a total of 1000 windows) within a realization. With this large number of realizations/trials, the error bars on the resulting time-dependent firing rates were small. Therefore we emphasize that the firing rate pattern is largely driven by network connectivity; while firing is driven by random fluctuations in the background noise, any cell-to-cell variability in the trial-averaged firing rates are not an artifact of the finite number of trials.
In general, computing the response of even a single neuron to an input requires solving a complicated, nonlinear stochastic process. However, it often happens that the presence of background noise linearizes the response of the neuron, so that we can describe this response as a perturbation from a background state. This response is furthermore linear in the perturbing input and thus referred to as linear response theory [56]. The approach can be generalized to yield the dominant terms in the coupled network response, as well; we will use the theory to predict the covariance matrix of activity.
We first consider the case of a single cell: an LIF neuron responding to a mean zero current ϵXi(t)
τ m d ν i d t = - ( ν i - E L ) + E i + σ i τ m ξ i ( t ) + ϵ X i ( t ) .
(otherwise, the mean of Xi can simply be absorbed into Ei).
For a fixed input ϵXi(t), the output spike train yi(t) will be slightly different for each realization of the noise ξi(t) and initial condition νi(0). Therefore we try to work with the time-dependent firing rate, νi(t) ≡ 〈yi(t)〉, which is obtained by averaging over all realizations and initial conditions. Linear response theory proposes the ansatz that the firing rate can be described as a perturbation from a baseline rate proportional to the input ϵXi:
ν i ( t ) = ν i , 0 + ( A i * ϵ X i ) ( t ) ; (17) νi,0 is the baseline rate (when X = 0) and Ai(t) is a susceptibility function that characterizes this firing rate response up to order ϵ [22, 29, 57].
We now consider the theory for networks; here cell i responds to the spike train of cell j, yj(t), via the synaptic weight matrix W, after convolution with a synaptic filter Fj(t):
τ m d ν i d t = - ( ν i - E L ) + E i + σ i τ m ξ i ( t ) + ∑ j W i j F j * y j ( t )
In order to consider joint statistics, we need the trial-by-trial response of the cell. We first propose to approximate the response of each neuron as:
y i ( t ) ≈ y i 0 ( t ) + A i * ∑ j ( J i j * y j ) ( t ) ; (18)
that is, each input Xi has been replaced by the synaptic input, and Jij = WijFj(t) includes both the i ← j synaptic weight Wij and synaptic kernel Fj (normalized to have area 1); Ai(t) is the susceptibility function from Eq 17. In the frequency domain this becomes
y ˜ i ( ω ) = y ˜ i 0 + A ˜ i ( ω ) ∑ j J ˜ i j ( ω ) y ˜ j ( ω ) (19)
where y ˜ i = F [ y i - ν i ] is the Fourier transform of the mean-shifted process (νi is the average firing rate of cell i) and f ˜ = F [ f ] for all other quantities. In matrix form, this yields a self-consistent equation for y ˜ in terms of y ˜ 0:
I - K ˜ ( ω ) y ˜ = y ˜ 0 ⇒ y ˜ = I - K ˜ ( ω ) - 1 y ˜ 0 (20)
where K ˜ i j ( ω ) = A ˜ i ( ω ) J ˜ i j ( ω ) is the interaction matrix, in the frequency domain. The cross-spectrum is then computed
⟨ y ˜ ( ω ) y ˜ * ( ω ) ⟩ = I - K ˜ ( ω ) - 1 ⟨ y ˜ 0 ( ω ) y ˜ 0 * ( ω ) ⟩ I - K ˜ * ( ω ) - 1 (21)
To implement this calculation, we first solve for a self-consistent set of firing rates: that is, νi is the average firing rate of
τ m d ν i d t = - ( ν i - E L ) + ( E i + E [ f i ] ) + σ i τ m ξ i ( t ) (22)
where E[fi] = ∑j Wijνj.
We must then compute the power spectrum 〈 y ˜ 0 ( ω ) y ˜ 0 * ( ω ) 〉 and the susceptibility Ai(ω), which is the (first order in ϵ) response in the firing rate r i ( t ) = r i 0 + ϵ A i ( ω ) exp ( ı ω t ) in response to an input current perturbation X(t) = ϵ exp(ıωt) (here ı is used for - 1, while i denotes an index). Both can be expressed as the solution to (different) first-order boundary value problems and solved via Richardson’s threshold integration method [47, 58].
In our simulations, we used conductance-based neurons; this requires modification, compared with the simpler current-based models. We first approximate each conductance-based neuron as an effective current-based neuron with reduced time constant, following the discussion in [59]. First, separate each conductance into mean and fluctuating parts; e.g. gE,i → 〈gE,i〉 + (gE,i − 〈gE,i〉). Then we identify an effective conductance g0,i and potential μi, and treat the fluctuating part of the conductances as noise, i.e. gE,i − 〈gE,i〉→σE,i ξE,i(t):
τ m d ν i d t = - g 0 , i ( ν i - μ i ) + σ E , i ξ E , i ( t ) ( ν i - E E ) + σ I , i ξ I , i ( t ) ( ν i - E I ) + σ i 2 τ m ξ i ( t ) (23)
where
g 0 , i = 1 + ⟨ g E , i ⟩ + ⟨ g I , i ⟩ (24) μ i = E L + E i + ⟨ g E , i ⟩ E E + ⟨ g I , i ⟩ E I g 0 , i (25) σ E , i 2 = Var g E , i ( t ) = E g E , i ( t ) - ⟨ g E , i ⟩ 2 (26) σ I , i 2 = Var g I , i ( t ) = E g I , i ( t ) - ⟨ g I , i ⟩ 2 (27)
We next simplify the noise terms by writing
ν i - E E = ν i - μ i + μ i - E E (28)
and assume that the fluctuating part of the voltage, νi − μi, is mean-zero and uncorrelated with the noise terms ξE,i(t) [59]. That allows us to define an effective equation
τ eff , i d ν i d t = - ( ν i - μ i ) + σ eff , i 2 τ eff , i η eff , i ( t ) (29)
where
τ eff , i = τ m g 0 , i (30) σ eff , i 2 = σ E , i 2 ( μ i - E E ) 2 + σ I , i 2 ( μ i - E I ) 2 + σ i 2 τ m g 0 , i τ m (31)
and the fluctuating voltage, νi(t) − μi, now makes no contribution to the effective noise variance.
Finally, we consider how to model the conductance mean and variance, e.g. 〈gE,i〉 and σ E , i 2. In our simulations, we used second order α-functions: each conductance gX,i is modeled by two equations that take the form
τ r , X d g X , i ( 1 ) d t = - g X , i ( 1 ) + τ r , X α ^ X , i ∑ k δ ( t - t k ) (32) τ d , X d g X , i d t = - g X , i + g X , i ( 1 ) (33)
where X = E, I and the summation is over all type-X spikes incoming to cell i. (For notation purposes, α ^ X , i includes all factors that contribute to the pulse size in Eq 16, including synapse strength and pulse amplitude.) The time constants τr,X, τd,X may depend on synapse type; the spike jumps α ^ X , i may depend on synapse type and target cell identity. We assume that each spike train is Poisson, with a constant firing rate: i.e. each spike train is modeled as a stochastic process S(t) with
⟨ S ( t ) ⟩ = ν ⟨ S ( t ) S ( t + τ ) ⟩ - ν 2 = ν δ ( τ )
Then a straightforward but lengthy calculation shows that
⟨ g X , i ( t ) ⟩ = α ^ X , i ν X , i τ r , X (34) Var g X , i ( t ) = 1 2 α ^ X , i 2 ν X , i τ r , X τ r , X τ r , X + τ d , X (35)
where νX,i is the total rate of type-X spikes incoming to cell i.
We now describe how these considerations modify the linear response calculation. First, for the self-consistent firing rate calculation, Eq 22 is replaced by an equation with a modified time constant, conductance, and noise (Eq 29).
We next compute the susceptibility in response to parameters associated with the conductance, i.e. 〈gE,i〉 and σ E , i 2. This differs from the current-based case in two ways: first, there is voltage-dependence in the diffusion terms, which results in a different Fokker-Planck equation (and thus a different boundary value problem to be solved for the power spectrum 〈 y ˜ 0 ( ω ) y ˜ 0 * ( ω ) 〉). Second, modulating the rate of an incoming spike train will impact both the mean and variance of the input to the effective equation, Eq 23 (via μi and σX,i). Furthermore, this impact may differ for excitatory and inhibitory neurons, giving us a total of four parameters that can be varied in the effective equation. However, neither consideration presents any essential difficulty [47].
Therefore we apply Richardson’s threshold integration method directly to Eq 23:
τ m d ν i d t = - g 0 , i ( ν i - μ i ) + σ E , i ξ E , i ( t ) ( ν i - E E ) + σ I , i ξ I , i ( t ) ( ν i - E I ) + σ i 2 τ m ξ i ( t ) (36)
When we compute susceptibilities, the parameter to be varied is either a mean conductance—〈gE,i〉 → 〈gE,i〉0 + 〈gE,i〉1 exp(ıωt) or 〈gI,i〉 → 〈gI,i〉0 + 〈gI,i〉1 exp(ıωt)—or a variance—σ E , i 2 → ( σ E , i 2 ) 0 + ( σ E , i 2 ) 1 exp ( ı ω t ) or σ I , i 2 → ( σ I , i 2 ) 0 + ( σ I , i 2 ) 1 exp ( ı ω t ). Thus we have a total of four susceptibility functions A ˜ 〈 g E 〉 , i ( ω ), A ˜ 〈 g I 〉 , i ( ω ), A ˜ σ E 2 , i ( ω ), and A ˜ σ I 2 , i ( ω ). Since the Fokker-Planck equation to be solved is linear, we can compute both susceptibilities separately and then add their effects. We now have the interaction matrix:
K ˜ i j ( ω ) = A ˜ ⟨ g E ⟩ , i ( ω ) J ˜ i j ( ω ) + A ˜ σ E 2 , i ( ω ) L ˜ i j ( ω ) , j excitatory A ˜ ⟨ g I ⟩ , i ( ω ) J ˜ i j ( ω ) + A ˜ σ I 2 , i ( ω ) L ˜ i j ( ω ) , j inhibitory (37)
where L ˜ ( ω ) plays a similar role as J ˜, but for the effect of incoming spikes on the variance of conductance. Its relationship to J ˜ (either in the frequency or time domain) is given by the same simple scaling shown in Eq 35: i.e., for j excitatory,
L ˜ i j ( ω ) = J ˜ i j ( ω ) × α ^ E , i 2 × τ r , E τ r , E + τ d , E (38)
where the first factor comes from the effective spike amplitude α ^ E , i (and is the scale factor proposed in [47], Eq (64)), and the second arises from using second-order (vs. first-order) alpha-functions.
We use a modified version of the implementation given by [29] for Richardson’s threshold integration algorithm [47, 58] to compute rate νi, power 〈 y ˜ i 0 ( ω ) y ˜ i 0 * ( ω ) 〉, and the various susceptibilities (A ˜ 〈 g E 〉 , i ( ω ), A ˜ 〈 g I 〉 , i ( ω ), A ˜ σ E 2 , i ( ω ), and A ˜ σ I 2 , i ( ω )) for an LIF neuron. We validated our code using exact formulas known for the LIF [60], and qualitative results from the literature [61].
Linear response theory yields the cross spectrum of the spike train, 〈 y ˜ i ( ω ) y ˜ j * ( ω ) 〉, for each distinct pair of neurons i and j (see Eq 21). To recover a representative set of statistics, we rely on several standard formulae relating this function to other statistical quantities.
The cross correlation function, Cij(τ), measures the similarity between two processes at time lag τ, while the cross spectrum measures the similarity between two processes at frequency ω:
C i j ( τ ) ≡ ⟨ ( y i ( t ) - ν i ) ( y j ( t + τ ) - ν j ) ⟩ (39) C ˜ i j ( ω ) ≡ ⟨ y ˜ i ( ω ) y ˜ j ( ω ) ⟩ (40)
The Weiner-Khinchin theorem [56] implies that { C i j , C ˜ i j } are a Fourier transform pair: that is,
C ˜ i j ( ω ) = ∫ - ∞ ∞ C i j ( t ) e - 2 π ı ω t d t (41)
In principle, the crosscorrelation C(t) and cross-spectrum C ˜ ( ω ) matrices are functions on the real line, reflecting the fact that correlation can be measured at different time scales. In particular, for a stationary point process the covariance of spike counts over a window of length T, ni and nj, can be related to the crosscorrelation function Cij by the following formula [4]:
Cov T ( n i , n j ) = ∫ - T T C i j ( τ ) T - ∣ τ ∣ d τ (42)
The variance of spike counts over a time window of length T, ni, is likewise given by integrating the autocorrelation function Cii:
Var
T
(
n
i
)
=
∫
-
T
T
C
i
i
(
τ
)
T
-
∣
τ
∣
d
τ
(43)
It can be helpful to normalize by the time window, i.e.
Cov T ( n i , n j ) T = ∫ - T T C i j ( τ ) 1 - ∣ τ ∣ T d τ ; (44)
we can now see that for an integrable cross correlation function (and bearing in mind that the cross-spectrum is the Fourier transform of the cross correlation), that
lim T → ∞ Cov T ( n i , n j ) T = ∫ - ∞ ∞ C i j ( τ ) d τ = C ˜ i j ( 0 ) (45)
while
lim T → 0 Cov T ( n i , n j ) T 2 = 1 T ∫ - T T C i j ( τ ) 1 - ∣ τ ∣ T d τ ≈ C i j ( 0 ) (46)
Thus, we can use C ˜ i j ( 0 ) and Cij(0) as measures of long and short time correlations respectively.
Finally, the Pearson’s correlation coefficient of the spike count defined as:
ρ T , i j = Cov T ( n i , n j ) Var T ( n i ) Var T ( n j ) (47)
is a common normalized measure of noise correlation, with ρ ∈ [−1, 1]. While CovT and VarT grow linearly with T (for a Poisson process, for example), ρT,ij in general will not (although it may increase with T). In general, ρT,ij depends on the time window T; however for readability we will often suppress the T-dependence in the notation (and use ρij instead).
We next explain how we can use the results of linear response theory to give insight into the role of different paths in the network. We begin with our predicted cross-spectrum (Eqs 21 and 40) and apply a standard series expansion for the matrix inverse:
C ˜ ( ω ) = I - K ˜ ( ω ) - 1 C ˜ 0 ( ω ) I - K ˜ * ( ω ) - 1 (48) = ∑ k = 0 ∞ K ˜ ( ω ) k C ˜ 0 ( ω ) ∑ l = 0 ∞ K ˜ ( ω ) l (49) = ∑ k = 0 ∞ ∑ l = 0 ∞ K ˜ ( ω ) k C ˜ 0 ( ω ) K ˜ ( ω ) l (50)
where C ˜ 0 ( ω ) is a diagonal matrix containing the power spectra of the unperturbed processes; i.e. C ˜ i i 0 ≡ 〈 y ˜ i ( ω ) y ˜ i ( ω ) 〉. This double sum will converge as long as the spectral radius of K ˜ is less than 1 [29].
By truncating this double sum to contain terms such that k + l ≤ n, we define the nth approximation to the cross-spectrum:
C ˜ ( ω ) ≈ C ˜ n ( ω ) (51) = C ˜ 0 ( ω ) + ∑ k = 1 n ∑ l = 0 k K ˜ ( ω ) k - l C ˜ 0 ( ω ) K ˜ * ( ω ) l (52)
Each distinct term in the inner sum can be attributed to a particular undirected path of length k. Terms of the form K ˜ k C ˜ 0 and C ˜ 0 ( K ˜ * ) k account for unidirectional paths from j → i and i → j respectively; the term ( K ˜ ( ω ) ) k - l C ˜ 0 ( ω ) ( K ˜ * ( ω ) ) l captures the contribution from a cell that has a length l path onto cell j and a length k − l path onto cell i. Thus, we can use Eq 52 to decompose the correlation into contributions from different motifs ([28], see also [31, 62]).
We can also consider the contribution from all length-n paths; that is,
P ˜ n = C ˜ n ( ω ) - C ˜ n - 1 ( ω ) = ∑ l = 0 n K ˜ ( ω ) n - l C ˜ 0 ( ω ) K ˜ * ( ω ) l
If the sum in Eq 50 converges, we should expect the magnitude of contributions to decrease as n increases.
We will also show the normalized contribution from length-n paths, which we define as follows: let Λ(ω) be the diagonal matrix with Λ i i ( ω ) = C ˜ i i ( ω ). Then we define the matrix of contributions from length-n paths R ˜ n as follows:
R ˜ n ( ω ) = Λ - 1 / 2 ( ω ) P ˜ n ( ω ) Λ - 1 / 2 ( ω ) (53)
Equivalently, R ˜ i j n ( ω ) = P ˜ i j n ( ω ) / C ˜ i i ( ω ) C ˜ j j ( ω ). This effectively normalizes the cross correlation by the autocorrelation; in particular, we can use this to decompose the correlation coefficient (Eq 47) for long time windows, because lim n → ∞ ∑ k = 0 n R ˜ k ( 0 ) = lim T → ∞ ρ T , i j.
In general, we will show long-timescale correlation (e.g. C ˜ ( 0 ) or R ˜ n ( 0 )) (Eq 45); results were qualitatively similar for other timescales.
We next consider how to quantify the (linear) susceptibility of correlation to a change in parameter. Returning to Eq 17, but written in terms of the single-cell response:
y i ( t ) = y i , 0 + ( A μ , i * X μ ) ( t ) ⇒ (54) y ˜ i ( ω ) = y ˜ i , 0 ( ω ) + A ˜ μ , i ( ω ) X ˜ μ ( ω ) (55)
Here, Xμ(t) is a (possibly) time-dependent change in a parameter, such as input current or mean inhibitory conductance; yi,0 is the baseline spike train (when X = 0). Aμ,i(t) is a susceptibility function that characterizes the cell’s response (to the parameter variation) as long as Xμ(t) is small [22, 29, 57]. Following [22], the cross-spectrum of y can now be approximated as:
C ˜ i j ( ω ) ≡ ⟨ y ˜ i * y ˜ j ⟩ ≈ ⟨ y ˜ i , 0 * y ˜ j , 0 ⟩ + ⟨ A ˜ μ , i * X ˜ μ * y ˜ j , 0 ⟩ + ⟨ A ˜ μ , j X ˜ μ y ˜ i , 0 * ⟩ + A ˜ μ , i * A ˜ μ , j ⟨ X ˜ μ * X ˜ μ ⟩ (56) = A ˜ μ , i * ( ω ) A ˜ μ , j ( ω ) C ˜ μ ( ω ) (57)
where C ˜ μ ( ω ) is the spectrum of the parameter variation. The susceptibility has an appealing interpretation in the limit ω → 0, as the derivative of the classical f-I curve:
lim ω → 0 A ˜ μ , i ( ω ) = d ν i d μ (58)
where νi is the steady-state firing rate of cell i, assuming we can measure it for specific values of the parameter μ.
lim T → ∞ ρ T , i j = lim T → ∞ Cov T ( n i , n j ) Var T ( n i ) Var T ( n j ) = C ˜ i j ( 0 ) C ˜ i i ( 0 ) C ˜ j j ( 0 ) (59) ≈ A ˜ μ , i ( 0 ) A ˜ μ , j ( 0 ) C ˜ i i ( 0 ) C ˜ j j ( 0 ) C ˜ μ ( 0 ) (60)
This motivates the definition of a correlation susceptibility, which approximates the change in pairwise correlation induced by a parameter change experienced by both cells i and j:
S i j μ = A ˜ μ , i ( 0 ) A ˜ μ , j ( 0 ) C ˜ i i ( 0 ) C ˜ j j ( 0 ) (61)
If this increases with firing rate—that is, if d S i j μ d ν > 0—then correlations will also increase with firing rate.
We can further analyze this quantity by making an assumption for asynchronous spiking, that spike count variance is equal to spike count mean; i.e. Var T ( n i ) = T ν i ⇒ C ˜ i i = ν i. Then
S i j μ ≈ 1 ν i ν j A ˜ μ , i ( 0 ) A ˜ μ , j ( 0 ) = A ˜ μ , i ( 0 ) ν i A ˜ μ , j ( 0 ) ν j (62)
which motivates the definition of the single-cell quantity
S i ⟨ g I ⟩ ≡ A ˜ ⟨ g I ⟩ , i ( 0 ) ν i
In general, the firing rate depends on all single cell parameters included in Eqn.; i.e. there exists some function f such that
ν i = f ⟨ g I , i ⟩ , σ I , i , ⟨ g E , i ⟩ , σ E , i , σ i , θ i (63) A ˜ ⟨ g I ⟩ , i ( 0 ) = ∂ f ∂ x 1 ⟨ g I , i ⟩ , σ I , i , ⟨ g E , i ⟩ , σ E , i , σ i , θ i (64)
(recall that the susceptibility for ω = 0 is the derivative of the firing rate with respect to the appropriate parameter (here, mean inhibitory conductance 〈gI〉).
We consider the correlation matrix of spike counts, as measured from Monte Carlo simulations; while these are in principle related to the cross-correlation functions C(t) defined in Methods: Computing statistics from linear response theory we will use CT to denote the matrix of correlation coefficients measured for time window T; i.e.
C T i j = ρ T , i j (65)
Furthermore, we will restrict to the E-E correlations; i.e. CT will be a nE × nE matrix, with ones on the diagonal (as ρT,ii = 1).
When we examined the singular values of the E-E correlation matrices obtained from Monte Carlo simulations, we noticed a consistent trend: there was usually one large cluster with one positive outlier. This motivates the following simple idea: by subtracting off a multiple of the identity matrix, λI, we shift the cluster towards zero; consequently CT − λI is close to a rank-1 matrix. We then propose to use the sum of the two as an approximation to CT:
C
T
≈
λ I
+
(
σ
1
-
λ
)
u
1
u
1
T
.
(66)
We seek the value λ which maximizes the fraction of the Frobenius norm explained by the first singular vector: i.e. in terms of the singular values,
λ = max λ σ ˜ 1 2 ∑ j = 1 r σ ˜ j 2 (67) = max λ ( σ 1 - λ ) 2 ∑ j = 1 r ( σ j - λ ) 2 (68)
Since CT is symmetric semi-positive definite, the singular values σj are equal to the eigenvalues λj: here σ1 ≥ σ2 ≥ ⋯ ≥ σr ≥ 0 and r is the rank of CT. This has an exact solution:
λ = λ 1 - ∑ j > 1 ( λ 1 - λ j ) 2 ∑ j > 1 λ 1 - λ j (69)
Because we have subtracted a multiple of the identity matrix, none of the singular vectors will have changed. We then have
C T ≡ λ I + ( C T - λ I ) (70) = λ I + ∑ i = 1 r ( σ i - λ ) u i u i T (71)
By truncating this sum, we approximate C with a shifted low-rank matrix:
C T ≈ C T diag + R 1 ≡ λ I + ( σ 1 - λ ) u 1 u 1 T (72)
This procedure is similar to factor analysis, in which one seeks to explain a data vector as the sum of a random vector (u) and the linear combination of some number of latent factors (z) [48]:
x = Λ z + u ;
the entries of x would then have the correlation matrix Ψ + ΛΛT, where Ψ is a diagonal matrix containing the variances of u.
|
10.1371/journal.ppat.1003338 | Phosphorylation of CDK9 at Ser175 Enhances HIV Transcription and Is a Marker of Activated P-TEFb in CD4+ T Lymphocytes | The HIV transactivator protein, Tat, enhances HIV transcription by recruiting P-TEFb from the inactive 7SK snRNP complex and directing it to proviral elongation complexes. To test the hypothesis that T-cell receptor (TCR) signaling induces critical post-translational modifications leading to enhanced interactions between P-TEFb and Tat, we employed affinity purification–tandem mass spectrometry to analyze P-TEFb. TCR or phorbal ester (PMA) signaling strongly induced phosphorylation of the CDK9 kinase at Ser175. Molecular modeling studies based on the Tat/P-TEFb X-ray structure suggested that pSer175 strengthens the intermolecular interactions between CDK9 and Tat. Mutations in Ser175 confirm that this residue could mediate critical interactions with Tat and with the bromodomain protein BRD4. The S175A mutation reduced CDK9 interactions with Tat by an average of 1.7-fold, but also completely blocked CDK9 association with BRD4. The phosphomimetic S175D mutation modestly enhanced Tat association with CDK9 while causing a 2-fold disruption in BRD4 association with CDK9. Since BRD4 is unable to compete for binding to CDK9 carrying S175A, expression of CDK9 carrying the S175A mutation in latently infected cells resulted in a robust Tat-dependent reactivation of the provirus. Similarly, the stable knockdown of BRD4 led to a strong enhancement of proviral expression. Immunoprecipitation experiments show that CDK9 phosphorylated at Ser175 is excluded from the 7SK RNP complex. Immunofluorescence and flow cytometry studies carried out using a phospho-Ser175-specific antibody demonstrated that Ser175 phosphorylation occurs during TCR activation of primary resting memory CD4+ T cells together with upregulation of the Cyclin T1 regulatory subunit of P-TEFb, and Thr186 phosphorylation of CDK9. We conclude that the phosphorylation of CDK9 at Ser175 plays a critical role in altering the competitive binding of Tat and BRD4 to P-TEFb and provides an informative molecular marker for the identification of the transcriptionally active form of P-TEFb.
| The release of the transcription elongation factor P-TEFb from the 7SK RNP complex and its binding to the HIV Tat transactivator protein enables the efficient transcription of HIV proviruses. In resting memory T-cells, which carry the bulk of the latent HIV viral pool, limiting the cellular levels of P-TEFb ensures that the provirus remains silenced unless the host cell is activated. Here we demonstrate that T-cell receptor (TCR) activation induces phosphorylation of Ser175, a residue which is located at the interface between CycT1, CDK9 and Tat. Phosphorylation of Ser175 occurs on free or 7SK-dissociated P-TEFb and genetic experiments indicate that this modification enhances P-TEFb interaction with Tat resulting in Tat-dependent reactivation of HIV proviral transcription. Modification of Ser175 appears critical for controlling the competitive binding of Tat and the bromodomain protein BRD4 to P-TEFb. Activation of P-TEFb in resting T-cells thus involves both the initial assembly of the 7SK snRNP complex and the subsequent mobilization of P-TEFb by cellular signaling and Tat. Therefore, pSer175 provides an informative molecular marker for the identification of the transcriptionally active form of P-TEFb that can be used to monitor the extent of T-cell activation during therapeutic interventions aimed at virus eradication.
| HIV infections persist throughout the lifetimes of patients due to the creation of a latent viral reservoir that is refractory to both antiviral immune responses and antiretroviral therapy (ART) [1], [2], [3], [4], [5]. Genetic and biochemical evidence strongly suggests that the major latent viral reservoir comprises a small population of resting memory CD4+ T-cells (∼1 in 106 cells) [6], [7], [8] that are created when effector T-cells acquire a Go resting memory phenotype [9] or when resting memory T-cells become infected [10]. Interruption of ART invariably leads to a rebound of virus production, even in patients that have been suppressed to below detectable levels of viremia for decades [11], [12], [13], [14], [15], [16]. The need to develop novel therapeutic tools to attack the latently infected population is now a widely recognized goal [1], [5], but implementation of this will require both a more detailed understanding of mechanisms underlying proviral latency and the creation of improved analytical tools to monitor the state of the latent proviral reservoir [3], [4], [17].
A key feature that distinguishes HIV transcription from cellular gene transcription and permits efficient entry of proviruses into latency is that it is auto-regulated by the regulatory protein Tat (for reviews see [18], [19], [20], [21], [22]). Because of this feedback mechanism a disproportionate decline in HIV transcription ensues when Tat levels become restricted due to small changes in the efficiency of transcriptional initiation, typically initiated by epigenetic changes to the chromatin structure at the HIV LTR. Epigenetic restriction of the initiation of HIV-1 transcription has been documented in transformed cells [23], [24], [25], [26], ex vivo primary cell models for HIV-1 latency [27], [28], [29] and latently infected cells obtained from patients [30]. Additional blocks to HIV-1 transcription initiation found in resting CD4+ T-cells and transformed T-cell lines, such as Jurkat cells, include the sequestration of the transcription initiation factors NF-κB [31], [32] and NFAT [33], [34] in the cytoplasm.
In addition to these blocks to transcriptional initiation, Tat-dependent transcriptional elongation in latently infected CD4+ T cells is restricted by limiting the availability of the cellular elongation co-factor P-TEFb [35], [36], [37], [38]. P-TEFb is a heterodimer of the CDK9 serine/threonine kinase and a C-type regulatory cyclin, Cyclin T1 (CycT1). Human Cyclin T1 (hCycT1) binds directly to Tat and enhances the co-operative binding of P-TEFb/Tat to TAR RNA by binding to its apical loop [39], [40], [41]. P-TEFb stimulates HIV transcription elongation by phosphorylating a variety of positive and negative factors. Latent HIV proviruses typically carry promoter-proximally paused RNAP II complexes. Hyperphosphorylation of serine residues of the heptad repeats at the CTD of RNAP II by Tat-stimulated P-TEFb enhances its processivity [42], [43], [44], [45]. In addition, phosphorylation of the E-subunit of the negative elongation factor NELF by P-TEFb forces its dissociation from paused RNAP II complexes and allows resumption of productive elongation [46]. Similarly, phosphorylation of the C-terminal region of the SPT5 subunit of DSIF by P-TEFb transforms it into a positive elongation factor [47], [48]. It has also been recently discovered that in addition to stimulating P-TEFb recruitment to the promoter, Tat also mediates the recruitment of a large “super elongation complex” containing numerous additional elongation factors [49], [50], [51], [52].
Despite these overlapping restrictions, T-cell receptor (TCR) stimulation of latently infected T- cells provides all the intracellular signals that are necessary to reactivate productive HIV transcription and is generally regarded as the most efficient way to reactivate latent proviruses [27], [33], [53]. Within minutes of TCR stimulation there is an influx of the transcription initiation factors NF-κB and NFAT into the nucleus [24], [27], [32], [53]. Binding of these transcription factors to the HIV LTR activates chromatin modifying and remodeling events that reverse the epigenetic restrictions associated with proviral latency and the release of paused RNAP II located in the vicinity of the transactivation response element, TAR [24], [53], [54], [55], [56]. Initiation of new rounds of transcription and promoter clearance by RNAP II is then triggered by the recruitment of TFIIH [31], [57] which allows for additional accumulation of paused RNAP II bound by the negative elongation factors NELF and DSIF [58], [59], [60], [61]. The early rounds of transcription lead to the synthesis of additional Tat and a switch to constitutive proviral transcription where RNAP II promoter proximal pausing is continuously overcome by Tat leading to a 100- to 1000-fold increase in the production of viral transcripts.
TCR activation simultaneously stimulates P-TEFb activity by regulating the 7SK snRNP complex. 7SK RNA is a highly conserved 331-nucleotide RNA polymerase III transcript that serves as a platform for the binding of two P-TEFb molecules [62], [63] together with a dimer of the inhibitory proteins HEXIM1 and/or HEXIM2 which maintain CDK9 in a catalytically inactive state [64], [65], [66]. The 7SK snRNP complex also contains the 5′-end capping enzyme MEPCE and the 3′-end uridine-rich binding protein LARP7 that are believed to protect the ends of the RNA from exonuclease cleavage [67], [68].
The molecular mechanisms regulating the binding and release of P-TEFb from the 7SK snRNP are still poorly understood. Incorporation of P-TEFb into 7SK snRNP requires the phosphorylation of CDK9 at Thr186 of its activation loop [69], [70], [71]. This modification is also essential for activating the enzyme's kinase activity [69], [70], [71]. In resting CD4+ T lymphocytes CycT1 expression is highly restricted [27], [35], [36], [72] due to translational blocks mediated by miRNA [38], severely limiting the levels of functional P-TEFb in these cells. Activation of CD4+ T lymphocytes through the TCR is sufficient to trigger the expression of hCycT1 [36], [73] and the concomitant phosphorylation of Thr186 permitting the assembly of the 7SK snRNP complex.
Tat is able to extract P-TEFb from the 7SK complex by displacing HEXIM1 and inducing conformation changes in the 7SK RNA [74], [75], [76], [77]. When Tat levels are very low, such as during the reactivation of latent proviruses, activation of HIV transcription elongation might be enhanced by a Tat-independent activation signal that triggers the disassembly of 7SK snRNP. Consistent with this hypothesis we recently found that in Jurkat T-cells, TCR activation induces the rapid release of P-TEFb from the nuclear 7SK snRNP complex and enhances its recruitment to the HIV long terminal repeat (LTR) [53]. Fujinaga et al. [78] have recently confirmed that TCR signaling is a potent trigger for P-TEFb dissociation from 7SK RNP. They also found that during TCR mediated disruption of the 7SK RNP complex protein kinase C (PKC) phosphorylates HEXIM1 at Ser158. The Ser158 phosphorylated HEXIM1 protein is unable to bind to 7SK snRNA and is therefore unable to inhibit P-TEFb.
In the experiments to be described we used affinity purification-tandem mass spectrometry (AP-MS/MS) to define key post-translational modifications found on P-TEFb subunits in response to TCR signaling. Our results show that phosphorylation of CDK9 at Ser175 [79] is rapidly induced by TCR signals in memory CD4+ T cells. Ser175 phosphorylation occurs subsequent to the dissociation of P-TEFb from 7SK snRNP and this modification plays an important role in controlling the competitive interaction of CDK9 with Tat and the bromodomain protein BRD4, which recruits P-TEFb to cellular genes. These observations are consistent with previous studies showing that certain mutations in Ser175 can reduce P-TEFb interactions with BRD4 [80]. Additionally, using an antibody specific for pSer175 we have shown that phosphorylation of CDK9 at Ser175 provides a sensitive molecular marker for the transcriptionally active form of P-TEFb in primary CD4+ T-cells.
In latently infected Jurkat T-cells, TCR-induced dissociation of the 7SK snRNP complex coincides with enhanced Tat-dependent P-TEFb recruitment to the HIV LTR and the stimulation of proviral transcription elongation [53]. These effects are extremely rapid; within 30 min of engagement of the TCR with α-CD3 and α-CD28 antibodies, or exposure of cells to the protein kinase C (PKC) agonist, phorbol 12-myristate 13-acetate (PMA), 7SK snRNP complex dissociation was induced as measured by gel filtration chromatography or density gradient centrifugation [53], [78]. Both 7SK snRNP complex disruption and HIV transcription elongation could be partially blocked by U0126, an inhibitor of the MAPK/ERK pathway, suggesting that these kinases participate in the signaling cascade leading to P-TEFb activation [53]. To determine whether post-translational modifications are associated with the dissociation of P-TEFb from 7SK snRNP and its enhanced interaction with Tat we used an affinity purification–tandem mass spectrometry approach.
As shown in Fig. 1A, P-TEFb complexes were isolated from Jurkat T-cells expressing FLAG-CDK9 isoform 1 from a MSCV-based retroviral vector both before and shortly after stimulation with PMA. Gel electrophoresis of the purified complexes revealed a high degree of homogeneity in the samples, with several of the most abundant bands corresponding to CDK9, CycT1 and HEXIM1. Analysis of the individual bands by a nano LC-MS/MS method identified all known 7SK snRNP protein components each with sequence coverage between 42% and 80% of the proteins (Fig. 1A). As shown in Tables 1 to 5, CycT1, CDK9 isoforms 1 and 2, HEXIM1 and HEXIM2 each carry extensive known and novel post-translational modifications (PTMs) including numerous phosphorylation, acetylation and methylation sites that have not been previously identified.
The CDK9 isoform 1 (Table 1) was consistently identified with >75% sequence coverage that allowed for a near-complete mapping of its PTMs under both basal and activated conditions (Fig. S1 provides a representative listing of the identified peptides from CDK9 and a Mascot protein database search/analysis). The most pronounced increase in the phosphorylation of CDK9 following PMA treatment of cells was phosphorylation of Ser175 (pSer175), a highly conserved residue located in the activation loop of the kinase (Fig. 2). Quantitative MS data from a representative experiment are shown in Fig. 1C and Fig. S2. Phosphorylation in three separate experiments was enhanced 14.3±1.7 fold over basal levels after 1 h PMA stimulation. U0126, partially blocked, but did not eliminate pSer175 formation, suggesting that this modification is not directly the result of ERK kinase activity. Similar results were obtained following TCR stimulation of cells which led to a 2-fold increase in pSer175 levels (Table 1). While this work was in progress, Ammosova et al. [79] also identified Ser175 as an in vivo phosphorylation site. Their identification was based on Hunter peptide mapping of in vivo -labeled 32P-peptides followed by confirmation of the peptide sequence using mass spectroscopy.
A second major modification of CDK9 was the acetylation and methylation of Lys127 (Table 1). This residue is located on an external surface of CDK9 away from the T-loop and CycT1 interaction interface.
In contrast to phosphorylation at Ser175, the phosphorylation of the activation loop residue Thr186 (pThr186), which is essential for CDK9 enzymatic activity [36], [37], [73], did not change significantly after PMA stimulation. Moreover, while both the modified and unmodified precursor peptides for Ser175 could be readily identified by MS/MS (Fig. 1B), only the pThr186 precursor peptide could be identified under basal and stimulated conditions. These observations are consistent with published reports demonstrating that in cells with steady state levels of P-TEFb, CDK9 is constitutively phosphorylated at Thr186 and that pThr186 is not only important for promoting P-TEFb kinase activity but it is also required for P-TEFb to be sequestered into the 7SK snRNP complex [36], [37], [73].
In order to obtain unambiguous, high resolution MS/MS data as cross-verification of the Ser175 phosphorylation event detected in the Velos ion trap (IT) detector, commercially synthesized AFSLAK peptides with and without phosphorylation at Ser175 (ThermoFisher Scientific) were used as controls and subjected to high resolution CID MS/MS data collection (Fig. S3). The data showed that the major daughter ions observed in the ion trap (IT) detector (Fig. 1B) can be detected with higher accuracy in the FT detector. Due to the much higher precursor ion abundance obtained from the synthetic peptide samples, some additional fragment masses that were undetected in the IT experiment were also seen in the FT MS/MS spectra and provided further verification for the tandem MS-based detection and confirmation of the pSer175 modification.
Virtually all the modifications we found on the other 7SK RNP components (CDK9 isoform 2 (Table 2), CycT1 (Table 3), HEXIM1 (Table 4) and HEXIM2 (Table 5)) have not been previously described. An important exception is that in agreement with Cho et al. [81] we have found acetylation of CycT1 on K386, a region of the molecule that is predicted to form a coiled-coil structure. Acetylation of CycT1 has been associated with shifting the the balance between Hexim1-bound (inactive) and Hexim1-free (active) P-TEFb [81].
It should be noted that the identified PTMs in Table 1 are based on data obtained in Jurkat T-cells which have assembled 7SK RNP complexes. Primary resting memory CD4+ T-cells are severely restricted in CycT1 and therefore have only miminal levels of 7SK RNP complexes, it is therefore expected that many of these reported modifications are no found in resting memory T-cells, as is the case for pThr186. We therefore expect that the majority of the modifications detected in Jurkat T-cells will only be present in activated T-cells.
In order to facilitate the detection of CDK9 Ser175 phosphorylation, we developed a rabbit polyclonal antibody against a peptide containing pSer175. The crude antiserum was initially affinity purified using the phospho-serine peptide and the eluted antibody fractions were further purified by multiple rounds of negative selection with affinity resin for the corresponding unmodified peptide.
Wildtype CDK9 and a series of mutations in Ser175 and Thr186 were stably expressed as FLAG-tagged proteins in Jurkat cells using the MSCV-retroviral expression vector. As shown in Fig. 2A, when whole cell extracts from these cells were blotted using α-CDK9 antibody, both the endogenous CDK9 and the FLAG-tagged proteins, which migrate slightly slower than the native CDK9, were detected. Ectopic expression of the FLAG-tagged CDK9 protein reduced endogenous CDK9 expression by approximately 2-fold and resulted in a ratio of FLAG-CDK9 to CDK9 expression of between 2.0 and 3.0. Consequently, the majority of the CDK9 in cells can be recovered in the FLAG-immunoprecipitates.
The FLAG-tagged proteins were immunoprecipitated with α-FLAG antibody and then blotted using α-CDK9, α-pThr186 and α-pSer175 CDK9 antibodies. The blots using the α-pSer175 CDK9 antibody showed a 10.86-,27.49-, and 111.6-fold increase in pSer175 levels following 1 hr PMA stimulation in three different experiments (49.98±31.2) (Fig. 2A), consistent with the increases observed in the MS/MS experiments (Fig. 1). Further evidence that the antibody we developed is highly selective for pSer175, comes from its failure to react against CDK9 carrying the S175A or S175D mutations after 1 hr PMA stimulation. As an additional control, peptide blocking experiments performed using the phosphorylated peptide showed complete inhibition of antibody binding to the phosphorylated form of CDK9 (Fig. 2B).
To evaluate the impact of mutations in the T-loop on the assembly of the 7SK snRNP complex we performed Western blotting experiments using 7SK snRNP complexes recovered by affinity chromatography of FLAG-tagged CDK9. As shown in Fig. 3, the S175A and S175D mutations of CDK9 did not affect the assembly of the 7SK snRNP complex and permitted efficient co-purification of CDK9, CycT1 and the 7SK snRNP components HEXIM1 and LARP7. For the S175A mutation each of the components was immunoprecipitated with >90% efficiency. Compared to the wildtype CDK9 and the S175A mutant, co-IP of S175D showed somewhat reduced levels of the 7SK RNP complex components (e.g. 77.8%±2.9% CycT1, 67.5%±3.3% HEXIM1 and 77.2%±6.6% LARP7). However, we believe that the association of S175D with the 7SK complex is equivalent to the wildtype since this mutant was expressed at 1.5 to 2.0-fold higher levels than the wildtype CDK9 and the excess CDK9 may be found in association with chaperone proteins [82].
The CDK9 in these complexes was phosphorylated on T186, as demonstrated by Western blotting using a pT186-specific antibody [36] (Fig. 2A). Phosphorylation of CDK9 at Thr186 is not only essential for enabling its kinase activity but is also required for CDK9/hCycT1 P-TEFb to become assembled into the catalytically inactive 7SK snRNP [36], [37], [69], [73]. As expected, mutations in Thr186 prevented assembly into the 7SK snRNP complex and therefore co-precipitation of HEXIM1 and LARP7 was reduced by over 99% (Fig. 3A). The mutations in Thr186 limited, but did not abolish, CDK9 binding to CycT1. The T186A mutation strongly inhibited CDK9 binding to CycT1 (8.5±1.6% of the wildtype), whereas the T186D mutation only reduced binding to CycT1 to 50.5 5±4.2% (Fig. 3B).
Our observation that T186A fails to bind to CycT1 disagrees with previous observations by Yang et al. [10] and Li et al. [9]. A possible molecular explanation for why mutation of Thr186 compromises CDK9 interaction with Cyclin T1 is based on the study by Russo et al. [71] of who clearly demonstrated that in the analgous CDK2-Cyclin A structure the phosphate moiety at the highly conserved CDK T-loop T186 residue serves as an organizing center to coordinate a network of intramolecular and intermolecular hydrogen bonding interactions that stabilize heterodimerization with the Cyclin T1 subunit.
In summary, the preceding immunoprecipitation experiments demonstrate unequivocally that the phosphorylation of CDK9 at Ser175 is not required for the assembly of P-TEFb/7SK snRNP.
To determine whether CDK9 can be phosphorylated at Ser175 while assembled within 7SK snRNP, we also carried out an affinity isolation of the 7SK complex from Jurkat T-cells stably expressing FLAG HEXIM1 (Fig. 4). In contrast to FLAG-CDK9, FLAG-HEXIM1 is expressed at approximately 3.5-fold higher levels than the endogenous protein in these cell lines. The 7SK RNP complex components (CDK9, CycT1 and LARP7) that co-purified with HEXIM1 under these conditions were analyzed by Western blotting before and after treatment of the cells with PMA. Our previous gel filtration chromatography studies demonstrated that activation of MAPK/ERK by brief stimulation of Jurkat T-cells with PMA induces partial 7SK complex dissociation and release of P-TEFb [53]. Since HEXIM1 only associates with P-TEFb as part of the 7SK complex, any CDK9 that precipitates with HEXIM1 is present in the 7SK complex, while CDK9 that does co-precipitate represents released P-TEFb. As shown in Fig. 4, PMA stimulation for 1 h reduced HEXIM1 association with CDK9, Cyclin T1, and LARP7 by 30%, 31%, and 23% respectively, indicating that approximately 30% of the 7SK complexes were disrupted. While we could easily detect endogenous CDK9 in the HEXIM1 immunoprecipitates, it is important to note that pSer175 could not be detected within these complexes even after PMA stimulation when pSer175 levels are readily detected in the whole cell extracts (Fig. 4). Thus, CDK9 carrying pSer175 appears to be excluded from the 7SK snRNP complex, and is likely to represent a transcriptionally active form of the P-TEFb enzyme.
As shown in Fig. 5A, Jurkat T-cells latently infected with HIV express subthreshold levels of Tat that are undetectable by Western blotting [53]. Following stimulation of the cells by α-CD3 and α-CD28 antibodies, PMA, or TNF-α, Tat levels progressively increase reaching maximum levels by 24 hrs.
We took advantage of this induction system to examine the extent to which the phosphorylation of CDK9 at Ser175 may alter the affinity of P-TEFb for Tat and BRD4. Latently infected Jurkat T-cells engineered to stably express FLAG-tagged versions of CDK9 wildtype, S175A, S175D, T186A, or T186D in trans were induced by 16 hr treatment with TNF-α prior to the affinity isolation of P-TEFb complexes by anti-FLAG IP (Fig. 5B). After normalizing to the corresponding CDK9 levels, 59.2±5.9% of Tat remained associated with S175A CDK9 compared to wildtype (Fig. 5B). By contrast the S175D phosphomimetic mutation modestly increased the association of CDK9 with Tat (109.2%±11.9%).
Consistent with the results in Fig. 3, both the S175A and S175D mutants bound CycT1 with nearly wildtype efficiency (S175A: 98.9±5.7%; S175D: 78.9±1.9%) (Fig. 5B). However, it is important to note that the two mutations had disproportionate effects on BRD4 binding. The S175A mutation significantly reduced BRD4 binding to 1.9±1.7% whereas the S175D mutation bound BRD4 with 51.1±19.3% of the wildtype efficiency. There was no significant difference between the CDK9 association profiles of wildtype Tat and the H13L Tat in these experiments.
As shown in Fig. 5B, the T186A mutation severely reduced binding to CycT1 (8.0±2.5%) and consequently abrogated Tat interaction with CDK9 (0.3%±0.3%). The phosphomimetic mutation T186D showed somewhat reduced BRD4 binding (61.6±14.7% of wildtype) and a similarly reduced level of CycT1 binding (49.4±8.5%). However Tat binding to T186D CDK9 was severely restricted (7.0%±7.0%). Thus, both T186 mutations are likely to alter the conformation of the activation loop in ways that interfere with potential intermolecular electrostatic interactions between CycT1, BRD4 and the N-terminus of Tat.
To study the impact of Tat expression on 7SK snRNP complex dissociation, and to evaluate whether Ser175 phosphorylation could enhance Tat-dependent disruption of the 7SK RNP complex, we also employed an immunoprecipitation strategy. For these experiments, 239T cells stably expressing FLAG-tagged CDK9 were transiently transfected with HA-tagged Tat since only trace amounts of HA-Tat could be constitutively expressed in Jurkat cells from retroviral vectors. Phosphorylation of Ser175 was induced by treatment of the cells for 1 hr or 2 hrs with PMA. As shown in Fig. 5C, ectopic expression of HA-Tat caused disruption of the 7SK snRNP complex and resulted in release of 71% of the HEXIM1 and 82% of the LARP7 from the affinity purified P-TEFb fractions. Stimulation of cells with PMA increased the efficiency of the disruption of the 7SK snRNP complex by Tat, with the result that after 2 hrs of treatment with PMA of 2.4% of the HEXIM1 and 0.04% of the LARP7 copurified with the P-TEFb complex.
As, expected the HA-Tat copurified with the released P-TEFb. After normalization for CDK9 levels there was a moderate 12% increase in the amount of Tat associated with CDK9 after 1 hr of PMA treatment. These results are consistent with our hypothesis that Ser175 phosphorylation, while not obligatory for Tat binding, can enhance the interactions between P-TEFb and Tat.
Previous reports have suggested that phosphorylation of Ser175 might be due to CDK9 autophosphorylation [79]. Since Thr186 is essential for the kinase activity of CDK9 [66], [69], [71], we decided to test whether T186A and T186D could become phosphorylated at Ser175. In the experiment shown in Fig. 2A, Ser175 phosphorylation increased 37- and 32-fold, respectively for the T186A and T186D mutations after 1 hr PMA treatment of cells. Thus, neither the assembly into the 7SK RNP complex, nor the enzymatic activity of CDK9 is required for S175 phosphorylation.
We also performed in vitro kinase assays using affinity purified P-TEFb in the absence (Fig. 6A) and presence of a RNAP II CTD peptide (Fig. 6B). For these experiments, Tat was induced by overnight treatment of latently infected Jurkat cells with TNF-α prior to purification of complexes carrying P-TEFb. Immunoblots of these samples are shown in Fig. 5B (lower panel).
P-TEFb carrying wildtype CDK9 showed high enzymatic activity in both assays and was used to normalize the data. The kinase activity was also strongly inhibited by 100 nM flavopiridol, a potent CDK9 inhibitor, demonstrating that it is unlikely to be due to contaminating kinases.
The S175A mutation had wildtype activity (90.1% and 98.0% in the autophosphorylation and CTD kinase assays). By contrast, the S175D phosphomimetic mutation increased CDK9 kinase activity to 117% in the autophosphorylation assay and 158.7% in the CTD assay, consistent with a stabilization of the Tat-P-TEFb interface by this mutation. In partial disagreement with our results, Yang et al. [80] have reported that the Ser175Ala (S175A) mutation inactivates P-TEFb catalytic activity as assessed by in vitro kinase assays while the S175D had wildtype activity. Using an in vitro transcription assay they were able also to demonstrate that S175D CDK9/hCycT1 can mediate Tat-dependent full length transcription from the HIV LTR [80].
Since the S175A and S175D, sequences cannot be phosphorylated on Ser175, the CDK9 autophosphorylation events detected in these assays are clearly occurring on other sites, most likely in the C-terminal region [83]. As expected the T186A and T186D mutations, which fail to assemble properly with CycT1 severely inhibited P-TEFb kinase activity and reduced kinase activity to less than 2% of the wildtype.
We conclude that the signal-dependent phosphorylation of Ser175 is independent of CDK9 autophosphorylation and can occur only on P-TEFb that has dissociated from the 7SK snRNP complex.
Since mutations in Ser175 have an impact on the binding of P-TEFb to both Tat and BRD4 we decided to evaluate whether expression in trans of CDK9 carrying mutations in Ser175 and Thr186 had an effect on latent HIV proviral expression. For these experiments we compared cells carrying proviruses with H13L Tat (2D10), wildtype Tat (E4) and the inactive C22G Tat (2B2D). The proviruses in these experiments also carried a d2EGFP fluorescent marker permitting measurements of proviral expression by flow cytometry.
Essentially identical results were obtained with both the H13L Tat and wildtype Tat proviruses. As shown in Fig. 7A, the stable expression of S175D CDK9 in latently infected Jurkat E4 cells induced significant basal HIV proviral gene expression (Wildtype CDK9, 5.12±0.06% (E4); S175D 13.66±0.79% (E4)). This is consistent with the idea that introduction of a negative charge at position 175, either through phosphorylation, or because of the phosphomimetic mutation, enhances CDK9 binding to Tat. Similarly, in our earlier study, we showed that both PMA and TCR activation enhanced HIV proviral transcription [53].
We were surprised to observe that expression of the S175A mutation more potently reactivated latent HIV proviral gene expression than the S175D mutation (Wildtype CDK9, 5.12±0.06% (E4); S175A 26.28±2.70% (E4)) (Fig. 7B). Ammosova et al. [79], also observed that expression of S175A is able to activate HIV transcription.
Measurable induction of latent proviral gene expression by the S175A and S175D CDK9 mutants required the presence of functional Tat protein. There was no induction of HIV proviral expression above control levels in cells carrying the inactive C22G Tat (Fig. 7B).
Since we have observed that S175A somewhat reduces CDK9 association with Tat, and we and Yang et al. [80] have found that this mutation nearly completely abolished its association with BRD4 [80], [84], it seems likely that the proviral reactivation induced by expression of CDK9 carrying the S175A mutation results from altering competition between Tat and BRD4 for P-TEFb. Consistent with this idea, and in agreement with several recent publications [85], [86], [87], the stable knockdown of BRD4 expression by shRNA also led to a robust reactivation of latent proviral gene expression (Fig. 7B). Proviral reactivation due to the knockdown of BRD4 expression was Tat-dependent and proviral expression was not detected following BRD4 knockdown in latently infected cells carrying the inactive C22G Tat (Fig. 7B).
The recently published Tat/P-TEFb X-ray structure by Tahirov et al. demonstrated that though Tat forms the majority of its contacts with hCycT1, it may also form electrostatic interactions with the activation loop of CDK9 [88]. Two hydrogen bonds are postulated to be formed between Pro182 and Asn183 of CDK9 and Lys12 and Trp11 of Tat, respectively. These Tat/CDK9 contacts are thought to have a role in stabilizing an N-terminal α-helical fold in Tat that contains both Trp11 and Lys12 [88]. Ser175 is unmodified in the X-ray structure and is situated within the activation loop about 6.7 Å away from Lys12 of Tat (Fig. 8A).
To examine whether modification of Ser175 could be accommodated in the Tat/P-TEFb structure, and assess whether it can potentially make a contribution to Tat/CDK9 intermolecular interactions we modeled and energy minimized a wide range of sequence variations (Fig. 8). The modeling revealed that pSer175 is able to form an intermolecular hydrogen bond with Tat Lys12, bringing the phosphate to 2.6 Å of the amino group of Lys12 (Fig. 8A). Similar results were obtained when the modeling was performed using Tat carrying the H13L mutation (Fig. 8B). In both cases the phosphate fits nicely into a “pocket” in the structure and can be accommodated with no significant change in the conformation of the CDK9 activation loop backbone (Fig. 8C). Lys 12 is present in the consensus clade B sequence. However, Asn12 is the consensus residue in each of the non-clade B Tat sequences [89]. The modeling suggests that Asn12 is also easily accommodated in the structure and is also able to form a hydrogen bond with pSer175 (Fig. 8D).
Additional modeling was performed using mutations in Ser175. Modeling of these substitutions in CDK9 suggested that S175A induces a slight change in the Tat backbone structure but does not lead to disruption of the interactions between Tat Trp11 and CDK9 Asn183, or the interactions between Tat Lys12 and CDK9 Pro182 (Fig. 8E). Modeling of the phosphomimetic substitutions S175D and S175E suggests that they are both able to form hydrogen bonds with Tat Lys12 (Fig. 8F).
In summary, consistent with the experimental data described above, the modeling studies suggest that Ser175 phosphorylation and the S175A and S175D mutations are easily accommodated in the P-TEFb-Tat structure and that S175 phosphorylation potentially stabilizes the interactions between Tat and CDK9.
In primary resting memory CD4+ T cells the transcriptional activity of CDK9 is restricted due to the absence of its major regulatory partner hCycT1 and lack of Thr186 phosphorylation [36], [37], [38]. The activation of P-TEFb in resting memory CD4+ T cells may therefore require multiple sequential events that involve induction of hCycT1 expression, formation of P-TEFb and Thr186 phosphorylation of CDK9, initial assembly of P-TEFb into 7SK snRNP, and release of P-TEFb from the inactive 7SK complex and its eventual mobilization toward RNAP II transcribed genes (see Discussion).
Support for this model comes from studies of the subcellular localization of hCycT1, pSer175 CDK9, and pThr186 CDK9 in resting and activated CD4+ T cells by immunofluorescence microscopy (Fig. 9). In unstimulated resting memory T-cells hCycT1 levels are severely restricted and CDK9 is largely restricted to the cytoplasm (Fig. 9A). and pSer175 levels are undetectable (Fig. 9B).
Upon activation of the cells through the TCR there is upregulation of hCycT1 and concomitant increases in CDK9 and pSer175 CDK9 levels. In a recent publication, Budhiraja et al. [72] also observed upregulation of CDK9 and pThr186 upon activation of resting T-cells. We found that after TCR activation, hCycT1, pSer175 CDK9, colocalize with each other and with a marker of nuclear speckles and exhibited a punctate nucleoplasmic staining (Fig. 9B). These observations are consistent with those of Dow et al. [73] who have proposed that nuclear speckles are regions of active cellular gene transcription and sites of P-TEFb/7SK snRNP localization. The confinement of P-TEFb within nuclear speckles would therefore allow activated P-TEFb to be conveniently transferred from the 7SK RNP complex to transcribed genes in response to the appropriate extracellular signals of T cell activation. As shown in Fig. S4A, the immunofluorescent signal to pS175 is specific for this modified form of CDK9 since it can be effectively blocked by the pS175 phosphopeptide.
To determine whether cells expressing high levels of pSer175 CDK9 also carried T-cell activation markers, memory CD4+ T cells were purified from healthy donor peripheral blood using negative bead selection and stimulated for 16 hr anti-CD3 and anti-CD28 antibodies to activate the TCR. These cells were analyzed by flow cytometry after staining with fluorophore-conjugated antibodies against pSer175 CDK9 and hCycT1 (Fig. 10). As expected, after stimulation through the TCR 76.22% of the resting cell population shifted to an activated memory CD4+ T cell phenotype (CD25+CD69+) (Fig. 10B). More than 95.5% of the activated cells showed significantly elevated expression of pSer175 CDK9 and hCycT1 (Fig. 10B). As shown in Fig. S4B the pSer175 peptide effectively blocked the pSer175 signal in the flow cytometry assay, again demonstrating the specificity of the antibody binding.
The alterations in P-TEFb levels following TCR exposure are extremely rapid. As shown in Fig. 11 and Figs. S5 to S10, the flow cytometric assay described above was used to monitor the kinetics of P-TEFb activation in resting memory T-cells. As shown in Fig. 11A, increases in CycT1 levels and pSer175 CDK can be detected as early as 30 min after stimulation through the TCR. A detailed kinetic analysis is shown in Fig. 11B based on the data shown in Figs. S5 to S10. In Experiment 1 (Figs. S5, S6), following TCR activation pSer175 CDK9 levels rose from undetectable (1.4% positive cells) to 70.3% positive cells during the first 4 hrs. During the next 20 hrs there was a gradual rise in the pSer175 CDK9+ cells reaching 84.7% at 24 hrs. There was a parallel rise in CycT1 and pThr186 levels, but because these markers were present at low, but measurable levels, a significant fraction of the resting cell population scored as a positive signal (60.1% CycT1 and 50.7% pThr186 CDK). Therefore, the use of these markers to define the activated cell phenotype is less reliable than the pSer175 marker.
Similar results were obtained in Experiment 2 ((Figs. S7, S8), where total CDK9, CycT1 and pSer175 CDK9 levels were monitored using flow cytometry. In this experiment essentially all the resting cells were positive for CDK9. However, total CDK9 levels rose gradually during the next 24 hrs, consistent with previous reports that total CDK9 expression is largely unaffected by TCR stimulation of resting CD4+ T cells [36], [72]. Therefore, the rapid induction of pSer175 could not be attributed to an elevation of CDK9 expression. In this experiment, approximately 10% of the resting cells were positive for CycT1 but less than 3% were positive for pSer175 CDK9. After TCR stimulation CycT1 and pSer175 CDK9 levels rapidly increased showing peaks at 30 min before reaching plateau levels by 4 hrs. The complex kinetics of P-TEFb induction seen in this experiment probably reflect fluctuations in TCR signaling due to the cyclical downregulation of the receptor [90], [91], [92].
We also examined P-TEFb responses to PMA stimulation of resting memory T-cells (Experiment 3, Figs. S9, S10). Compared to TCR stimulation, we consistent found that increases in pSer175 and pThr186 levels were comparatively small following PMA activation of the primary T-cells. This may help to explain relatively poor activation of latent HIV in resting memory cells by PMA [33], [93].
In Jurkat T-cells, PMA is a highly effective inducer of CDK9 Ser175 phosphorylation. Flow cytometry demonstrates that after 2 hr exposure to PMA, 18.8% of the Jurkat T-cells showed elevated pSer175 levels and 76.8% showed an elevation in RNAP II Ser2 C-terminal domain (pSer2-CTD) phosphorylation, a marker of P-TEFb phosphorylation of RNAP II (Fig. S11). Similar activation profiles were seen following 4 hr stimulation through the TCR, but TNF-α treatment did not elevate either pSer175 or pSer2-CTD levels.
In recent studies using Jurkat T-cells we demonstrated that TCR signaling regulates P-TEFb activity [53]. Examination of P-TEFb complexes by gel filtration chromatography showed that both PMA and TCR signaling led to the rapid dissociation of the large inactive 7SK snRNP complex and release of lower molecular weight P-TEFb complexes. The disruption of the 7SK snRNP correlated with a global increase in the association of P-TEFb with chromatin, recruitment of P-TEFb to the HIV provirus and a rapid increase in HIV elongation. Both P-TEFb recruitment to the HIV LTR and enhanced HIV processivity were blocked by the ERK kinase inhibitor U0126 but not by AKT and PI3 kinase inhibitors. Thus, TCR signaling, mediated through the PKC and ERK kinase pathways provides the first example of a physiological pathway that can shift the balance between the inactive and active P-TEFb pools and thereby stimulate proviral reactivation.
The effect of TCR signaling on HIV transcription could only be demonstrated in latently infected cells carrying proviruses that encoded functional Tat genes. Although Tat is itself capable of serving as an activator of P-TEFb by physically extracting the kinase complex from 7SK snRNP [74], [75], [77], in latently infected cells this is not occurring to a measurable extent, suggesting that there is a critical threshold level of Tat that will allow it to compete with HEXIM1 for P-TEFb. We therefore hypothesized that the disruption of 7SK snRNP triggered by signal-dependent changes in posttranslational modifications of one or more of its protein components is able to increase the interactions between extremely low levels of presynthesized Tat and P-TEFb.
In the present study we employed AP-MS/MS using Jurkat T-cells stably expressing FLAG-CDK9 to define the molecular requirements for the signal-dependent dissociation of P-TEFb from 7SK snRNP and its enhanced interaction with Tat to mediate proviral transcription elongation. This approach successfully led to the isolation of all the known 7SK snRNP protein components, mapping of substantial portions of their amino acid sequences, and the identification of novel PTMs on each of these proteins (Tables 1 to 5). PTMs involving acetylation, methylation and phosphorylation of each of the protein components of the complex were identified, but relatively few of these modifications were enhanced in response to PMA or TCR signaling. The identified PTMs in Tables 1 to 5 are based on data obtained in Jurkat T-cells which are actively dividing and have assembled 7SK RNP complexes. Many of these reported modifications are likely to be restricted in resting memory T-cells, as is the case for pThr186. We therefore expect that the majority of the modifications detected in Jurkat T-cells could only be present in activated primary T-cells. Although the identification of these modifications can be further strengthened using high resolution MS/MS data collection for site assignment, the tandem MS spectral quality of these modified peptide precursors was sufficient under the optimum (LTQ Velos ion trap) MS/MS data collection conditions to allow us to unambiguously make site assignments.
The phosphorylation event that showed the greatest increase in response to TCR or PMA signaling was the phosphorylation of CDK9 at Ser175, a highly conserved residue located in the activation loop of the kinase. In the Tat/P-TEFb X-ray structure published by Tahirov et al. [88], Ser175 is situated at the Tat/CDK9 interface about 6.5 Å away from Lys12 of Tat. Quantitative MS/MS analysis revealed that stimulation of Jurkat T cells with PMA, a potent activator of the MAPK/ERK pathway, induced >10-fold increase in Ser175 phosphorylation in three different experiments. We were able to confirm these MS/MS findings by pSer175 immunoblotting analysis of FLAG-CDK9 affinity purified complexes isolated from non-stimulated and PMA-treated cells. Ammosova et al. [79] also identified Ser175 as an in vivo phosphorylation site using Hunter peptide mapping of in vivo -labeled 32P-peptides followed by confirmation of the peptide sequence by mass spectroscopy.
Our structural models revealed that phosphorylation of CDK9 at Ser175 permits formation of a hydrogen bond with Lys12 of Tat and also strengthen intermolecular interactions postulated to exist between the activation loop of CDK9 and the N-terminal region of Tat. In support of this model is the observation that the S715A CDK9 mutation resulted in a 1.7-fold disruption of CDK9 association with Tat without interfering with CDK9/hCycT1 interaction while the phosphomimetic S175D CDK9 mutation resulted in a modest increase in Tat association (Fig. 5).
Mutations in S175 also had profound effects on interactions between CDK9 and BRD4. In agreement with the original studies of Yang et al. [80], who performed similar experiments using HeLa cells transfected with FLAG-tagged CDK9 plasmids, and Ammosova et al. [79], we found that S175A resulted in an almost complete block of CDK9 binding to BRD4 (1.9% of wildtype). However, we found in contrast to their results that S175D had 50% BRD4 binding capacity compared to wildtype. Therefore, we believe that BRD4 is probably also able to interact with Ser175-phosphorylated CDK9.
A P-TEFb interacting domain (PID) has been identified at the C-terminal end of BRD4 [94] and was shown to be sufficient to induce the dissociation of P-TEFb from 7SK snRNP [77], [95]. Overexpression of the PID alone in cells induced HEXIM1 and 7SK snRNA dissociation from P-TEFb, but it is not sufficient to activate Tat-independent transcription of the HIV LTR [95]. Although the region(s) of P-TEFb that interact with the PID of BRD4 have not been fully defined, our current observations suggest that the activation loop of CDK9 carrying phosphorylated Ser175 plays an important role in mediating P-TEFb interaction with BRD4.
Expression of CDK9 carrying mutations in Ser175 unexpectedly resulted in the activation of latent proviruses. The stable expression of the phosphomimetic mutation S175D CDK9 resulted in a 2.7-fold (E4 cells, wildtype Tat) and 3.7-fold (2D10 cells, H13L Tat) enhancement of HIV proviral gene expression (Fig. 7). These observations provide additional support for the idea that phosphorylation of Ser175 promotes a favorable interaction between CDK9 and Tat. However, the mutation with the strongest phenotype was S175A, which induced a 45-fold (E4 cells, wildtype Tat) and 20-fold (2D10 cells, H13L Tat) increase in HIV proviral gene expression. Since S175A reduced CDK9 association with Tat, we were surprised to observe that this mutation so effectively reactivated HIV proviral gene expression. We have attributed the S175A phenotype to elimination of the competition between Tat and BRD4 binding to P-TEFb for two reasons: a) S175A caused a much more severe disruption of CDK9 association with BRD4 than with Tat; and b) Stable knockdown of BRD4 expression in latently infected Jurkat T cells also led to a robust reactivation of proviral gene expression in a Tat-dependent manner [85], [86], [96]. Ammosova et al. [79] have also observed that expression of S175A is able to activate HIV transcription.
Transactivation of HIV by CDK9 mutants could only be observed when proviruses expressed functional Tat genes. This implies that BRD4 is not used to sustain basal HIV transcription in the absence of Tat. Instead, it appears likely that Tat and BRD4 compete for P-TEFb binding and that high basal BRD4 levels serve to restrict HIV transcription when Tat levels are low. Reductions in the levels of BRD4 in the cell, or reductions in the affinity of P-TEFb for BRD4, are therefore expected to permit extremely low levels of Tat to bind to P-TEFb and initiate HIV transcription.
Our model that phosphorylation of S175 enhances its interactions with Tat and promotes HIV transcription differs from the model of Ammosova et al. [79] who reported that dephosphorylation of Ser175 by PP1 upregulated HIV-1 transcription. However, since PP1 is a promiscuous enzyme, and CDK9 carrying unphosphorylated Ser175 is able to interact with Tat, it is possible that they have observed an indirect effect of the PP1 treatment rather than a phenotype directly ascribable to Ser175 dephosphorylation.
Resting memory CD4+ T-cells isolated from healthy donor peripheral blood were screened for Ser175 phosphorylation using a highly specific antibody exclusively recognizing CDK9 carrying pSer175. Using a novel flow cytometric assay, we found that resting memory CD4+ T-cells are highly restricted in hCyT1 expression and Thr186 phosphorylation of CDK9 (pThr186), in agreement with previous work from the Rice laboratory [36], [37], [38], [73].
Consistent with the results seen in Jurkat T-cells, unactivated primary cells also showed no detectable pSer175 CDK9. Stimulation of resting memory CD4+ T-cells through the TCR with α-CD3 and α-CD28 antibodies or by PMA resulted in a rapid elevation of hCycT1, and pThr186 CDK9 levels. Similar results were obtained using immunofluorescence, which also demonstrated that CDK9 carrying the pSer175 modification associates with nuclear speckles, a region believed to correspond to the site of active transcription. pSer175 levels rose with slightly delayed kinetics compared to that of hCycT1 expression and pThr186 and fluctuated substantially during a 24 hr time course while hCycT1 and pThr186 were relatively constant after 2 hr post-stimulation. The fluctuating pSer175 levels are probably a reflection of fluctuations in TCR receptor signaling due to the cyclical downregulation of the receptor [90], [91], [92]. Thus, pSer175 phosphorylation represents a separate and specific step in the activation of p-TEFb. Similar results were obtained using immunofluorescence, which also demonstrated that CDK9 carrying the pSer175 modification associates with nuclear speckles, a region believed to correspond to sites of active transcription [73].
We are currently adapting this flow assay to monitor the activation state of T-cells in clinical studies for the evaluation of compounds that reactivate HIV as part of the “shock and kill” strategy for viral eradication. Preliminary results in collaboration with Dr. David Margolis have shown that both pThr186 and pSer175 levels rise in CD4+ T-cells recovered from patients exposed to SAHA.
The flow cytometry and immunofluorescence data are consistent with a model where the initial incorporation of P-TEFb into 7SK snRNP requires hCycT1 and pThr186 but not pSer175 (Fig. 12). Consistent with this we found that CDK9 phosphorylated at Ser175 is absent from complexes purified using tagged-HEXIM1, which is highly enriched for the 7SK snRNP complex. Furthermore, the T186A and T186D mutants of CDK9 which cannot be incorporated into 7SK snRNP are modified at Ser175 in a signal-dependent manner. Finally, U0126, which is a potent inhibitor of MAPK/ERK signal-dependent disruption of the 7SK snRNP complex [53], only modestly inhibited formation of pSer175, suggesting that it is unable to block phosphorylation of the pre-existing “free” P-TEFb. Thus we conclude that Ser175 phosphorylation is probably not a primary signal leading to the disruption of the 7SK snRNP complex, but is instead a modification that is introduced subsequent to the liberation of P-TEFb from 7SK snRNP. The separate acquisition of the pThr186 and pSer175 modifications during T-cell activation is also consistent with the idea that pSer175 is absent from the inactive 7SK snRNP complex. It is interesting to speculate that modifying CDK9 at Ser175 not only enhances its interactions with BRD4 and Tat but also inhibits the reassociation of P-TEFb with HEXIM1 and 7SK snRNP. Follow-up studies are underway to identify the kinase that is responsible for pSer175 in order to precisely define the P-TEFb activation pathway.
In addition to the phosphorylation of CDK9 Ser175 that we have reported here, Fujinaga et al. [78] have recently reported that HEXIM1 is also phosphorylated in response to TCR signaling. Specifically they showed that there is a protein kinase C (PKC)-dependent phosphorylation of HEXIM1 on Ser158, a site believed to be within the RNA-binding domain of HEXIM1. Direct binding experiments showed that the phosphorylated HEXIM1 protein has a reduced affinity for 7SK snRNA and is unable to inhibit the enzymatic activity of P-TEFb [78].
Thus, it seems likely that multiple post-synthetic modifications of P-TEFb and 7SK RNP components are used to regulate P-TEFb assembly and disassembly. Besides the phosphorylation of CDK9 Ser175 and HEXIM1 Ser158, there are acetylation and methylation events that take place on CDK9, CycT1 and HEXIM1. Cho et al. [81] have reported that acetylation of CycT1 on 4 residues (K380, K386, K390 and K404) triggers dissociation of the 7SK snRNA complex and activates the transcriptional activity of P-TEFb. Consistent with their results we found that acetylation of K386 was enhanced almost 2-fold after PMA stimulation. Additional modifications we are currently evaluating for their potential functional significance include, acetylation and methylation of CDK9 Lys127, and acetylation of HEXIM1 Lys284.
In conclusion the activation of P-TEFb in these cells is a multi-step process that involves the initial assembly of P-TEFb into 7SK snRNP, signal-dependent release of P-TEFb from this inactive complex, and the phosphorylation of CDK9 at Ser175. In this paper, we have identified pSer175 as a modification at the activation loop of CDK9 that plays a critical role in altering the competitive binding of Tat and BRD4 to P-TEFb. pSer175 also provides an easily detected molecular marker for the transcriptionally active form of P-TEFb in primary CD4+ T-cells.
RPMI 1640 medium and fetal bovine serum were purchased from Hyclone. PMA and LARP7 antibody were purchased from Sigma. U0126 was obtained from Calbiochem. α-CD3 and α-CD28 antibodies were obtained from BD Biosciences. CDK9, hCycT1, and TRITC-conjugated hCycT1 antibodies were from Santa Cruz Biotechnology. Phospho-Thr186 CDK9 antibody was purchased from Cell Signaling Technology. HEXIM1 and BRD4 antibodies were custom synthesized and affinity purified by Covance Research Products. The MSCV retroviral expression system (Clontech) was used in the current study for the creation of stable Jurkat T-cell lines expressing FLAG-CDK9 wt, the indicated FLAG-CDK9 point mutants, or FLAG-HEXIM1.
Rabbit polyclonal antiserum towards a 19-residue phospho-Ser175 epitope for CDK9 (ADFGLARAFpSLAKNSQPNR) was generated at Covance Research Products. After affinity isolation of the antibody using the phospho-serine peptide, eluted antibody fractions were subjected to multiple rounds of purification by negative selection with affinity resin for the corresponding unmodified peptide. Ser175Ala and Ser175Asp mutants of CDK9 were used to confirm by Western blotting that the purified antibody was selective towards the phospho-Ser175 epitope.
8.0×108 Jurkat T cells stably expressing FLAG-CDK9 or FLAG-HEXIM1 were treated or not with 50 ng/mL PMA or 25 µM DRB for 1 h. After washing the cells twice with 1× PBS, whole cell extracts (WCEs) were prepared using cell lysis buffer A [150 mM NaCl, 10 mM KCl, 1.5 mM MgCl2, 0.5% NP-40, 1 mM DTT, 10 mM Hepes pH 8.0] or RIPA buffer [150 mM NaCl, 0.5% Triton X-100, 0.5% Sodium deoxycholate, 0.1% SDS, 5 mM EDTA, 20 mM Tris HCL, pH 7.5] containing a cocktail of protease and phosphatase inhibitors. WCEs were cleared by centrifugation at 2000 rpm for 5 min and 2 h incubation with protein A sepharose beads prior to incubating them overnight at 4°C with anti-FLAG M2 agarose beads (Sigma). Immunoprecipitates were washed extensively with cell lysis buffer A or RIPA buffer and elution of FLAG-CDK9 or FLAG-HEXIM1 complexes was performed by overnight incubation at 4°C with 200 µg/mL FLAG peptide (Sigma) in buffer A or RIPA buffer.
FLAG peptide protein eluates were concentrated by methanol-chloroform precipitation and rehydrated with 1× LDS loading buffer containing 50 mM DTT before being resolved by 1D SDS-PAGE on a 4–12% Bis-Tris gel. After SDS-PAGE the gel was stained with SYPRO Ruby (Invitrogen) as recommended by the manufacturer. Protein bands on the gel were visualized on a UV light source whose surface had been cleaned with 50% isopropanol. Selected bands were excised from the gel with a sterile blade, crushed in 50 mM ammonium bicarbonate buffer, pH 8, before being subjected to a standard in-gel digest protocol involving a reduction step with 20 mM DTT and alkylation with 55 mM iodoacetamide in the dark. Reduction was preceded by wetting the gel pieces to its brim with 50 mM Ammonium bicarbonate buffer (pH 8.0) for 15 minutes followed by addition of 50% acetonotrile/25 mM Ammonium bicarbonate buffer for 15 minutes. Appropriate washing step of adding 50 mM Ammonium bicarbonate buffer (50 mM) was also done in between the reduction and alkylation steps. An extra wash was done just before drying the gel pieces in 100% acetronitrile. After removal of the organic solvent followed by drying of the gel pieces, overnight tryptic digestion step was performed at 37°C with the addition of 200 ng of sequencing grade trypsin (Promega Inc, WI) in 50 µL of 50 mM ammonium bicarbonate buffer, pH 8. Afterwards, 0.2% formic acid was added to stop the proteolytic processing and the resulting peptides were extracted from the supernatant fraction of the in-gel digest along with the recovered fractions from 2 rounds of 50% acetonitrile/0.3% formic acid extractions. Upon drying and resolubilization in 25 µL of 0.1% formic acid, the samples were processed for the LC-MS/MS analysis as described below.
The digests prepared above were analyzed by LC-MS/MS using a Waters nano Acquity UPLC system (Waters Inc, MA) that was interfaced to a LTQ Velos-Orbitrap mass spectrometer (Thermo-Finnigan, Bremen, Germany). The platform was operated in the nano-LC mode using the standard nano-ESI API stack fitted with a picotip emitter (uncoated fitting, 10 µm spray orifice, New Objective, Inc., Woburn, MA). The solvent flow rate through the column was maintained at 300 nL/min using the split-free Acquity system. The protein digests (7 µL) were injected into a reversed-phase symmetry C18 trapping column (0.18×20 mm, 5 µm particle size, Waters Inc.) equilibrated with 0.1% formic acid (FA)/2% acetonitrile (v/v) and washed for 5 min with the equilibration solvent at a flow rate of 15 µL/min, using the sample trapping mode of UPLC. After the washing step, the trapping column was switched in-line with a reversed-phase C18 nanoacquity UPLC column (0.075×250 mm, Waters Inc.) and the peptides were separated using a linear gradient of acetonitrile from 5% to 45% in aqueous 0.1% formic acid over a period of 60 min (0.67% gradient) at the above-mentioned flow rate such that the eluate was directly introduced to the mass spectrometer. A 100% acetonitrile elution step was subsequently performed for 10 minutes prior to resetting the analytical column to the initial equilibration conditions for 15 more minutes at the end of the chromatographic run, making a total run time of 90 min. for the LC-MS/MS analysis. The mass spectrometer was operated in a data-dependent MS to MS/MS switching mode, with the 10 most intense ions in each MS scan subjected to MS/MS analysis. The full scan was performed at 60000 resolution in the Orbitrap detector and the MS/MS fragmentation scans were performed in the Velos dual ion trap detector (IT) CID mode. The threshold intensity for the MS/MS trigger was always set at 1000 and the fragmentation was carried out using the CID mode using a normalized collision energy (NCE) of 35. The data was entirely collected in the profile mode for the full scan and centroid mode for the MS/MS scans. Dynamic exclusion function for previously selected precursor ions was enabled during the analysis such that the following parameters were applied: repeat count of 2, repeat duration of 45 seconds, exclusion duration of 60 seconds and exclusion size list of 450. Xcalibur software (version 2.0.7), Thermo-Finnigan Inc., San Jose, CA) was used for instrument control, data acquisition, and data processing.
LC MS/MS files exported from XCalibur were directly used for the protein database searching step. For all the search tasks, Mascot was the preferred search engine with the public IPI human (Version June 2010) database. Search parameter file for probing S/T/Y phosphorylation included the following settings: precursor mass tolerance of 5 ppm, fragment mass tolerance of 0.8 Da, 1 missed cleavage, use of decoy database and post-translational modification (PTM) search options including 1) cysteine carbamidomethylation, 2) methionine oxidation and 3) S,T/Y phosphorylation. In case of the database search for alternate PTMs such as methylation and acetylation, separate searches were initiated on the same raw files that included 1) Methylation (R,K) 2) Dimethyl (R,K) 3) Acetylation (K) apart from Cysteine carbamidomethylation as possible modifications to be probed at the same time. Protein identification were subject to strict data QC filtering within MASCOT at peptide level such as (a) peptide expectation value of 0.05 or less and b) single peptide hits being ignored whereas the peptide identifications made any for post-translation modifications (PTMs) were subject to manual verification using an extended MS/MS interpretation routine that accommodates potential neutral losses in the mass spectrometer collision cell (CID mode).
For quantitative data analysis to determine the abundance of the selected peptides of interest in control and treated samples using a relative peptide quantitation approach, in which the modified and unmodified forms of the peptide were successfully identified , the 325–1800 m/z full-scan, MS1, high resolution MS data was used to perform the extracted ion chromatogram (XIC) analysis based on a label-free quantitation method that relies on chromatographic peak area calculation after peak smoothing. This process was done manually since there were only a few precursor masses of interest in this study. In the case of CDK9's T186 containing-peptide 179-NSQPNRYTNR-188, the T186 site was detected only in the phosphorylated form and hence the relative abundance of this peptide under different conditions was calculated with reference to an internal standard-like peptide 295-LLVLDPAQR-303 using the commonly accepted abundant peptide ratiometric approach [97]. To maintain consistency in our quantitation approach, the relative quantitation for the S175 phosphorylation event was also performed using the detected 173-AFSLAK-178 peptide with reference to the 295-LLVLDPAQR-303 abundant peptide ratiometrically even though both the modified and unmodified forms of the 173-AFSLAK-178 were detected most of the time and that the peak areas of the S175-phosphorylated/modified and unmodified forms can be used independently to make a determination of their relative abundances under different conditions of interest. To calculate the variance of quantitation encountered using the XIC analysis, n = 3 data collection was performed for S175 phosphorylation and the data analysis procedure followed the method described above with the exception that abundance peptide approach was applied only if needed and that the relative peptide approach was the preferred approach.
Jurkat T cells were engineered to stably express FLAG-CDK9, FLAG-CDK9 mutants or FLAG-HEXIM1. Cells were maintained in RPMI 1640 medium supplemented with 5% fetal bovine serum, penicillin (100 IU/mL), streptomycin (100 µg/mL), and 25 mM Hepes at 37°C in 5% CO2. After the appropriate experimental treatments as discussed in the Results section, WCEs were prepared using cell lysis buffer A and cleared by microcentrifugation at 5000 rpm for 5 min. Protein concentration of WCEs was determined using the BCA protein assay kit (Pierce) and equal amounts of WCEs were resolved by NuPAGE 1D-SDS PAGE (Invitrogen). Following transfer of resolved proteins to a nitrocellulose or PVDF membrane, Western blotting analysis was performed with antibodies against CDK9, HEXIM1, Cyclin T1, LARP7, phospho-Thr186 CDK9, phospho-Ser175 CDK9, Tat, FLAG, and HA epitopes.
Cleared WCEs were also subjected to immunoprecipitation with anti-FLAG M2 agarose beads or anti-HA agarose beads (Sigma). After extensive washes of the immunoprecipitates with cell lysis buffer A, protein elution was performed with 200 µg/mL of either FLAG or HA peptide, and the eluates were analyzed by Western blotting as discussed above.
Whole cell extracts were prepared from Jurkat T cells (5×108) belonging to uninfected, CDK9 wt, S175A, S175D, T186A, and T186D and used to immunopurify FLAG-CDK9 complexes using anti-FLAG M2 affinity beads (Sigma). The in vitro kinase reaction was set up as follows with or without the addition of 250 ng His-tagged full length human RNAP II CTD repeat substrate (Abcam): 20 µL of the anti-FLAG protein immune complex in kinase dilution buffer (50 mM Hepes, pH 7.5, 1 mM DTT), 2 µCi of γ32 ATP, 1.5 µM ATP, and 25 µL of 2× Standard assay buffer (100 mM Hepes, pH 7.5, 2 mM DTT, 6 mM MgCl2, 6 mM MnCl2) containing a phosphatase inhibitor cocktail. The kinase reaction was performed at 30°C for 1 h. Thereafter, reactions were stopped by boiling the samples in LDS sample loading buffer, subjected to SDS-PAGE on a 4–12% Bis-Tris gel (Invitrogen), and the dried gel was analyzed by autoradiography.
Heparinized peripheral blood was obtained from a healthy donor and used to isolate peripheral blood mononuclear cells (PBMCs). Memory CD4+ T cells were purified from PBMCs using negative bead selection and challenged with a combination of anti-CD3 and anti-CD28 antibodies to activate the T-cell receptor. For surface staining with anti-CD25 and anti-CD69, cells were first washed with autoMACS buffer (Miltenyi Biotech) and then incubated with each fluorochrome-conjugated antibody for 30 min in the dark and on ice. For intracellular staining with anti-Cyclin T1, anti-phospho-Ser175 CDK9, and anti-phospho-Thr186 CDK9, surface stained cells were washed with autoMACS buffer and permeabilized and fixed using the Foxp3 Fixation/Permeabilization Kit (eBioscience) according to the manufacturer's protocol. Thereafter, these cells were rinsed twice with 1× permeabilization buffer resuspended in 1× permeabilization buffer containing 2% mouse serum, and incubated for 15 min at room temperature. Anti-phospho-Ser175 CDK9, and anti-phospho-Thr186 CDK9 antibodies were conjugated using Zenon labeling kit (Invitrogen) immediately prior to use. Fluorochrome-conjugated antibodies were added to the cells and incubation was performed for 30 min in the dark and on ice. After rinsing the cells twice with 1× Permeabilization buffer they were subjected to flow cytometry analysis using the LSR Fortessa instrument equipped with a red 640 laser with a 780/60 and 670/30 filter.
Memory CD4+ T cells were allowed to adhere to cover slips coated with poly-L-lysine for 5 min at 37°C. The cells were fixed and permabilized for 30 minutes using the FoxP3 Fixation/Permabilization kit (eBioscience). Afterwards the cells were blocked using 2% normal mouse IgG (The Jackson Laboratory) for 15 min. Antibodies for anti-phospho-Ser175 CDK9, anti-phospho-Thr186 CDK9, anti-SC35 Nuclear Speckle Marker (Abcam), anti-Cyclin T1(Santa Cruz Biotechnologies) and DAPI were added for 15 min. Cells were washed three times and Cy5, Cy2 secondary antibodies were added for 15 min and subsequently washed three times. The cover slips were mounted using gel mount (Electron Microscope Sciences) and viewed using a DeltaVision epifluorescent microscope (Applied Precision). Images were captured in z series, deconvolved, and processed using the Softworx analysis program (Applied Precision). Images were exported as TIFF or JPEG files, and Figures were composed using Adobe Photoshop CS.
The 2.1 Å X-ray structure of HIV-1 Tat complexed with human P-TEFb was retrieved from protein data bank (PDB ID: 3MIA). A model was prepared for the phosphorylated serine at position 175 of CDK9. The structural conformation of this modified Tat/P-TEFb complex was energy minimized to remove any unfavorable contacts and to allow for productive interactions which might occur as a result of the modification. Energy minimization was performed in explicit solvent environment by solvating the structures in TIP3P waters. Two-step energy minimization was performed – first only the water molecules were energetically relaxed keeping the protein fixed and in the second step the whole system was energy minimized without any restraints. Each step involved minimization employing combination of steepest descent and conjugate gradient algorithms. AMBER 11 suite of programs was used perform the calculations using FF99SB force field. Parameters for phosphorylated serine were obtained from AMBER parameter database.
Procedures for obtaining anonymous blood donations from healthy volunteers were approved by the University Hospitals of Cleveland IRB (Number: 12-11-33). Adult healthy volunteers were recruited by the Case CFAR Clinical Core from the immediate community by flyer advertisements. The only procedure that the volunteers (healthy nonpregnant adults who weigh at least 110 pounds) underwent was venipuncture to obtain peripheral blood samples. This was performed by trained phlebotomists in the Case CFAR Clinical Core. Verbal and written consent was obtained from adult volunteers participating in this study by persons who have been certified in human subject protection regulations. Blood samples were assigned a code number unrelated to the volunteer's medical record number.
|
10.1371/journal.pgen.1000594 | Notch and Prospero Repress Proliferation following Cyclin E Overexpression in the Drosophila Bristle Lineage | Understanding the mechanisms that coordinate cell proliferation, cell cycle arrest, and cell differentiation is essential to address the problem of how “normal” versus pathological developmental processes take place. In the bristle lineage of the adult fly, we have tested the capacity of post-mitotic cells to re-enter the cell cycle in response to the overexpression of cyclin E. We show that only terminal cells in which the identity is independent of Notch pathway undergo extra divisions after CycE overexpression. Our analysis shows that the responsiveness of cells to forced proliferation depends on both Prospero, a fate determinant, and on the level of Notch pathway activity. Our results demonstrate that the terminal quiescent state and differentiation are regulated by two parallel mechanisms acting simultaneously on fate acquisition and cell cycle progression.
| Despite substantial progress that has been made, we still know little about how single precursor cells undergo a limited number of cell divisions before arrest. Discovering the mechanisms by which terminal cells maintain cell division arrest is essential for understanding “normal” development, as well as the origin of pathological deregulations. Using the bristle cell lineage, a model system widely employed to analye cell identity acquisition, we observed that only two out of four terminal cells in this lineage are unable to re-enter the cell cycle and proliferate. Our study shows that in these cells, cell division arrest is maintained by the action of the transcription factor Prospero and the signalling pathway Notch. Since both of these factors also control cell identity in this lineage, this finding demonstrates that common elements acting simultaneously and in parallel regulate the terminal quiescent state and differentiation. This system provides a unique animal model in which to understand how the mechanisms involved in cell fate acquisition and those controlling cell division intermingle to produce cell lineages resulting in terminal cells in the right number and at the right place and time.
| A high degree of coordination between cell proliferation, cell cycle arrest and cell differentiation is essential for proper development. Disruption of this coupling can lead to malformations and eventually cancer. Cell cycle progression relies primarily on the activity of cyclin-dependant kinases (Cdk) that are regulated by their association with factors like cyclins or cyclin kinase inhibitors (CKI) and by phoshorylation or dephosphorylation [1]. In worms and vertebrates, this mechanism is redundant and inactivation of certain cell cycle factors can be compensated by the activation of others. This has been observed for Cyclin-E (CycE), which modulates Cdk2 activity and controls the transition from the G1 to S phase. In mouse and C. elegans, cell divisions are not completely blocked after genetic ablation of cycE [2],[3]. However, cycE−/− mouse cells are resistant to oncogenic transformations suggesting that normal and oncogenic proliferation have different requirements for CycE [3]. In Drosophila, the core mechanism of the cell cycle is not redundant and down-regulation of CycE arrests the cell cycle. Thus, CycE appears to be the most important G1 cyclin in all Drosophila cell divisions studied so far. In addition, it has been shown that ectopic expression of CycE after terminal mitosis induces re-entry into the S-phase resulting in additional cell cycles [4]–[6]. This shows that terminal cells continue to respond to CycE, and suggests that, as in vertebrates, Drosophila CycE seems to be central to the generation of ectopic rounds of cell divisions after cell cycle deregulation.
In the Drosophila bristle lineage, which produces the external mechanosensory organs called microchaetes, cell cycle progression and cell determination are intimately related [7]. Each adult microchaete is composed of four cells: two outer cells, the socket cell and the shaft cell, and two inner cells, the neuron and the sheath cell [8]. Each cell differs from the other by its size, localisation in the cluster and expression of specific markers (Figure 1). All four cells arise from a unique precursor cell, pI, after four asymmetric cell divisions occurring during early pupal development. At each division, one daughter cell (N-off) acts as a Notch ligand-producer and the other (N-on) as a Notch signal-receiver [9],[10]. The bias in the activation of the N-pathway assured the acquisition of different fates by both daughter cells. During the first round of division, the pI cell divides at about 16 h after puparium formation (APF) and generates two secondary precursor cells, pIIa and pIIb. During the second round of mitosis, the pIIb cell divides prior to the pIIa cell giving rise to a glial cell and a tertiary precursor cell pIIIb. The division of pIIa generates the socket and the shaft cell. Finally, the pIIIb cell divides to produce the neuron and the sheath cell [11]. Later in development, between 21 to 24 h APF, the glial cell undergoes apoptosis [12]. Thus, only four cells of the bristle lineage form each sensory organ. Upon completion of the lineage, cells stop proliferating and terminally differentiate. This stereotyped lineage has become an excellent model to analyse the relationship between cell cycle and cell determination. In particular, to address questions as to how cells maintain their terminal quiescent state or whether all terminal cells are responsive to proliferative signals. To analyse these issues, we tested the capacity of post-mitotic cells to re-enter the cell cycle in response to the overexpression of CycE that mimicked the cell's response to a proliferative condition. Surprisingly, not all cells in the lineage are sensitive to this overexpression and we show that the responsiveness to ectopic proliferation depends on both Prospero (Pros), a transcription factor, and the level of Notch (N) pathway activity.
In order to analyse the mechanisms that maintain arrest in terminal cells of the bristle lineage, we forced proliferation by specific overexpression of CycE using the UAS/Gal4 system [13]. The neuralized p72-GAL4 line (neur) was used as a driver to simultaneously overexpress CycE and Histone2B::YFP (H2B::YFP), which highlights the DNA [14]. After CycE overexpression, 84% of the organs contained two sockets as revealed by scanning electron microscopy (Figure 2A). At the cellular level, each sensory organ cell was identified by the expression of H2B::YFP (see Materials and Methods), socket cells by the accumulation of high levels of Suppressor of Hairless (Su(H)), sheath cells by the presence of Prospero, and neurons by the presence of ELAV or 22C10/Futsch (see Figure 1) [9],[11],[12],[15]. At 28 h APF, we observed three types of clusters. 13,5% of the clusters were wild-type and formed by four cells (Figure 2B, upper row, n = 22, and Figure 2C), 70% of the clusters contained one additional socket cell (Figure 2B, middle row, n = 115, and Figure 2C), and 16,5% of the clusters exhibited two additional cells namely, an extra-socket cell and an extra-neuron (Figure 2B, bottom row, n = 28, and Figure 2C). We never observed clusters with more than one shaft and one sheath cell. When overexpression of CycE was carried out at 30°C, where the GAL4/UAS system is more efficient, cluster composition was similar to that of pupae maintained at 25°C (Figure 2C, 92% of the clusters showed duplicated socket cells, n = 71, and 58% showed multiple neurons, n = 131). Similar results were obtained when cluster composition was analysed up to 48 h APF showing that after a time-lapse longer than six cell-cycles neither extradivisions nor apoptosis occurred (Figure S1 and data not shown). These data indicate that the duplication of neurons and socket cells are reproducible events induced after CycE overexpression and suggest that bristle lineage cells have differential sensitivities to proliferating signals.
To identify the origin of these additional differentiated cells, we carried out in vivo imaging to follow the formation of the bristle lineage in neur>H2B::YFP, PON::GFP (see supplementary data in [16] and Video S1) and in neur>CycE, H2B::YFP, PON::GFP pupae (Video S2). We used the PON::GFP fusion protein to identify the normal set of sensory cells during the time-lapse recording. Precisely, PON::GFP was inherited by the pIIb cell, the glial cell, the shaft cell and by the neuron [17].
Overexpression of CycE did not affect the sequence or the asymmetry of the first four divisions. Thereafter, up to two supplementary divisions were observed. One involved a pIIa daughter cell identified as the future shaft cell by its position in the cluster and its inheritance of Pon::GFP during pIIa mitosis (observed in 80% of clusters analysed, n = 124). The second involved a pIIIb daughter cell identified as the future neuron by its position and its inheritance of Pon::GFP during pIIIb mitosis (observed in 25% of clusters analysed, n = 124) (see Video S2 and Figure S2). This supplementary division was always observed in clusters where the pIIa daughter cell had also undergone an extra-division. Consistently, overexpression of CycE at 30°C led to 94,5% of the shaft cells and 45% of the neurons undergoing an extra-mitosis (n = 102).
To confirm the identity of the ectopically dividing cells, triple stainings were performed labelling metaphasic cells (phospho-ser10 histone H3 (PH3) immunoreactivity) together with bristle lineage cells, neurons or socket cells. Figure 3A shows a mitotic cell in a five-cell cluster from a pupae at around 22 h APF. This mitosis is the first additional division that we observed with time-lapse imaging. The mitotic cell was identified as the future shaft cell by its nuclear size, position and its lack of Su(H) accumulation. The second additional mitosis was observed in clusters from pupae at around 24 h APF. In this case, the mitotic cell corresponded to the neuron since it expressed ELAV (Figure 3B) and 22C10/futsch (Figure 3C) markers. The fact that in five-cell clusters an extra socket cell was observed suggests that the extra division of the future shaft cell gives rise to a shaft and a socket cell. Similarly, in later clusters, two neurons were observed suggesting that the extra-division of the future neuron gives rise to two neurons. These observations were confirmed in in vivo time-lapse followed by immunodetection experiments (not shown). Taken together these data indicate that the shaft cell and neuron are the only two cells in the terminal bristle lineage to undergo an ectopic division when CycE is overexpressed (Figure 3D).
To further analyse cell behaviour in response to CycE overexpression, we sought to determine the critical period during which cells were competent to undergo additional mitosis. To do so, CycE was expressed under the control of a heat-shock promoter at different moments in the cell lineage. Antibody staining was carried out twelve hours after heat shock to distinguish socket cells and neurons. Figure 3E shows the percentage of clusters with duplicated socket cells or neurons following heat shock treatment at different times APF. Two socket cells were generated only when CycE was overexpressed at the time of pIIa mitosis (between 18 h and 19 h45 APF). In contrast, the period in which CycE overexpression was able to trigger a division of the neuron was longer (from 19 h45 to 26 h APF). Since axogenesis starts at about 23 h30 APF [12], ectopic neuronal divisions can occur at a time when neurons appear to be fully differentiated. This is shown in Figure 3C in which a dividing cell was positive for 22C10/Futsch, a late neuronal marker that reveals differentiated neurons [15]. Interestingly, in some cases we observed that the future neuron underwent two consecutive rounds of supplementary divisions (not shown). These data indicate that neurons retain their competence to divide late into differentiation.
The data presented above indicate that the first extra mitosis is asymmetric and generates two different cells (a shaft and a socket cell) while the second one is symmetric giving rise to two identical cells (two neurons). In the normal bristle cell lineage all mitosis are asymmetric. The cell-fate determinants, Numb and Neuralized (Neu), co-segregate to one daughter cell, where both act to bias the Notch-mediated fate decision [18],[19]. We analysed the distribution of these factors in cells undergoing extra divisions following CycE overexpression. During the mitosis of the future shaft cell, Numb and Neu were detected and formed a crescent at the anterior pole of the mitotic cell (n = 14, Figure 3A, arrowhead and data not shown). The differential segregation of these factors was in agreement with the unequal repartition of the Pon::GFP fusion protein observed during in vivo recordings (see Video S2 and Figure S2). In contrast and in accordance with the symmetric characteristic of this division, Numb was never detected during the extra division of the future neuron (n = 9, Figure 3B). This is in agreement with in vivo recording data showing that Pon::GFP seems to be distributed uniformly between both daughter cells (Video S2). These data indicate that when cell divisions are forced in the pIIa sub-lineage, the future shaft cell behaves as its mother cell, pIIa, dividing asymmetrically and giving rise to another shaft and socket cell.
The extra division of the future shaft cell was not correlated to different levels of CycE overexpression (Figure S1). Moreover, both cells responded similarly to CycE overexpression, exhibiting precocious and synchronous entries into the S phase [20]. This suggests that certain factors favours the progression of the mitotic cell cycle in the shaft cell or alternatively, prevent division in the socket cell. As the N pathway is off in the shaft cell, its capacity to respond to CycE overexpression could be related to the absence of N pathway activation. To analyse this possibility, we controlled the activation of the N pathway concomitantly in both pIIa daughter cells with CycE overexpression (neur>CycE). N pathway was inhibited by expressing Numb under the control of a thermo-inducible promoter in both pIIa daughter cells (30 min pulse at 38°C at 20 h APF, [21]). Pdm1 was used as a marker of all pIIa progeny [16] and socket cells were identified by their accumulation of Su(H) (Figure 4A–4D). The percentage of clusters containing 2, 3 or 4 pIIa daughter cells obtained under different experimental conditions is shown in Figure 4E. In each case, the proportion of clusters harbouring 1 or 2 socket cells is indicated. Importantly, under these conditions Numb overexpression induced a very low rate of socket to shaft cell transformation (Figure 4E, hs-numb column 2). When CycE was overexpressed together with mild induction of Numb, we observed that 28% of the clusters harboured four pIIa daughter cells with two socket cells and two shaft cells (Figure 4C, E hs-numb, neur>CycE column 4). Under these same conditions, in vivo analysis showed that both pIIa daughter cells divided (not shown). Such clusters were never observed when CycE was overexpressed alone (Figure 4E, neur>CycE column 4). Thus, when N activity was reduced and CycE overexpressed, the future socket cell divided asymmetrically giving rise to a shaft and a socket cell. This suggests that, similarly to the shaft cell, the socket cell retains pIIa features shortly after birth. Taken together, these data show that full activation of the N pathway is necessary to prevent the future socket cell from entering mitosis upon CycE overexpression.
To further analyse the action of N on the proliferative capacity of terminal cells, we performed reciprocal experiments and activated the N pathway in the shaft cells. This pathway was activated after overexpression of the intracellular domain of the N receptor (Nintra) [22]. The expression of Nintra was induced at 20 h APF by a 30 min pulse at 34°C. Under these conditions, only 45% of clusters were transformed (clusters harbouring two socket cells, Figure 4E, hs-Nintra, column 2). We took advantage of this mild penetrance to analyse the effect of CycE overexpression together with weak Notch dependant transformation. We observed that the number of clusters containing three pIIa daughter cells (two of which were socket cells) was reduced from 91% under CycE overexpression alone to 38% under conditions of N and CycE overexpression (Figure 4E compare neur>CycE column 3 and hs-Nintra, neur>CycE column 3). This reduction was associated with an increase in the number of clusters having a normal set of pIIa daughter cells (a socket and a shaft cells, Figure 4D and 4E, hs-Nintra, neur>CycE column 2). This indicates that the extra divisions induced by CycE overexpression were blocked by mild activation of the N pathway. These data show that in the pIIa sub-lineage, activation of the N pathway is involved in maintaining the future socket cell in a quiescent state. As a consequence, CycE overexpression induced cell division exclusively in the future shaft cell. A similar anti-mitotic action of the N pathway has been reported in Drosophila follicle cells. In these cells, lack of N activity has been shown to induce extra mitoses at the expense of endocycles [23],[24].
Similar to the situation in the pIIa sub-lineage, only the N-on cell of the pIIIb sub-lineage, namely the sheath cell, did not undergo extra mitosis upon CycE overexpression. To test whether activation of the N pathway was necessary to prevent ectopic division of the sheath cell, we reduced the activation of this pathway by overexpressing Numb (30 min heat shock at 38°C, at 21 h30 APF) together with CycE (neur>CycE). Under these conditions, the overexpression of Numb was not sufficient to modify cell identity, since both an ELAV and a Pros positive cell (neuron and sheath cell respectively) were present in all clusters. Surprisingly, even though the conditions of Numb overexpression were similar to previous experiments, we observed no change in the number of clusters with duplicated sheath cells when CycE was overexpressed (0% vs 4% in neur>CycE (n = 126) and in hs-numb, neur>CycE (n = 118) respectively). Similarly, no change was observed in the number of clusters with duplicated neurons (19% vs 22% in neur>CycE (n = 126) and in hs-numb, neur>CycE (n = 118) respectively).
The fact that the response of pIIIb daughter cells to CycE overexpression was invariant after Numb overexpression suggests that factors other than N can prevent mitosis in these cells. One candidate that may impede extra mitoses is the cell determinant Pros, as loss of function of pros results in ectopic mitotic activity in the Drosophila central nervous system (CNS) [25],[26]. In the bristle lineage, Pros is detected in the pIIIb cell during its division, is inherited by the neuron where it disappears rapidly and, at the same time, is expressed in the sheath cell (Figure 1) [11]. To analyse the putative anti-mitotic role of Pros in the bristle lineage, CycE was overexpressed in a pros17/+ heterozygous background. Results depicted in Figure 5A show that 26% of the clusters contained duplicated sheath cells in pros17/+, neur>CycE (n = 129) compared to 1% in neur>CycE alone (n = 71). Interestingly, we also observed an increase in the percentage of clusters containing multiple neurons (86% n = 129). These results suggest that with CycE overexpression, Pros affects the proliferative properties of the sheath cell and the neuron even if in this later Pros is transiently expressed. Similar results were obtained by overexpressing CycE using a hs-CycE construct at the time of pIIIb division (22 h APF) in a pros17/+ background (data not shown). To determine the origin of the duplicated cells, we combined time lapse imaging to follow the pattern of cell division and immunostaining to identify the fate of pIIIb daughter cells. Within the 25 lineages followed, three types of extra divisions were observed: (i) in 64% of the cases, only one pIIIb daughter cell underwent an ectopic mitosis, giving rise to two neurons; this is similar to what was observed upon overexpression of CycE alone (Figure 3); (ii) in 8% of the cases, three neurons were observed, the third one resulting from an extra cell division of a duplicated neuron (not shown); (iii) in 24% of the cases, both pIIIb daughter cells divided, giving rise to two neurons and two sheath cells respectively. An example of an in vivo recording of such a lineage is shown Figure 5B and Video S3. These data show that both pIIIb daughter cells underwent a symmetric division, giving rise to two neurons and two sheath cells respectively. The symmetric nature of these divisions was confirmed by the lack of Numb staining during these additional mitoses (Figure 5C) and Pros was equally distributed between both daughter cells (Figure 5D). These data indicate that in a pros17/+ background, both pIIIb daughter cells are able to undergo an ectopic and symmetric mitosis in response to CycE overexpression. To determine whether loss of function of pros alone could also induce extra mitoses, we analysed the bristle lineage in pros null clones. Although CycE staining of pIIIb daughter cells inside pros clones was more intense, we failed to detect supplementary cell divisions (Figure S3). The absence of extra mitoses inside the pros null clones could be explained by the absence of Dacapo downregulation, Dacapo (Dap) being a CycE inhibitor.
Taken together, these data indicate that Pros has a dual function in the bristle lineage. In addition to its involvement in neuron and sheath fate determination [27], Pros acts as a cell cycle regulator in these two cells since it prevents extra cell divisions under conditions conducive to proliferation.
The data presented above indicate that activation of the N pathway prevents mitosis of the socket cell upon CycE overexpression. A similar role is played by Pros in the sheath cell. Since the N pathway is active in the sheath cell, N and Pros could act redundantly to prevent extra divisions in this cell. To analyse this possibility, we overexpressed CycE (neur>CycE) while downregulating N activity (hs-numb) in a pros17/+ background. Induction of Numb expression (30 min heat shock at 38°C) was performed under visual control on living pupae followed by immunostaining to identify sheath cells and neurons. Only clusters heat shocked within one hour following the pIIIb division were analysed. We observed a significant increase in the proportion of sheath cells that underwent an extra division (Figure 6A, black box): 62% in hs-numb, pros17/+, neur>CycE (n = 26) versus 19% in pros17/+, neur>CycE (n = 72) and 4% in hs-numb, neur>CycE (n = 104). These data indicate that decreasing N pathway activity in a pros17/+ mutant background favours the division of the future sheath cell upon CycE overexpression.
The joint action of Notch and Pros on the maintenance of the quiescent state of pIIIb daughter cells has been revealed in a sensitized background of CycE overexpression. In order to analyse whether both N and Pros act similarly under normal conditions, we studied the composition of sensory clusters inside null pros17 clones in conditions of down regulation of the N activity. This downregulation was obtained using pupae heterozygous for the thermosensitive allele of N, Nts-1, maintained at restrictive temperature of 30°C [28]. In control conditions (at a permissive temperature of 18°C) and using Cut immunoreactivity to identify lineage cells, we always observed that all sensory clusters inside pros17 clones were composed by four cells, two pIIa and two pIIIb daughter cells, the former harbouring a nucleus bigger than the latter (Figure 6B, [12]). Similar results were observed under restrictive temperature (30°C) in heterozygous Nts-1 pupae alone or in pros17 heterozygous tissue (data not shown and arrows in Figure 6C–6D). However, in 8% of clusters inside null pros17 clones, when N-function was reduced (at 30°C), we observed a supplementary pIIIb daughter cell identified by its small nucleus (arrowheads in Figure 6C–6D). In half of these cases, the supplementary cell was ELAV–positive indicating a neuronal identity. In the other half, the ELAV-negative cell probably corresponded to a sheath cell. These results suggest that ectopic cell divisions occurred when both Pros and Notch activities were reduced without forcing cell divisions by CycE overexpression. In the future sheath cell, the action of Pros appears predominant since reduction of the activity of the N pathway alone did not induce extra divisions after CycE overexpression.
In this study we have shown that in the bristle lineage terminal cells, the shaft cell and the neuron, but not the socket and sheath cells, undergo supplementary cell divisions after CycE overexpression. These supplementary cell divisions were not due to a cell-specific differential expression by neur>Gal4 of cycE. After CycE overexpression, CycE was detected in all cells of the cluster during the entire period analysed in neur>cycE pupae. This is in contrast with the absence of CycE detection in terminal cells under normal conditions [20]. However, between clusters, the level of CycE detected was variable in each cell type suggesting that the CycE accumulation was very dynamic (Figure S1). Nevertheless, we never observed a correlation between these variations and the cell-specific extradivisions reported. This absence of correlation was also found under the different genetic backgrounds used (Figure S1). In addition, we observed a similar cell-specificity using a heat-shock promoter construction to drive CycE. Taken together, these data show that the cell-specificity of the extra cell divisions observed was not due to an un-even CycE level driven by neur>Gal4. Furthermore, in previous studies, we have shown that all terminal cells are arrested in G1-phase [20]. This indicates that resistance to cycE overexpression was not due to a differential cell-cycle arrest in terminal cells of the sensory cluster. Finally, an ectopic S-phase was induced after CycE overexpresion even in cells that did not undergo extra divisions ([20], data not shown). This shows that the level of CycE driven by neur-Gal4 was sufficient to force cell-cycle progression in all cells. This suggests that inhibitory role of CKI, like Dacapo, was overridden by the accumulated levels of CycE. As such, these extra cell divisions are due to a bone fide behavioural difference of the shaft cell and the neuron. Indeed, we show that this difference is due to the action of the Notch pathway and/or Prospero in maintaining cell cycle arrest in the socket and the sheath cells. After CycE overexpresssion, (1) supplementary cell divisions were observed only in those terminal cells in which the Notch pathway is not endogenously activated, (2) supplementary cell divisions were observed in sheath cells and neurons in a pros17/+ background; (3) activation of the N pathway blocked the ectopic division of the shaft cell; (4) socket cells underwent an extra division after reduction in N pathway activity and (5) additional sheath cells underwent extra divisions when both N and Pros activity was reduced.
In this study, the proliferative capacity of terminal cells was analysed when cell divisions were forced by CycE overexpression under different Notch and Prospero backgrounds. The analysis was performed under conditions in which either no cell transformation per se was observed (as in pros null background or pros heterozygous pupae) or in which only a small proportion of cells were transformed (as after mild overexpression of Numb or Nintra). As such, the observed effects were considered to be the result of N and Pros acting directly on the maintenance of the state of cell cycle arrest rather than a modulation of cell fate acquisition. Furthermore, we never observed changes in bristle cell identities after CycE overexpression. This is in contrast to Drosophila neuroblasts, in which CycE seems to control cell identity independently of its role in the cell cycle [29]. Thus, in the bristle lineage system, CycE seems to act exclusively as a cell cycle regulator.
When N-activity was impaired and cell divisions were forced, both pIIa daughter cells divided like their mother, pIIa, each producing a shaft and a socket cell. These observations reveal that, just after birth, pIIa daughter cells are not yet committed to their final fate and retain pIIa characteristics. These results suggest that cell fate in pIIa daughter cells is acquired in a sequential manner. Initially, both pIIa daughter cells appear to be equivalent. Then, the N-pathway is rapidly activated in one daughter cell due to a N-signal originating from its sister cell [30]. The activation of the N-pathway arrests the cell cycle and this cell then acquires a socket terminal fate. Later on, the future shaft cell is committed, by a cell cycle independent and as yet unknown mechanism, to its terminal cell fate and becomes post-mitotic. Taken together, our results suggest that initially, precursor cell division is self-renewed and that an active process is required to commit both daughter cells to their final identities.
Our results show that 15 minutes after birth, pIIa daughter cells are committed to their normal fate and do not respond to CycE overexpression. This suggests that extra divisions observed in these cells result from a delay in the cell cycle exit rather than a re-entry into the cell cycle. In contrast, cycE expression induced extra mitosis even in fully differentiated neurons that were identified by 22C10/Futsch immunodetection [15]. Interestingly, these extra mitoses were always symmetric and gave rise exclusively to neurons. Similarly, extra divisions of sheath cells, under conditions that impaired the activity of Pros and N, were mainly symmetric producing two sheath cells. This suggests that sheath cells, like neurons, can also re-enter the cell cycle long after being committed to their final fate. Similar mitotic capacity has been also observed in other differentiated cells, in particular in neurons, suggesting that terminal differentiation and cell cycle exit are distinct events [31]–[35]. Why pIIIb daughter cells only divide symmetrically and how this characteristic is related to the capability of these cells to retain proliferative capacities are open questions.
The fact that CycE induced extra divisions in terminal cells suggests that these cells are arrested in G1. Two main mechanisms trigger G1-arrest: repression of Cyclin/Cdk2 activity by CKI and repression of E2F activity by RB proteins (reviewed in [36]). Both of these mechanisms are involved in the maintenance of a quiescent state. Our data show that in both neurons and future shaft cells of the sensory organs, a high level of CycE alone is sufficient to induce extra mitoses. This is in contrast to differentiated neurons of the eye and the anterior wing margin where extra mitoses after sustained cycE expression were induced only in a rbf1 mutant background [6]. We suggest that, in our conditions CycE overexpression override the action of downregulateurs like Rbf1. Alternatively, factors other than Rbf1 can regulate cell cycle exit and the quiescent state maintenance in neurons and sheath cells. Interestingly, sensory organs containing two neurons were observed in null dap background (AA unpublished results). Thus, Dap is involved in neuronal cell arrest. Since Dap is also expressed in shaft cells [20], this factor probably plays a similar role in these cells.
Our data show that Prospero, together with N, cooperate in maintaining a quiescent state in sheath cells. Without overexpression of CycE, ectopic cell divisions occurred resulting in supplementary pIIIb daughter cells when Pros and N activity was reduced. Since the future neuron is a N-off cell, we expected that this cell will not be affected by the reduction in the N-function. As such, we anticipate that the future sheath cell undergoes a supplementary division. Thus, clusters containing two sheath cells and those containing two neurons would result from symmetric and asymmetric cell divisions respectively. Since these data were obtained in a non-sensitized condition, these results suggest that Pros and Notch are actively involved in maintaining a quiescent state in terminal cells during the normal development of bristles.
Between Pros and Notch, the former appears to predominate over the later to restrict cell proliferation. Indeed, in pIIIb daughter cells, CycE overexpression induced few extra divisions when N activity was reduced and significantly increased extra divisions in pros heterozygous cells. A role for Pros in maintaining cell cycle arrest is in agreement with the observation that neuronal proliferation was increased in pros loss of function embryos. It has been shown that this proliferation was associated with both an upregulation of cycE, stg and e2f genes [25] and a delay in the appearance of dap transcripts [26]. In our system, cell cycles were not resumed in pros null clusters per se. In addition, we observed only a mild increase in CycE expression and no decrease in Dap expression inside pros null clones. It is likely that in the absence of Dap downregulation, the increase in CycE levels is insufficient to induce extra mitosis.
N-signalling is pleiotropic and either promotes or represses cell cycle progression depending on the cellular context [37]. Thus, in Drosophila eye development, N-activity is necessary and sufficient to trigger cell cycle progression in G1 arrest in cells of the morphogenetic furrow, by derepressing the inhibition of E2F1 by RBF1 [38],[39]. Furthermore, glial precursor cells are maintained in an undifferentiated proliferative state by both N pathway activation and Dap downregulation [40]. In contrast, in follicle and wing cells, the N pathway has an anti-mitotic action [24],[41]. In follicle cells, N activity stops mitotic cycles and promotes endocycles by repressing the expression of both String/cdc25 and Dap and upregulating Fizzy-related/Cdh1 expression [23],[24]. In wing disc cells, N promotes G1-arrest by reducing E2F activity [41],[42]. The mechanism involved in this G1-cell cycle arrest seems to involve downregulation of the dmyc proto-oncogene and the bantam micro-RNA, both of which act positively on E2F activity [42]. If a similar mechanism is involved in the N-mediated G1-arrest observed in socket and sheath cells, we anticipate that dmyc and bantam would be downregulated exclusively in N-on bristle cells, in particular in pIIa, sheath and socket cells. However, the analysis of dmyc and bantam expression in sensory organ cells was not consistent with this idea. Unexpectedly, bantam and dMyc were detected in N-on cell such as pIIa, socket or sheath cells (AA, not shown). Thus, the mechanisms by which N and Prospero maintain cell cycle arrest in terminal cells of the bristle organ remain to be elucidated.
In conclusion, our results demonstrate that fate determination factors such as Notch and Prospero participate in maintaining a quiescent state in terminal cells. Under normal conditions, several mechanisms act in concert to ensure cell cycle arrest. Dap appears as a plausible candidate to ensure this role in N-off cells like the neuron and the shaft cell. N cooperates to maintain a quiescent state in sheath and socket cells and, finally, Pros acts only on the sheath cell. Consequently, the terminal quiescent state and cell differentiation do not seem to be regulated by mutually exclusive mechanisms. We favour the notion that these phenomena are regulated by parallel mechanisms involving factors having dual actions on fate acquisition and cell cycle progression.
The neur-GAL4 driver was used to specifically express H2B::YFP [14]; Partner Of Numb::GFP (PON::GFP, [17]), and CycE were expressed using the UAS/Gal4 system [13]. CycE overexpression was carried out using a line harbouring two copies of the UAS-CycE construct, one on the second and one on the third chromosome [5]. To increase viability of the neur>CycE and neur>CycE, pros17/+ pupae, overexpression was performed using a strain bearing GAL80ts (gift from D. Coen). Fly crosses, embryonic and larval development were carried at 18°C, and white pupae were transferred to 30°C to allow the expression of GAL4. Overexpression induced by heat-shocks were performed using the hs-CycE (Bloomington); hs-numb [21]; hs-Nintra [22] lines. In order to precisely stage pupae, pupae were collected at puparium formation and timed while considering that the developmental time at 18°C was twice longer than that at 25 or 30°C. Heat shocks were performed at 34°C or 38°C for 30 min and pupae were kept at 25°C for recovery. Somatic clones were obtained using the FLP/FRT recombination system [43]. The FRT82B pros17 [27] line alone or combined with Nts-1 [28] was crossed to the y, w,UbxFLP; FRT82B ubi-nls::GFP (gift of J. Knoblich) to generate pros-null somatic clones.
Live imaging of the bristle lineage in neur>CycE, UAS-H2B::YFP, UAS-PON::GFP pupae was carried out as described previously [11]. Images were acquired every 3 minutes on a confocal microscope (20× or 40× objective) driven by Metaview (Universal Imaging). Temperature was maintained at 25°C. Time-lapse movies were assembled using ImageJ (free software). Sensory clusters were identified according to their relative positions on the thorax. For each mitosis, asymmetric localisation of the PON::GFP fusion protein allowed the identification of the daughter cells. Live imaging in experiments combining time-lapse imaging and immunodetection was realised using an spinning disc microscope (Ropert Scientific France) (20× or 40× objective) driven by Metaview (Universal Imaging). Images were acquired every 4 minutes. Temperature was controlled by a thermo-regulated chamber (home-made).
Pupal nota were dissected between 17 h and 35 h APF and processed as previously described [9]. Primary antibodies were: mouse anti-Cut (DSHB, 1∶500); rat anti-CycE (gift from H. Richardson, 1/1000); rabbit anti-Dap (gift from C. Lenher, 1∶300); rat anti-ELAV (DSHB, 1∶10); mouse anti-ELAV (DSHB, 1∶100); mouse anti-Futsch (22C10) (DSHB, 1∶100); rabbit anti-GFP (Interchim, 1∶1000); mouse anti-GFP (Roche, 1∶500); mouse anti-Pros (gift from C. Doe 1∶5); rat anti-Su(H) (gift from F. Schweisguth, 1∶500); rabbit anti-phospho-Histone H3 (Upstate, 1∶10000). Alexa 488- and 568-conjugated secondary anti-mouse, anti-rat, anti-rabbit, anti-guinea pig were purchased from Molecular Probe and used at 1∶1000. Cy5 conjugated antibodies anti-mouse, -rat or -rabbit were purchased from Promega and were used at 1∶2000. In addition to antibody immunodetection, we also used other criteria to identify cells. (1) Nuclear size, bigger in pIIa daughter cells than in pIIb daughter cells. (2) Cell location relative to both the antero-posterior axis and the cell arrangement into the cluster, socket cell posterior to shaft cell and both cells posterior to pIIb daughter cells. (3) Relative YFP intensity, neuron nucleus is less intense than that of the sheath nucleus, (4) Small and bright YFP staining, reflecting apoptotic DNA condensation, to distinguish the glial cell. Images were processed with ImageJ and Photoshop. Counting of terminal cells was carried on-fixed pupae and was restricted to the clusters forming the two middle rows (24 and 28 h APF or after one night recovery when a heat shock was applied).
|
10.1371/journal.pgen.1005629 | A Point Mutation in Suppressor of Cytokine Signalling 2 (Socs2) Increases the Susceptibility to Inflammation of the Mammary Gland while Associated with Higher Body Weight and Size and Higher Milk Production in a Sheep Model | Mastitis is an infectious disease mainly caused by bacteria invading the mammary gland. Genetic control of susceptibility to mastitis has been widely evidenced in dairy ruminants, but the genetic basis and underlying mechanisms are still largely unknown. We describe the discovery, fine mapping and functional characterization of a genetic variant associated with elevated milk leukocytes count, or SCC, as a proxy for mastitis. After implementing genome-wide association studies, we identified a major QTL associated with SCC on ovine chromosome 3. Fine mapping of the region, using full sequencing with 12X coverage in three animals, provided one strong candidate SNP that mapped to the coding sequence of a highly conserved gene, suppressor of cytokine signalling 2 (Socs2). The frequency of the SNP associated with increased SCC was 21.7% and the Socs2 genotype explained 12% of the variance of the trait. The point mutation induces the p.R96C substitution in the SH2 functional domain of SOCS2 i.e. the binding site of the protein to various ligands, as well-established for the growth hormone receptor GHR. Using surface plasmon resonance we showed that the p.R96C point mutation completely abrogates SOCS2 binding affinity for the phosphopeptide of GHR. Additionally, the size, weight and milk production in p.R96C homozygote sheep, were significantly increased by 24%, 18%, and 4.4%, respectively, when compared to wild type sheep, supporting the view that the point mutation causes a loss of SOCS2 functional activity. Altogether these results provide strong evidence for a causal mutation controlling SCC in sheep and highlight the major role of SOCS2 as a tradeoff between the host’s inflammatory response to mammary infections, and body growth and milk production, which are all mediated by the JAK/STAT signaling pathway.
| Mastitis is an inflammation of the mammary gland mainly caused by invading bacteria. Ruminants show natural variability in their predisposition to mastitis, and therefore provide unique models for study of the genetics and physiology of host response to bacterial infection. A genome-wide association study was conducted in a dairy sheep population for milk somatic cell counts as a proxy for mastitis. Fine mapping, using whole genome sequencing, led to the identification of a mutation in the Suppressor of Cytokine Signaling 2 gene (socs2). This mutation was shown to cause a loss of functional activity of the SOCS2 protein, which suggested impairment of feedback control of the JAK/STAT signaling pathways in susceptible animals. Additionally, size, weight and milk production were increased in animals carrying the susceptible variant suggesting a pleiotropic effect of the gene on production versus health traits. Results gave strong evidence of the role of SOCS2 in the host’s inflammation of the udder and provided new insights into the key mechanisms underlying the genetic control of mastitis.
| Mastitis is an inflammation of the mammary gland mainly caused by bacteria, which develop in the gland cistern after penetration through the teat canal. Mastitis is the main infectious disease of dairy ruminants, with respect to industry and public concern, economic impact, zoonotic potential and animal welfare [1,2]. This disease has occasionally been reported in breast-feeding women [3,4].
How mammals defend against a microbial intra-mammary infection is still poorly understood. Animal models, especially ruminants, have provided useful information about the mechanisms underlying immunity. Mammary defense depends on the early recognition of invading pathogens by sentinels, such as the dendritic cells and macrophages, and also mammary epithelial cells [2]. Early activation leads to the production of soluble factors like cytokines and chemokines that are able to recruit blood cells into the parenchyma and milk to fight the infection by phagocytosis and bacterial killing. This process is finely tuned to avoid collateral damage to the secretory epithelia. Literature data [2,5] corroborate the importance of a rapid influx of neutrophils into the mammary gland, to control the inflammatory process and allow effective and early elimination of the pathogens.
The existence of a genetic basis for mastitis resistance has been well documented in dairy ruminants [6,7,8,9]. Initially, evidence was essentially based on quantifying the polygenic variation of indirect predictors of udder health that could be measured on a large scale in dairy operations. One of the most widely studied predictors was milk somatic cell count (SCC). Indeed, the milk SCC mainly reflects the number of neutrophils that migrate from blood to the mammary gland in response to infection. Measured on a monthly basis, SCC can be interpreted as indicating the consequences of infection and repeatedly high SCC can be associated with the presence of chronic mastitis. In sheep, presence of major pathogens (staphylococci, streptococci, and enterococci organisms) in milk was strongly related to a sharp inflammatory response with increased log SCC means when compared to uninfected half udder [10,11]. Further, Albenzio et al. [12] added to the evidence that immune competence of the mammary gland is related to levels of somatic cell and presence of pathogenic bacteria in ewes with subclinical mastitis. Genetic studies, as reviewed in [2], showed that the estimated heritability for SCC ranged from 10 to 20%, indicating that much of the variation is of genetic origin. The trait is also genetically strongly correlated with resistance to mastitis. This has been further demonstrated in a divergent selection experiment in dairy sheep based on extreme values of log-transformed SCC, i.e. SCS [13]. Measurements of the frequency and duration of bacteria in milk showed that Low-SCS and High-SCS ewes have lower and higher rate of intra mammary infections in natural conditions, respectively. Additionally, bacteriological titer after experimental challenge [14] in the two genetic lines (High vs Low- SCS) demonstrated that bacterial clearance is more efficient in the Low-SCS ewes than in the High-SCS ewes. Accordingly, although SCC and resistance to mastitis are not exactly the same trait (genetic correlation is not 1), genes and mechanisms that underlie those traits are partially common.
The availability of genome sequencing data has opened up new fields of investigation that can be applied to domestic ruminant species with already-sequenced genomes [15,16]. Indeed, the development of high-density single nucleotide polymorphism (SNP) arrays and their application in genome-wide association studies has facilitated the identification of regions controlling production and health traits. These approaches have recently led to the localization of several regions of the genome responsible for some of the variability for udder health traits [2]. However, to our knowledge, except for the direct association with the MHC locus, to date only one QTL for mastitis related traits has been fully described. Indeed, Sugimoto et al. [17] showed that a polymorphism of the bovine forebrain embryonic zinc finger-like gene (FEZL), located in the region of a QTL for SCC on bovine chromosome BTA22, was associated with high and low SCC. Susceptible animals displayed lower expression of FEZL and consequently lower expression of a number of cytokines including TNF-alpha and IL-8. This down-regulation of cytokines was mediated by lower SEMA5A expression.
We mapped the genetic loci determining resistance/susceptibility to SCC as a proxy for mastitis by performing a large-scale QTL analysis using an outbred population of dairy sheep. Following discovery of such loci using a 50K SNP chip, we used fine mapping and whole genome sequencing for molecular dissection of a major region identified on ovine chromosome 3. A possible candidate mutation in the Socs2 gene was validated by implementing i) genotyping and analysis of the mutation in the discovery population, ii) functional analysis of the isoforms of the SOCS2 protein and, iii) phenotyping of animals carrying contrasted genotypes at the Socs2 gene. Our results provide new insights into the key mechanisms underlying the genetic control of the host’s response to intra mammary infections.
QTLs associated with SCC were mapped by performing a genome scan (26 autosomes) in 1009 dairy sheep distributed in 33 half-sib families. The SCC trait pertaining to mastitis susceptibility was the lactation average of somatic cell score (LSCS) as this trait is highly correlated with intra mammary infections. The phenotype was the daughter yield deviation provided by the routine genetic evaluations. These represent the average performance of the daughters of a sire, corrected for the environmental effects and the genetic value of the mates.
All animals were genotyped with the 50K OvineSNP50Beadchip (Illumina, San Diego, CA). Haplotype-based linkage and association analyses were used to detect QTLs on 22 chromosomes at a 5% chromosome-wide threshold (S1 Table). Five regions on chromosomes OAR3, 4, 11, 16 and 23 exceeded the 5% genome-wide threshold. One highly significant QTL on chromosome 3 (OAR3) was similarly located in the two association and linkage analyses (Fig 1A and 1B; S1 Table). For this OAR3-QTL, the association study provided haplotypes of contrasting susceptibility based on four consecutive SNP: ss836339510—ss836339511—ss836339512—ss836339513, spanning a length of 416kb (129,685,397bp—130,103,393 bp).
The OAR3-QTL was subjected to further fine mapping using whole genome sequencing in a trio of rams (average read-depth of 12X). This trio included a segregating sire carrying the most susceptible haplotype Q (Qq), and two homozygous sons for alternative haplotype alleles (QQ and qq), whose progenies were extremely divergent for the SCC phenotype. Out of the total of 1543 SNPs found in a region of 0.5 Mb, 207 SNPs were retained that were heterozygous (0/1 genotype) for the sire and homozygous, with different genotypes (0/0 or 1/1) for the sons and annotated with the current gtf file for sheep genome (ftp://ftp.ensembl.org/pub/current_gtf/ovis_aries/) (S2 Table). The distribution of the 207 SNP over the OAR3 QTL region is given in Fig 1C. These 207 variations were then submitted to the NCBI databases, dbSNP and dbVar. SNPs were released in the dbSNP Build (B143). Sixty-six SNPs were homozygous for the reference genome base in the susceptible son (0/0), whereas 141 SNPs were homozygous for the alternate base in the susceptible son (1/1). We focused our interest on the variants located in coding or untranslated regions (UTR). Two SNPs were located in the 3’UTR of CRADD gene. Neither of the 3’UTR variants appeared to be conserved, based on GERP scores [18]; the difference between observed and expected GERP scores was -2 for the 130009096 position and -9 for the 130009121 position. Only a single SNP mapped to the coding region of a gene with a non-synonymous change in an amino acid. Modification of the C base in the reference sequence (OAR3, 129722200 bp) to a T in susceptible animals encoded an arginine to cysteine substitution at position 96 (p.R96C) of the Suppressor of Cytokine Signaling 2 (SOCS2) protein. It is noteworthy that this sequence of the SOCS2 protein is highly conserved across species, especially in the region surrounding the mutation (Fig 2A and 2B). The predicted impact of the mutation on the 3D-structure of the protein is shown in Fig 2C and 2D based on a model predicted by Homotopy Optimization Method, HOPE. The mutation is located within the SH2 domain and replaces arginin, a polar positively charged amino-acid by a cystein which is a small and mostly hydrophobic amino-acid. This change can disturb the tri-dimensional structure of the domain and abolish the protein functions.
It has been previously shown that human SOCS2 interacts with endogenous receptors, such as growth hormone (GHR) or erythropoietin (EpoR) receptors, and that this interaction occurs in a phosphorylation-dependent manner at pY595 and pY401, respectively, with the SH2 domain of the protein [19]. We performed biomolecular interactions of recombinant ovine wild type (WT) and p.R96C SOCS2 proteins with phosphorylated peptide from the GHR fixed to a biosensor chip, to compare the rate constants and affinity binding of the two isoforms. When a high concentration of pY-GHR peptide was bound to the support, the observed difference in protein binding was very large (Fig 3A). A similar difference was confirmed after single-cycle kinetics using a wide range of protein concentrations (Fig 3B). Rate constants and affinities for binding of rOvSOCS2 WT on immobilized pY595 GHR were the following: Ka = 8.28 X 102 M-1 S-1, Kd = 1.6 X 10−3 S-1, Kt = 3.06 X 107 RU M-1 S-1 and KD = 2.0 μM. KD were close to those previously published for rhuSOCS2, i.e. 2 vs 1.6 μM [19], an affinity value could not be determined for the p.R96C isoform due to the lack of interaction. These results show that the p.R96C mutation precludes SOCS2 binding to the phosphorylated peptides from at least two endogenous receptors, and is probably associated with a loss of SOCS2 functional activity, suggesting that the p.R96C mutation is associated with a functional knock-out of the Socs2 gene in T/T homozygous sheep.
A PCR genotyping test (KASPar) was implemented to genotype 468 rams in the discovery population for the C/T (p.R96C) mutation in the Socs2 gene. The frequency of the T mutation, associated with increased inflammation (i.e. elevated SCC), was 21.7%. The frequencies of wild-type (C/C), heterozygous (C/T) and homozygous (T/T) carriers were 58.5%, 40% and 1.5%, respectively. Analysis of variance confirmed that the T mutation involved a dramatic increase in cell counts in the progeny of these rams, especially in homozygous animals, as shown by the average sire breeding values in Fig 4A. Overall, considering the frequency of carrier animals, the Socs2 genotype explained 12% of the variance of the SCC trait in this population.
To explain the high frequency of the deleterious T mutation in this dairy sheep population, we hypothesized that the Socs2 genotype might be associated with a beneficial effect on other traits under selection. Indeed, heterozygous C/T rams displayed significantly (p < 0.05) higher progeny milk yields than C/C animals (Fig 4B). Assuming an average milk yield of 280 L and genetic standard deviation of 36.5L (JM Astruc, Institut de l’Elevage, personal communication), the milk increase in homozygous (T/T) carriers is equivalent to + 4.4% when compared to the wild type (C/C).
It has been previously shown that Socs2 knock out mice exhibit an unusual gigantism phenotype (Metcalf et al., 2000). We therefore analyzed the growth curve and size of half-sister ewes in relation to their Socs2 genotype to see if the p.R96C mutation impaired the function of the SOCS2 protein associated with body development. The weights of sheep carrying the T mutation were very much higher than their wild type counter-mates from 17 months of age onwards, i.e., after the end of the first lactation (Fig 5). The average difference between TT and CC was 16Kg at 3 years of age, which was equal to 18% of the average body weight. Similarly, body morphometric measurements showed that 8 out of 14 measures were significantly impacted by the Socs2 genotype (Fig 6). These included a notable increase in height, width and bone length (tibia) of sheep carrying the T mutation. The average significant difference between TT and CC was 9.2%, and up to 24% for height at elbow. The T mutation in the Socs2 gene showed an additive effect for both weight and size, the data for heterozygous C/T sheep being mostly intermediate between those of C/C and T/T sheep. These results suggest that SOCS2 is effectively disabled in p.R96C sheep, with regard to well established effects of the gene, and provide a further explanation for the balancing effect of the mutation.
A highly significant QTL for SCC, as a proxy for predisposition to mastitis, was mapped on ovine chromosome 3 (v3: 129,685,397bp—130,103,393 bp). It is interesting to note that among the numerous QTL regions for SCC-based trait reported in the ruminants literature [2] or in the QTL data base (http://www.animalgenome.org/cgi-bin/QTLdb), a dairy cattle study [20] also pointed to the syntenic region on the cattle chromosome BTA5 (23.5–23.9 Mbp) for clinical mastitis occurrence (14.4–26.7 Mbp) and SCS (26.7–30.5 Mbp). Rodriguez-Zas et al. [21] also reported a QTL for SCS on BTA5 (about 7.7 Mbp; http://www.animalgenome.org/cgi-bin/QTLdb). Although the confidence intervals associated with the latter QTLs were large, the present study provides a noteworthy candidate for these QTLs. More generally it appears appealing to systematically screen for mutation in socs2 in various livestock species and ruminants in particular.
We have presented several lines of evidence that the p.R96C mutation of Socs2 is responsible for this QTL for SCC and explains significant differences in levels of udder inflammation in the studied dairy sheep population. First, by genotyping the mutation in the discovery population, we showed that its frequency was high, i.e., 21.7%, and that the Socs2 genotype explained up to 12% of the variance of the SCS trait. Second, amongst 207 candidate SNPs, the Socs2 mutation was the only SNP in the coding region of a gene. This mutation causes an amino acid change in the SH2 domain that is highly conserved across species. Indeed, SOCS2 belongs to a family of 8 proteins: CIS, for cytokine inducible SH2 domain containing protein and SOCS1 to 7 for suppressor of cytokine signaling [22,23]. The proteins in this family share a common structural analogy with a central Src homology 2 (SH2) domain, a C-terminal domain called the SOCS box and a variable N-terminal domain [24,25]. Whereas the C- and N-terminal regions show some variation, the sequence of the SH2 and SOCS box domains of the SOCS proteins is extremely conserved across species, thus suggesting that any non-synonymous sequence variation may have important consequences on the functional role of the protein. Finally, a functional assay for wild type and p.R96C mutated SOCS2 protein showed that the p.R96C mutation prevents SOCS2 binding to phosphorylated peptides from at least two endogenous receptors, and is probably associated with a loss of SOCS2 functional activity. This result suggests that the p.R96C mutation causes with a functional knock-out of the Socs2 gene.
Other identified SNP were not examined in additional detail after finding the non-synonymous variant in Socs2, supported with functional evidence. Especially, there were two 3’UTR SNPs in the CRADD gene but neither of these variants appear conserved (GERP scores). We cannot exclude, however, that some of those other SNP within the 207 identified may exert a minor role beyond the p.R96C substitution. Genotyping those SNP in independent populations of dairy sheep might help validating the results and refining the relative contribution of each variant to the trait expression.
Many studies have documented the essential role of SOCS proteins (CISH, SOCS1-7) in the regulation of cytokine, growth factors and hormones such as prolactin, growth hormone and erythropoietin [22,23,26]. Proteins In the SOCS-family play roles as negative feedback regulators for those cytokines and growth factors by switching off the Janus Kinase (JAK)/signal transducers and activators of the transcription (STAT) signaling pathway. Gene expression in the nucleus is therefore reduced, resulting in negative control of the cytokine-mediated signal. SOCS1 and SOC3 can directly inactivate the JAK function through a specific region called KIR [22]. The SH2 domain of all SOCS proteins, including SOCS2, binds phosphorylated tyrosine residues in the cytoplasmic tails of cytokine receptors. It is therefore involved in stimulation by ligands and in the downstream competition with STATs. The mutation identified herein, located in the SH2 domain of Socs2, most probably prevents activation of SOCS2 by ligands and limits direct or indirect competition with STATs.
The results suggest that a functional knock-out of Socs2 causes an in vivo uncontrolled inflammatory response. This hypothesis is in agreement with the QTL phenotype, i.e., high milk SCC that reflects chronically elevated infiltration of white blood cells in milk of animals carrying the p.R96C mutation in the Socs2 gene. Because milk SCC are strongly associated with intra mammary infections in sheep [10,11,12] and as both traits are genetically correlated in sheep [14,27], we can hypothesize that elevated SCC associated with the p.R96C mutation may actually be due to more frequent or severe infections, increased mastitis susceptibility and some other immunological dysfunction. However, additional data on health status and immune response between mutant and wildtype animals are needed to confirm this hypothesis. A few other published data corroborate the association of dysfunction of SOCS2 with an impaired immune response. Machado et al. [28] showed that Socs2 -deficient mice exhibited in vivo “uncontrolled production of proinflammatory cytokines “(peritoneal CCL2), “decreased microbial proliferation, aberrant leukocyte infiltration and elevated mortality” upon infection with Toxoplasma parasite. In their study, Posselt et al. [29] demonstrated the role of SOCS2 in the counter-regulation and limitation of inflammatory activity of dendritic cells after TLR stimulation. Additionally, SOCS2 appears to be a late-induced SOCS protein [29,30] and has been shown, like SOCS6 and SOCS7, to interact with the SOCS box of other members of the SOCS family, and to be involved in a cross-talk regulation of other SOCS proteins [23,30]. One possible explanation for the role of SOCS2 in immune homeostasis and disease may therefore be the presence of defects in the ability of SOCS2 to regulate SOCS3 expression [30]. Finally, considering the importance of the JAK–STAT signaling pathway in host immunity, Usman et al. [31] investigated single nucleotide polymorphisms in the STAT5A and JAK2 genes in association with mastitis indicator traits (SCC) and some serum cytokines and production traits in Chinese Holstein cattle. They found SNPs in the two genes that were significantly associated with cytokine IL-6, IL-17 and with SCC. Altogether these studies support our finding that genetic variants in the genes involved in the JAK/STAT/SOCS pathway can alter actual functions in this signaling pathway and significantly contribute to the control of inflammatory response to infection. Which biochemical mechanisms, cytokine and associated JAK/STAT molecules are modified by this reported dysfunction of SOCS2, however, remains elusive and needs to be addressed in further studies.
We found a high frequency of the deleterious mutation p.R96C in the population examined and a significant association with higher milk production, growth and size, which raises the question of a balancing selection for increased frequency of p.R96C due to beneficial pleiotropic effects. The large scope of signalling pathways controlled by the SOCS2 protein upholds a pleiotropic effect of the mutation on both immune response and growth and production in the lactating mammal. Indeed, while SOCS2 has been associated with control of immune response and disease outcome in the few above-mentioned studies, SOCS2 is also a well-established negative regulator of GH signalling [22,23,26,32,33] and prolactin signalling [22,23,26]. On the one hand, two models of Socs2 knockout mice [34] and transgenic mice overexpressing SOCS2 [32] revealed an enlargement of most organs, increased bone growth, weight gain and size in modified mice compared to their wild-type littermates. The results reported by authors, for both models, reinforced the hypothesis that the underlying mechanisms involved suppression of GH signaling by SOCS2 in wild-type animals [32,33,34] mediated by STAT5b [33]. Growth and bone size were both significantly increased in our homozygous p.R96C SOCS2 sheep, substantiating a direct effect of the mutation mediated by the GHR/JAK/STAT signaling pathway. GH is also known to affect mammogenesis and increase milk production in dairy ruminants. Indeed, intra-mammary administration of GH in cattle [35], goat and sheep tends to stimulate milk production during lactation. Zhang et al. [36] showed that udder-size was increased in transgenic goats overexpressing GH in the mammary gland, compared to the controls, suggesting a larger capacity for milk production. Additionally a polymorphism in the GHR has been shown to explain, at least in part, a major QTL for milk production in Holstein-Friesian cattle [37]. In Sarda sheep, significant differences in milk traits were observed among genotypes at polymorphic GHR loci [38]. On the other hand, prolactin (and progesterone) is necessary for the initial development of the alveolar bud of the mammary gland and for further differentiation during secretory initiation and activation, as showed by mouse [39] and cow [40] prolactin/prolactin receptor mutants. Additionally, using gene expression profiling, Harris et al. [41] showed that SOCS2 (and E74-like factor 5) counterbalanced the defects in mammary gland development produced in two prolactin-deficient models, thereby demonstrating the role of SOCS2 in the control of mammary gland development. These results imply a potential direct effect of the p.R96C mutation in the Socs2 gene on mammary gland development and physiology, and on subsequent milk production, through modified feedback control of GH and the prolactin signaling pathways. More detailed analyses of growth hormone, prolactin and cytokine signaling and SOCS proteins expression are needed to further clarify the role of SOCS2 in the immune response and on production traits in our Socs2-deficient sheep model.
To our knowledge, this is the first report of a mutation with pleiotropic effect in which a key molecule is identified as a potential tradeoff between health and production traits in dairy ruminants. Similarly Fasquelle et al. [42], found a mutation in the MRC2 gene which accounts for the outbreak of the Crooked Tail Syndrome in Belgian Blue Cattle while associated with enhanced muscular development in the general population. Further, Kadri et al. [43] showed that a 600-b deletion accounted for antagonistic effect on fertillity and milk yield in dairy Nordic Red Cattle. Both health and fertility deleterious variants were found at high frequency most probably because of selective advantage on production trait. The present results may, at least partly, explain the adverse genetic correlation between SCC and milk production that exists in our sheep population, i.e., 0.18 between SCS and milk yield in first lactation [13]. Such a genetic opposition has been widely documented in other dairy ruminant species [2,6,7,8,9] and corroborates the earliest genetic studies which provided evidence that the highly successful selection for milk production had probably led to a deterioration of mastitis resistance in cattle [44]. The genetic opposition between health and production caused by such genetic variants with pleiotropic effect can therefore not be disrupted by conventional- or SNP- assisted selection. We will therefore need to address the issue of how fast to minimize the frequency of the unfavorable Socs2 genotype (if relevant) while minimizing the loss of genetic progress regarding production traits. The present finding and above mentioned cattle literature [42,43] add to the evidence that pleiotropy and tradeoffs are not uncommon in bred livestock. They highlight the need for knowledge and methods on how to achieve optimal balancing selection on both health and production traits in livestock, when specific variants with pleiotropic effect are identified.
We have identified a mutation in the suppressor of cytokine signaling 2 (Socs2) gene that likely underlies a QTL for SCC in a dairy sheep model. This mutation is associated with persistently high milk cell counts, indicating chronic inflammation of the mammary gland.
The mutation modifies an amino acid (p.R96C) in the SH2 domain of SOCS2 and prevents binding to phosphorylated peptides from at least two endogenous receptors. Although the biochemical mechanisms by which SOCS2 alters the host’s response remain unknown, the fact that SOCS2 is known to play a key role in the negative feed-back of the cytokine-meditated response via JAK/STAT signalling pathways, supports the idea that our Socs2-deficient sheep fail to control the inflammatory process in response to intramammary bacterial infection, leading to impaired resistance to the disease.
The Socs2 mutation was also associated with increased milk production, growth and size, which suggests a pleiotropic effect due to impairment of retro-control of the GH and prolactin signalling pathways. Altogether these results uphold further detailed analyses of the effect of SOCS2 on the immune response and production traits in our Socs2-deficient sheep model.
For association mapping, data from a total of 1009 commercial French dairy rams were used. These 1009 males were distributed in 33 half-sib families in a so called grand-daughter design. Family size averaged 30.2 (±9.7) sons and ranged from to 18 to 54.
The phenotypic measurement was the milk somatic cell count (SCC) measured in lactating ewes and available from the national data base, (Centre de Traitement de l’Information Génétique, CTIG, Jouy en Josas, France), as well as pedigree information, as part of the official data system for livestock (ministerial order NOR: AGRT1431011A, 24th March 2015, Ministry of Agriculture, France). Sampling methods and SCC analyses followed standard recommendations by the IDF (International Dairy Federation) and ICAR (International Committee for Animal Recording). SCC was measured on average three times per lactation in first and second parity. SCC was then log-transformed to somatic cell score [SCS = log2 (SCC/100) +3] to normalize the data distribution and averaged per lactation to compute the analyzed trait LSCS as described in [9]. For association mapping we used twice the daughter yield deviation (DYD) [45] for LSCS from the national genetic evaluation procedure [46]. DYDs correspond to the average performance of the daughters of a ram, corrected for the environmental effects and the genetic value of the dams.
All 1009 rams were genotyped using the Illumina Ovine SNP50 BeadChip assay. DNA extraction from blood samples and genotyping were performed at the Laboratoire d’Analyses Génétiques pour les Espèces Animales, Jouy en Josas, France (LABOGENA; www.labogena.fr). Blood samples were collected as part of the official national preservation collection and the coded samples stored at LABOGENA. Data were cleaned using in-house pipelines automated for the whole dairy sheep population genotyping data. In brief, any individual with a call rate below 98% or showing pedigree inconsistency had been previously discarded (less than 3%). SNP quality control included the following inclusion criteria: call rate above 97%, minor allele frequency above 1% and Hardy-Weinberg P-value above 10−6. After edits, a total of 41,501 autosomal SNPs, distributed on ovine chromosomes OAR1 to OAR26 were included for further analyses. The marker order and positions were based on the Ovine Assembly v3.1 (http://www.livestockgenomics.csiro.au/sheep/oar3.1.php).
For discovery association, both linkage analyses (LA) and genome-wide association study (GWAS) were applied to the data using the QTLMap software (Elsen et al., 1999; http://dga7.jouy.inra.fr/qtlmap/). For LA, interval mapping [47] was performed by likelihood ratio test (LRT) using within-sire linear regression [48]. The QTL effect (average substitution effect) was expressed in deviation units (SD) for the trait. GWAS was based on a regression analysis of the phenotypes on founder sires’ haplotypes for every haplotype of 4 consecutive SNPs along the chromosome [49]. Chromosome-wise significance levels were calculated with QTLMap, using the current family structure and phenotypes. For LA, the empirical 5% and 1% chromosome-wise significance levels of the test statistics were estimated from 1000 within-family permutations [50] for each chromosome. For GWAS, the empirical chromosome-wise significance level of the test statistics was estimated from 1000 simulations for each chromosome. The genome-wise thresholds were obtained by applying the Bonferroni correction: Pgenomewise = (1 – Pchromosomewise)n, where n is the number of chromosomes, i.e., 26 in sheep [51]. The 95% confidence intervals of the QTL locations were estimated by logarithm of odds drop-off [47] implemented in the QTLMap software. In practice, the bounds of the interval were the 2 locations where the likelihood was equal to the maximum likelihood minus 3.84
(=χ(1,0.05)2)
.
One paired-end library with a 300 bp insert size was generated for each animal using the Illumina TruSeq sample prep v2 Kit (PN FC-121-2001). 3 μg of each DNA were fragmented on a Covaris M220 focused-ultrasonicator and used to generate the libraries on a Tecan EVO200 automate. Quality controls were assessed with the High Sensitivity DNA chip (PN 5067–4626) on a 2100 Bioanalyzer (Agilent), and libraries were quantified using the Kappa Library Quantification kit (PN KK4824) on an ABI 7900 HT (Life Technologies). Each library was paired-end sequenced (2x100 bp) on a single lane of an Illumina HiSeq2000 flowcell using the Illumina TruSeq SBS Kit v3 (FC-401-3001).
FASTQ sequences were aligned to the sheep genome v3 (http://www.ensembl.org/Ovis_aries/Info/Index) with BWA software, version 0.7.0-r313 [52] (“aln” algorithm with default settings). The resulting SAM format files were processed using samtools view, sort and merge functions [52]. Then, we applied duplicate removal (http://picard.sourceforge.net, version 1.91), GATK (version 2.4.9) base quality score recalibration [53], indel realignment, and performed SNP and INDEL discovery and genotyping across all 3 samples simultaneously using standard hard filtering parameters. snpEFF (version 3.3a) and snpSIFT [54,55] were used to classify the filtered-by-genotype variations according to their functions as synonymous, nonsynonymous, nonsense, missense, insertions, deletions or splice variations. Results were further compared with the previously generated Illumina Ovine SNP50 BeadChip genotypes. Within the QTL localization interval, SNP satisfying the following characteristics were filtered: heterozygous in the father, homozygous (two different genotypes in the sons). Polymorphisms were visualized using IGV software [56].
Genotyping of the Socs2 mutation was performed by allele-specific amplification using the KASPar SNP genotyping system, followed by fluorescence detection on an ABI7900HT. KASPar assays were carried out in 5 μL reactions according to the KBioscience published conditions (http://www.kbioscience.co.uk/). The primers used for this genotyping test are listed in S3 Table.
Peripheral blood mononuclear cells (PBMC) from a Socs2 heterozygous sheep were isolated from EDTA blood samples using centrifugation over Ficoll gradient, and cultured overnight in complete RPMI medium with 10% fetal calf serum. For cDNA preparation, a two-step purification of RNA was used with phenol-chloroform extraction followed by purification on silica columns (RNeasy mini kit, Qiagen). RNA integrity number (RIN) was determined using an Agilent 2100 bioanalyzer, and was above 6.5. cDNA was synthesized on 1 μg RNA with Superscript III reverse transcriptase (Invitrogen) following the manufacturer’s recommendations.
WT and p.R96C SOCS2 cDNA were cloned by PCR using primers OAR_Socs2-5’ and OAR_Socs2-3. PCR products were cloned in pCR2.1 plasmid using Strataclone PCR (Stratagene) and Sanger sequencing was performed with universal primers.
Sequences of each variant were transferred into prokaryotic vectors downstream of the Thioredoxin and 6xHis-tag sequences. Upon expression in BL21 E. coli strain, recombinant proteins were purified using His-tag and dialyzed overnight against a large volume of Tris-HCl 50 mM NaCl 150 mM pH 9 for refolding. Protein concentration was then determined using the Bradford reaction and purity of the recombinant proteins was monitored by SDS/PAGE and Coomassie blue staining.
All binding and kinetics studies based on surface plasmon resonance (SPR) technology were performed on a BIAcore T200 optical biosensor instrument (GE Healthcare). Immobilization of biotinylated phosphopeptides (Proteogenix) was performed on a streptavidin-coated (SA) sensorchip in HBS-EP buffer (10 mM Hepes pH 7.4, 150 mM NaCl, 3 mM EDTA, 0.005% surfactant P20) (GE Healthcare). All immobilization steps were performed at a flow rate of 2 μl/min with final peptide concentrations of 2.5 μg/ml and 25 ng/ml for 350 RU or 75 RU immobilized peptides, respectively. A channel (Fc2) was used for immobilization of GHR peptide and a channel with an immobilized scramble peptide (Fc1) was used as a reference surface for non-specific binding measurements. Binding analyses were performed with solubilized purified SOCS2-WT and SOCS2-p.R96C proteins at different concentrations over the immobilized peptide surface at 25°C for 2 minutes at a flow rate of 30 μl/min.
A single-cycle kinetics (SCK) analysis to determine association, dissociation and affinity constants (ka, kd, and KD respectively) was carried out by injecting different protein concentrations (100 nM–8.1 μM). Binding parameters were obtained by fitting the overlaid sensorgrams with the 1:1 Langmuir binding model of the BIAevaluation software version 1.0.
For the 468 rams from the discovery population that were genotyped with the KASPar test, data from the national data base CTIG were used to quantify the effect of the Socs2 genotype on milk production traits. A single SNP test of association by analysis of variance (ANOVA) was performed using the mixed procedure of the statistical analysis system (SAS) programs (9.1) for the 468 genotyped animals. The dependent variables were DYD for LSCS and production traits (Milk Yield, Fat and Protein content) available from the national genetic evaluation procedure. The three possible Socs2 genotypes were fitted as a fixed explanatory variable and a significance threshold of p<0.05 was selected. The varcomp procedure of SAS was used to fit the genotype effect as random and estimate the proportion of variance explained by the genotype. The sire was included as a random effect in both mixed and varcomp models.
A group of 18 female sheep with the three Socs2 genotypes were bred in the INRA experimental unit of La Fage (Causse du Larzac, 43°54'54.52′′N; 3°05'38.11′′E, Aveyron, France). The 18 sheep were triplets of Socs2-C/C,-C/T and -T/T genotype, each triplet being sired from a common ram, except one (C/T and T/T from common sire and a non-related C/C match). The sheep were born in 2010, 2011 and 2012. Body weight was measured to the nearest 0.1 Kg at 12 time points during a 3 year period, from birth to third lambing. Number of records per ewe was 11 on average and varied from 7 to 12. Repeated body weight data within animal were analyzed with a linear model using the mixed procedure of SAS (SAS Institute, Cary, NC). Fixed effects were Socs2-genotype within time point, year of birth and triplet match, whereas the animal was treated as a random effect. Body size was measured in 17 out of the 18 sheep once in October 2013 to obtain fourteen measurement points of height, width and bone lengths as illustrated in S1 Fig. Morphometric data were analyzed with a linear model based on GLM procedure from SAS (SAS Institute, Cary, NC) including the fixed effects of genotype, age at measure (1, 2 or 3 years old) and triplet match. A significance threshold of p<0.05 was declared for both analyses of weight and size.
Commercial rams did not belong to any experimental design but were sampled by veterinarians and/or under veterinarian supervision for routine veterinary care and DNA collection. For the experimental animals (INRA, Domaine de La Fage), breeding conditions were similar to commercial sheep flocks. Blood collection and measurements followed procedures approved by the Regional Ethics Committee on Animal Experimentation, Languedoc-Roussillon (France), under the Agreement 752056/00.
|
10.1371/journal.ppat.1007085 | Multiple roles of core protein linker in hepatitis B virus replication | Hepatitis B virus (HBV) core protein (HBc) contains an N-terminal domain (NTD, assembly domain) and a C-terminal domain (CTD), which are linked by a flexible linker region. HBc plays multiple essential roles in viral replication, including capsid assembly, packaging of the viral pregenomic RNA (pgRNA) into nucleocapsids, viral reverse transcription that converts pgRNA to the genomic DNA, and secretion of DNA-containing (complete) virions or genome-free (empty) virions. The HBc linker is generally assumed to act merely as a spacer between NTD and CTD but some results suggest that the linker may affect NTD assembly. To determine its role in viral replication, we have made a number of deletion and substitution mutants in the linker region, in either the presence or absence of CTD, and tested their abilities to support capsid assembly and viral replication in human cells. Our results indicate that the linker could indeed impede NTD assembly in the absence of CTD, which could be partially relieved by partial linker deletion. In contrast, when CTD was present, the linker deletions or substitutions did not affect capsid assembly. Deletion of the entire linker or its C-terminal part resulted in a partial defect in pgRNA packaging and severely impaired viral DNA synthesis. In contrast, deletion of the N-terminal part of the linker, or substitutions of the linker sequence, had little to no effect on RNA packaging or first-strand DNA synthesis. However, the N-terminal linker deletion and two linker substitution mutants were defective in the production of mature double-stranded viral DNA. Secretion of empty virions was blocked by all the linker deletions and substitutions tested. In particular, a conservative linker substitution that allowed mature viral DNA synthesis and secretion of complete virions severely impaired the secretion of empty virions, thus increasing the ratio of complete to empty virions that were secreted. Together, these results demonstrate that the HBc linker region plays critical and complex roles at multiple stages of HBV replication.
| The hepatitis B virus (HBV) is a major human pathogen that infects hundreds of millions of people worldwide and represents a major cause of viral hepatitis, liver cirrhosis, and liver cancer. The HBV capsid protein (HBc) plays multiple roles in the viral life cycle and has emerged recently as a major target for developing antiviral therapies against HBV infection. HBc is divided into three separate regions, an N-terminal domain (NTD) responsible for capsid assembly, a C-terminal domain (CTD) that plays critical roles in the specific packaging of the viral pregenomic RNA (pgRNA) into replication-competent nucleocapsids and the subsequent reverse transcription of the pgRNA into the viral genomic DNA, and a linker region between the NTD and CTD. In contrast to the prevailing assumption that the linker merely serves to connect the NTD and CTD, we have discovered here that it plays a critical role in almost every stage of HBV replication. The linker likely exerted its pleiotropic effects via affecting the NTD and CTD as well as via direct interactions with other viral factors independent of the NTD or CTD. Our results thus not only deepen understanding of HBc structure and functions but also implicate the linker as a potential novel target for antiviral development against HBV infection.
| Hepatitis B virus (HBV), a major cause of viral hepatitis, liver cirrhosis, and hepatocellular carcinoma [1], replicates a small (ca. 3.2 kb), partially double-stranded (DS), relaxed circular (RC) DNA via reverse transcription of an RNA intermediate, the pregenomic RNA (pgRNA) [2,3]. Virus assembly begins with the formation of an immature nucleocapsid (NC) incorporating the pgRNA and the viral reverse transcriptase (RT), which then undergoes a process of maturation defined as the conversion of the pgRNA first to a single-stranded (SS) DNA and subsequently to the RC DNA, catalyzed by the RT protein [4]. The RC DNA-containing NC is defined as the mature NC, which can be enveloped by the viral envelope proteins and secreted extracellularly as complete virion.
HBc is a small (183 or 185 amino acids depending on the strains, ca. 21 kd) protein that forms the shell of the NC and also plays a critical role at multiple other stages of HBV replication [2,5,6]. It is composed of three regions, an N-terminal domain (NTD), a C-terminal domain (CTD), and a linker that connects the NTD and CTD. NTD encompasses amino acid residues 1–140 and forms the classical assembly domain, generally thought to be necessary and sufficient for capsid assembly [7–9]. CTD encompasses residues from 150 to the C-terminal end, is highly basic (enriched in R, protamine-like), displays non-specific nucleic acid-binding activity [7,10], and is functionally important in pgRNA packaging and reverse transcription but generally thought to be dispensable for capsid assembly [11–14]. Furthermore, CTD is known to undergo dynamic phosphorylation and dephosphorylation, which regulate HBc functions in pgRNA packaging and reverse transcription [15–22].
In between the NTD and CTD is a “linker” peptide with a conserved sequence, 141STLPETTVV149 (Fig 1) [23]. The linker is routinely included together with NTD (as in HBc149; Fig 1) for recombinant expression and capsid assembly in bacterial systems and in vitro assembly reactions using HBc proteins purified from bacteria under high protein and/or salt concentration conditions [5]. Under these assembly conditions, the linker clearly does not interfere with NTD assembly. Indeed, deletion of most of the linker (from 143–149) in the context of the full-length HBc, resulting in the fusion of CTD directly to NTD, abolished capsid assembly when expressed in E. coli, suggesting a positive role for the linker in capsid assembly by the full-length HBc [24]. Furthermore, permutation of the last 7 residues of the linker in the context of HBc149 also prevented capsid assembly but replacement of these same seven residues by the seven N-terminal residues of HBc (MDIDPYK) maintained assembly [24]. These results thus further indicated that the specific sequence of the linker can modulate capsid assembly by the full-length HBc under those conditions. On the other hand, the linker can be removed entirely and NTD alone is able to assemble into capsids under those conditions, in the absence of both the CTD and the linker [8,24,25]. Interestingly, truncation of the linker, in the complete absence of the CTD, affected the ratio of the T = 3 (with 90 HBc dimers) or T = 4 (with 120 dimers) capsids assembled under these conditions [24,25]. Thus, whereas most capsids formed when the linker is present belong to the T = 4 class, most capsids formed when the linker is removed belong to the T = 3 class. Other than this apparent effect on the dimorphism of capsid assembly, the mechanism of which remains elusive, the linker is not known to have any other specific functions in HBV replication.
As a para-retrovirus, HBV is selective in virion morphogenesis in that only mature NCs containing the DS, RC DNA, but not immature NCs containing either pgRNA or the SS DNA, are selected for envelopment and secretion as complete virions [26,27]. Situated between the genome and the envelope, the capsid plays an integral role in this selective virion formation process. Within the NTD and spatially located on the surface of the capsid shell, a so-called matrix binding domain (MBD) has been defined, through elegant genetic analysis, that is thought to interact with a short segment in the preS1 region of the viral large envelope protein (L), the so-called matrix domain (MD), for complete virion formation [28–30]. L is one of three HBV surface or envelope proteins (HBs) (the other two being the middle or M and small or S surface protein), which are also secreted as the classical HBsAg particles (the Australian antigen) that contain no capsid or genome, in huge excess over complete virions (by up to 100,000-fold) [27]. Surprisingly, recent studies have revealed that HBV also secretes very high levels (ca. 1011/ml) of genome-free (empty) virions, which contain the envelope and capsid but no DNA or RNA and are found at ca. 100-fold excess over complete (i.e., RC DNA-containing) virions in cell culture supernatant and in the blood of experimentally infected chimpanzees and naturally infected humans [27,31,32]. Neither the capsid nor the envelope requirements for empty virion formation are clear at present. Naked (non-enveloped) capsids are also released in cell cultures via an unknown mechanism that appears to be different from that for the secretion of virions [33]. However, the release of naked capsids seems to be a phenomenon in transformed cell lines, and has not been observed in vivo during HBV infection [31,32,34].
We have recently demonstrated that contrary to expectation, the HBc CTD is apparently needed for capsid assembly in living human cells and in the rabbit reticulocyte lysate (RRL), where both the protein concentration and salt conditions mimic more closely the conditions in authentic human host cells than the previous assembly systems using bacterial expression and purified HBc proteins [35]. An HBc construct containing both the NTD and the linker (i.e., HBc149) but no CTD was unable to assemble under these (near) physiological conditions. On the other hand, other CTD-lacking HBc constructs that also lack part of the linker attached to the NTD (truncated at position 147, 145, or 144) accumulated and assembled to varying but detectable levels [11,36–38], indicating that the exact truncation point within the linker region affects the capacity of NTD to assemble in the absence of CTD. These results raise the possibility that the linker sequence can somehow interfere with the assembly by NTD in the absence of CTD under (near) physiological conditions in RRL and in human cells, and this inhibitory effect of the linker on the NTD assembly function is somehow overcome by CTD in the full-length HBc.
Given the unexpected role of CTD, and potentially of the linker, in capsid assembly in RRL and in human cells, it is now important to further assess the role of the CTD, the linker and the interplay between the CTD and the linker, in capsid assembly under physiological conditions. Also, these results bring about the possibility that the linker may have potential roles in the other functions of HBc beyond capsid assembly, which has never been tested so far. Therefore, we have carried out a genetic analysis to test the role of the linker in capsid assembly, both in the presence and absence of CTD, under near physiological conditions in vitro and in cells. Furthermore, the effects of a panel of linker deletion and substitution mutants on pgRNA packaging, DNA synthesis, and virion secretion were assessed. Our results have revealed that the linker indeed can affect capsid assembly in a manner that is dependent on CTD, and furthermore, it plays a critical role in multiple stages of HBV replication beyond capsid assembly.
Since previous reports have found that HBc mutants with truncation of the linker region, in addition to CTD removal (i.e., C-terminal truncation beyond 149), could be expressed and assemble at appreciable levels [11,36–38] whereas NTD plus an intact linker (also without CTD) (i.e., HBc149) failed to assemble and accumulate [35,38], we reasoned that deletion of the linker may restore HBc expression and/or assembly, when the CTD was absent. Thus, we deleted the entire linker (from 141 to 149) or only part of the linker (from 144–149), both in the absence of CTD, to make HBc140 and HBc143 (Fig 1), respectively, and determined their expression and assembly in human hepatoma cells (HepG2 and Huh7). A second plasmid expressing the HBV pgRNA and all viral proteins except HBc (HBV-C-) was co-transfected in a trans-complementation assay to assess the ability of the mutant HBc proteins to carry out the other functions of HBc including pgRNA packaging, DNA synthesis, and virion secretion. We selected the plasmid pSVHBV1.5 to derive the HBc-defective genomic construct, as our pilot experiments showed that this plasmid secreted significantly higher levels of HBsAg than another genomic construct pCMVHBV (S1 Fig). Since the secreted HBsAg is known to be in great excess over virions during natural HBV infection [27,31], the higher levels of HBsAg produced from pSVHBV1.5 helped to ensure that the complementation experiment mimicked better the natural infection in terms of HBsAg expression and to avoid the potential situation where the expression of the envelope proteins might become limiting for virion secretion. Even though the HBc sequence used here was from HBV genotype D, and the complementing (HBV-C-) construct was from HBV genotype A, they complemented each other efficiently in all aspects of viral replication assayed here, as shown below.
In support of a negative effect of the linker on NTD expression/assembly as hypothesized above, the expression levels of HBc140 and HBc143, as assessed by SDS-PAGE and western blot analysis, approached those of the WT HBc in both HepG2 and Huh7 cells (Fig 2A and 2C, 3rd and bottom panels), much better than that of HBc149, which retains the entire linker [35,38]. As shown in Figs 2 and S2, the mAb T2221, recognizing an epitope towards the end of the HBc NTD [39], detected the WT and CTD- (and linker-) deleted HBc proteins very well, in comparison with two other mAbs targeted to the beginning of NTD, 10E11 [40] (commercially available) and the anti-WHc made against the very N-terminal sequences of the woodchuck hepatitis virus (WHV) core protein (WHc), which are identical to those in HBc [32,41] (see Materials and Methods). The levels of intracellular capsids (Fig 2A and 2C, 2nd panels), and the naked capsids released into culture medium (Fig 2B, top right), as assessed by native agarose gel electrophoresis and western blot analysis, were also higher than those of HBc149 although still lower than those of the WT HBc. For HepG2 cells, the naked capsids released into the culture supernatant by HBc140 and HBc143 (and even HBc149) were relatively abundant (though still less than the WT HBc) (Fig 2B, top right) although the levels of intracellular capsids from these mutants were very low (Fig 2A, 2nd panel). Thus, the release of naked capsids into the culture supernatant might be enhanced by the linker (and CTD) deletions in HBc140 and HBc143. An enhanced release of capsids plus a partial defect in capsid assembly could explain the relative abundance of HBc140 and HBc143 proteins detected by SDS-PAGE western blot analysis (Fig 2A, 3rd panel) but very low levels of intracellular capsids (Fig 2A, 2nd panel). For Huh7 cells, a similar phenomenon could have occurred but the released naked capsids from HBc140 (and to a lesser degree, HBc143) could have been rapidly disrupted/degraded in the supernatant (see also Fig 3 below). This could explain the relative abundance of these mutant proteins detected by the SDS-PAGE western blot analysis (Fig 2C, 3rd panel) but very low levels of intracellular and extracellular capsids (esp. for HBc140) (Fig 2C, 2nd panel; Fig 2D, top right).
To assess the potential role of the linker in the context of the full-length HBc (i.e., with both NTD and CTD), we constructed HBc/Δ141–149 (with the entire linker deleted) and HBc/Δ145–149 (deleting the C-terminal portion of the linker), which share the similar linker deletions as HBc140 and HBc143 but retain the CTD (Fig 1). Both of these linker deletion constructs were expressed and assembled into capsids like the WT HBc in both HepG2 and Huh7 cells (Fig 3A and 3C, middle and bottom panels). These results were rather surprising in light of the previous report showing that deletion of the linker, thus fusing the CTD directly to NTD, abolished capsid assembly in the bacterial expression system, which was taken as evidence to indicate a need for a flexible linker between the NTD and CTD to prevent the CTD from interfering with NTD assembly [24]. In light of this surprising result, we constructed another partial linker deletion construct, in the context of the full-length HBc, by deleting HBc residues 141–144 (i.e., the N-terminal portion of the linker) to make HBc/Δ141–144 (Fig 1). In addition, we made the same partial N-terminal deletion of the linker, in the absence of CTD, to construct HBc149/Δ141–144 (Fig 1). HBc/Δ141–144 was expressed and assembled just like the WT HBc in both HepG2 and Huh7 cells (Fig 4A and 4C, lane 5, middle and bottom panels). On the other hand, the same partial linker deletion, in the absence of the CTD, in HBc149/Δ141–144 did not rescue NTD expression or assembly (Fig 4A and 4C, lane 6, middle and bottom panels), unlike the deletion of the entire linker (in HBc140) or its C-terminal portion (in HBc143) described above.
It was previously shown that the sequence of the linker between the NTD and CTD could affect capsid assembly in the bacterial expression system [24]. We thus tested two different linker substitution mutations that were shown to be either compatible or not with assembly (Fig 1). The substitution that disrupted assembly was a randomized WT linker sequence (STETVPVLT, dubbed LR for “linker random” here), whereas the substitution that retained assembly was the replacement of the last seven residues of the linker by the first seven residues from the N-terminal end of HBc with the first two residues unchanged (STMDIDPYK, dubbed LN for “linker N-terminal” here). Interestingly, we found both of these linker substitutions were similar to the WT HBc in expression and assembly in human hepatoma cells (Fig 4A and 4C, lanes 2–3, middle and bottom panels), in contrast to their severe defect in assembly in bacteria [24], further attesting to the drastic effects of the expression host on the assembly behavior of the different HBc constructs. We also made a third linker substitution with a nine-residue segment (TTLPETTII) from a cellular protein (dubbed LC for “linker cellular” here) that is very similar to the WT HBc in sequence (the middle six residues being the same as the WT linker and the other three residues representing conserved substitution: S141T, V148I, and V149I) and in predicted secondary structure [24]. This substitution was also compatible with capsid assembly in hepatoma cells (Fig 4A and 4C, lane 4, middle and bottom panels).
We next tested the potential effect of the linker deletions and substitutions on the HBc function in pgRNA packaging into NCs. Given the known critical role of CTD in mediating pgRNA packaging, it was no surprise that none of the CTD deletion mutants with or without linker deletions (HBc140, HBc143, HBc149, HBc149/Δ141–144) were able to support packaging of viral RNA (Fig 2A and 2C, lanes 2–4, top panels; Fig 4A and 4C, lane 6, top panels). On the other hand, it was interesting that some of the linker mutations, in the presence of an intact CTD, also impaired pgRNA packaging. The complete linker deletion, HBc/Δ141–149, showed a decrease in pgRNA packaging by ca. 5–10 fold after normalizing to the amount of capsids (Fig 3A and 3C, lane 2, top panels), whereas partial deletion of the C-terminal portion of the linker, HBc/Δ145–149, decreased pgRNA packaging less severely, by ca. 3–4 fold (Fig 3A and 3C, lane 3, top panels). Partial deletion of the N-terminal portion of the linker, HBc/Δ141–144 had the weakest effect, decreasing pgRNA packaging by ca. 2 fold (Fig 4A and 4C, lane 5, top panels). In contrast to the linker deletions, none of the linker substitutions affected pgRNA packaging (Fig 4A and 4C, lanes 2–4, top panels), indicating that the specific sequence of the linker was not critical for this HBc function.
As expected from the essential role of CTD in pgRNA packaging as well as in facilitating viral reverse transcription, none of the CTD deletion mutants (HBc140, HBc143, HBc149, HBc149/Δ141–144) showed any viral DNA in NCs (Figs 1, 2B and 2D, top left, lanes 2–4). Intriguingly, even in the presence of the CTD, the complete linker deletion (HBc/Δ141–149) and the C-terminal partial linker deletion (HBc/Δ145–149) showed no viral DNA in NCs (Fig 3B and 3D, lanes 5, 6, top panels), indicating a critical role of the linker, particularly its C-terminal portion (145–149), in viral reverse transcription beyond its role in facilitating pgRNA packaging described above. On the other hand, the N-terminal partial linker deletion, HBc/Δ141–144, contained some viral DNA in NCs, although at reduced levels compared to the WT HBc (Figs 1 and Fig 4B, lane 5, top panel). The three linker substitutions apparently contained viral DNA in their capsids at levels similar to the WT (Fig 4B, lanes 2–4, top panel; Fig 4D, lanes 3–5, top panel).
To assess the species of DNA synthesized in mutant capsids, we extracted viral DNA from the WT and mutant capsids and analyzed their DNA content by Southern blot analysis. Previous results from us and others indicated that certain capsid mutants allow viral DNA synthesis but are unable to protect their DNA content from exogenous nuclease digestion, which is routinely used to remove plasmid DNA during core DNA extraction [19,42,43]. To avoid this potential issue so as to obtain a more accurate assessment of viral DNA synthesized in the mutant capsids, we extracted capsid-associated DNA (or core DNA) without nuclease digestion but then degraded the contaminating plasmid DNA in the resulting core DNA preparation with DpnI, which digests plasmid DNA (methylated in bacteria) but not viral DNA synthesized in hepatoma cells [42]. All capsids that contained viral DNA (the three linker substitutions and the partial N-terminal linker deletion, HBc/Δ141–144) based on the particle gel analysis (Fig 4B and 4D) had SS DNA (i.e., minus strand), although the SS DNA levels were reduced in HBc/Δ141–144 by ca. 2-fold compared to the WT HBc (Fig 5). As the SS DNA is reverse transcribed from pgRNA, this modest reduction of SS DNA in HBc/Δ141–144 was at least partly due to the moderately reduced levels of pgRNA packaging in this mutant described above. In contrast, HBc/Δ141–149 showed no DNA and HBc/Δ145–149 showed barely detectable levels SS DNA (Fig 5), consistent with the particle gel results (Fig 3B and 3D). Again, this DNA synthesis defect could be partly the consequence of the defect in pgRNA packaging by these two mutants. These results thus indicated that the specific sequence of the linker was not critical for the first step of reverse transcription to generate the minus strand DNA, and a linker that was only five (instead of the nine in WT) residues long was sufficient for SS DNA synthesis. Intriguingly, the partial N-terminal linker deletion (HBc/Δ141–144), as well as two linker substitutions (LR and LN), showed no RC DNA in their capsids in contrast to the WT HBc (Fig 5, lanes 5–7). These three mutants did make immature DS DNA intermediates (running as a smear between the SS DNA and RC DNA in Fig 5), indicating they were able to initiate plus strand DNA synthesis and elongate the plus strand to a limited extent. However, only the conservative linker substitution (LC) was competent in RC DNA synthesis (Fig 5, lane 8), thus implicating a critical role of the linker, in a sequence-specific manner, in the second step of reverse transcription (extensive plus strand DNA synthesis to generate RC DNA).
We next assessed the capacity of the linker mutants to be enveloped and secreted into the culture supernatant as virions. Viral particles (including both virions and naked capsids) released into the culture supernatant of transfected HepG2 or Huh7 cells were analyzed by native agarose gel electrophoresis, whereby naked (non-enveloped) capsids released into the culture supernatant were well separated from virions (enveloped) as the former migrated much faster than the latter on the gel (Figs 2–4, panels B and D). Complete virions were detected by Southern blot analysis of HBV DNA. Empty virions were detected by western blot analysis of the HBc protein in virions, assuming that the vast majority of HBc signal (99% or more) from virions was from empty virions, as shown in previous studies [18,31,32]. As expected, HBV DNA in (complete) virions (or naked capsids), readily detectable in WT virions, was not detected from HBc140, HBc143 or HBc149 in HepG2 cells (Fig 2B, top left) (true also for Huh7 cells; see below Fig 2D, top left), due to their lack of CTD, which is known to be essential for pgRNA packaging or DNA synthesis. On the other hand, the HBc protein signal detected in the WT virions (i.e., empty virions) was also undetectable from these mutants when tested in HepG2 cells (Fig 2B, top right). These results thus indicated that the linker, and/or CTD (see below also), was important for secretion of empty virions.
This suggestion was then confirmed by results obtained using Huh7 cells, when decreasing amounts of culture supernatant from the WT HBc transfection were analyzed along with that from the HBc140 and HBc143 transfection (Fig 2D). When the amount of supernatant from the WT HBc transfection was decreased by 10-fold, the levels of naked capsids released into the medium were similar to those from the HBc143 transfection (Fig 2D, top right, lanes 3 and 5); virion capsids were clearly detectable from the WT HBc even with this reduced loading whereas no virion capsids from either HBc143 or HBc140 were detected (Fig 2D, top right, lanes 3–5). As expected, the HBs signals were only detected with virions but not naked capsids (Fig 2B and 2D, bottom). As HBsAg particles (with no capsids or genome) are not separated from virions (either empty or complete) on the agarose gels under these conditions [31,32], the abundant HBsAg signals, in the absence of HBc signals at the top of the gel in the case of HBc140, HBc143 and HBc149 represented just HBsAg particles (no virions) (Fig 2B, lanes 2–4, bottom; Fig 2D, lanes 4, 5, bottom), as verified by the detection of HBsAg at the same position on the gel in the complete absence of HBc expression (Fig 2D, lane 6, bottom). These results thus indicated that capsids formed by NTD, in the absence of CTD and the linker, could not be enveloped for secretion as empty virions.
It was noticeable that the complete linker deletion (HBc/Δ141–149) showed little to no naked capsids in the culture supernatant either (Fig 3B and 3D, lanes 2 and 5), suggesting that the complete linker deletion might also have blocked the release of naked capsids into the culture medium, or if released, was rapidly degraded in the supernatant. On the other hand, we detected a smeary HBc signal migrating just below the virions and much slower than naked capsids, detectable only from this mutant (in Huh7 but not HepG2 cells), in a manner that was independent of the viral envelope proteins (Fig 3B and 3D, lanes 2 and 5, bottom). This result suggested that some naked mutant capsid might be disrupted once released extracellularly under certain conditions. The exact nature of the slowly-migrating HBc smear (in a non-capsid form) from this mutant, and its apparent cell line dependence, remained unclear. If the HBc/Δ141–149 capsid was indeed blocked from release from the cell, the excess mutant capsid might be degraded intracellularly such that its level in the cell did not exceed that of the WT HBc (Fig 3A and 3C).
The role of the linker in virion secretion, both complete and empty, was assessed in the context of HBc linker mutants which retain an intact CTD. Both the complete linker deletion (HBc/Δ141–149) and the two partial linker deletions (HBc/Δ141–144 and HBc/Δ145–149), despite being competent for capsid assembly intracellularly, did not show any virion secretion (Fig 3B and 3D, lanes 5, 6; Fig 4B, lane 5; Fig 4D, lane 2). As two of these three mutants (HBc/Δ141–149 and HBc/Δ145–149) failed to synthesize any viral DNA and the third linker deletion mutant (HBc/Δ141–144) failed to make RC DNA (which is a prerequisite for complete virion secretion) (Figs 3–5), the specific effect of these mutants on secretion of complete virions could not be ascertained from these experiments. However, these results clearly indicated that both parts of the linker were required for secretion of empty virions. The critical role of the linker in virion secretion was further confirmed with the linker substitution mutants. All three linker substitution mutants were defective in secreting empty virions (Fig 4B, lanes 2–4, bottom; Fig 4D, lanes 3–5, bottom), although the conservative substitution (LC) showed a low level of empty virions (ca. 10% of WT) (Fig 4B, lane 4, bottom; Fig 4D, lane 5, bottom). Again, since the LR and LN substitution mutants failed to make RC DNA (Fig 5), the specific effects of these mutations on DNA virion secretion could not be determined from these experiments. Interestingly, the conservative substitution (LC) allowed secretion of complete virions (virion DNA), despite severely blocking the secretion of empty virions (virion HBc) (Fig 4B, lane 4; Fig 4D, lane 5).
The results presented above indicated that the linker was required for virion secretion (Figs 3B and 3D and 4B and 4D), but a role for CTD could not be excluded, since when the CTD alone was deleted and the linker was retained (as in HBc149), there was little to no accumulation of intracellular capsids (Fig 2A and 2C, 2nd panel), precluding an assessment of its virion secretion capacity in the absence of the CTD. To overcome this limitation, we appended four positive R residues (4R) to HBc149, reasoning that the supply of the positive charges might rescue assembly of HBc149, in the absence of CTD, by either interacting with non-specific RNA or with NTD of HBc [35,44]. Indeed, HBc149-4R, in contrast to HBc149, accumulated substantial, though still lower than WT, levels of intracellular capsids that were released in the culture medium (Fig 6B and 6D). As expected, the HBc149-4R mutant capsids failed to package pgRNA or synthesize viral DNA due to the lack of a complete CTD (Fig 6A and 6C). Since the capsid levels of HBc149-4R were still lower than those of the WT HBc, we titrated the amount of WT HBc and HBc149-4R plasmids used for transfection, relative to the HBc-defective genomic construct, and measured levels of capsids and virions across the titration to facilitate a direct comparison of virion secretion efficiency of HBc149-4R relative to the WT HBc. Importantly, HBc149-4R, was secreted as virions (empty) as efficiently as the WT, when normalized to the capsid levels (Fig 6B and 6D). Thus, the linker was able to support efficient secretion of (empty) virions in the absence of a complete CTD.
Since the state of CTD phosphorylation is known to play a critical role in capsid assembly, pgRNA packaging, and reverse transcription, which were affected by the linker mutants studied here, we decided to test if the various linker mutants could affect the CTD phosphorylation state. Since HBc assembles into capsid particles rapidly in hepatoma cells, which can affect CTD phosphorylation state indirectly by influencing the accessibility of the CTD phosphorylation sites to host kinases and phosphatases, and CTD also undergoes dynamic phosphorylation and dephosphorylation associated with pgRNA packaging and reverse transcription, we decided to use the RRL in vitro translation system for HBc expression and assembly that we developed recently [35]. In this cell-free system, HBc is phosphorylated during translation by endogenous cellular kinases, at (at least) some of the same CTD sites as in vivo, which is independent of capsid assembly, pgRNA packaging or DNA synthesis [35], and HBc assembly does not occur until triggered by exogenous phosphatase treatment. We therefore examined the CTD state of phosphorylation of the WT HBc and various linker mutants immediately after translation, before triggering capsid assembly, to determine HBc phosphorylation state in the absence of capsid assembly.
Following resolution of HBc by SDS-PAGE, we used an NTD-specific mAb (T2221) to measure the total HBc levels, irrespective of CTD state phosphorylation (Fig 7, top panel) and two CTD-specific mAbs, B701 that is selective for the phosphorylated CTD with an epitope between 155–164 (Fig 7, middle panel), and 25–7 that is selective for the non-phosphorylated CTD with an epitope between 164–182 (Fig 7, bottom panel) [18,35], for western blot analysis. The specificity of the mAbs was verified by using the non-phosphorylated HBc protein purified from E. coli (Fig 7, lane 1). The complete linker deletion mutant (HBc/Δ141–149), as well as the two partial deletion mutants (HBc/Δ141–144 and HBc/Δ145–149), showed strongly increased (by ca. 5- to 7- fold) B701 signal relative to the WT HBc after normalization of the total HBc signal (as detected by mAb T2221) (Fig 7, lanes 4–6), indicative of enhanced CTD phosphorylation at the B701 epitope. The linker substitution mutant, HBc-LN, also showed a similar effect on CTD phosphorylation to the linker deletion mutants, albeit to a lesser degree (by ca. 3-fold) (Fig 7, lane 8). On the other hand, the 25–7 signal for the complete linker deletion (HBc/Δ141–149), the C-terminal partial linker deletion (HBc/Δ145–149), and the LN substitution mutants was modestly (by ca. 2-fold) increased relative to the WT HBc (Fig 7, lanes 4, 5, 8), suggesting the 25–7 epitope was less phosphorylated in these mutants as compared to the WT HBc.
We have demonstrated here that mutations of the HBc linker affected multiple steps in HBV replication, including modulation of capsid assembly, pgRNA packaging, DNA synthesis, and virion secretion, implicating a critical role for the linker in multiple stages of HBV replication (Fig 1). The mechanisms of action for these linker functions remain to be elucidated. As the nine-residue long linker peptide is not known to have any enzymatic function or biochemical activity (such as nucleic acid binding), we consider it plausible that the effects of the linker on the HBc functions in capsid assembly, pgRNA packaging, and reverse transcription, are exerted through its effects on the NTD or CTD (Fig 8). This is supported by our findings that the linker affected NTD assembly and CTD state of phosphorylation. On the other hand, the linker may function in a more direct manner (independent of its effects on the NTD or CTD) to facilitate virion secretion by interacting with the viral envelope proteins (Fig 8).
As introduced earlier, in the absence of CTD, the linker is not required for capsid assembly in bacteria or under in vitro assembly conditions with high HBc and/or salt concentration [24,25]. However, deletion of the linker, thus fusing the CTD to NTD, or substitution of the linker sequence, interfered with NTD assembly in bacteria (Fig 1) [24]. In sharp contrast, we have shown here that in human hepatoma cells, the linker interfered with NTD assembly if the CTD was absent, but in the presence of the CTD, linker deletions or substitutions did not interfere with capsid assembly. Consistent with the inhibitory effect of the linker on assembly by NTD reported here, a recent study also found that HBc149 failed to accumulate in a mouse hepatocyte cell line but HBc144 did (similar to HBc143 here) [38]. It thus appears that the NTD alone is sufficient, at least to a limited extent, for capsid assembly in human cells, but the presence of the linker, in the absence of the CTD, interferes with NTD assembly specifically in human cells but not in bacteria. Furthermore, we have shown here that in the absence of CTD, deletion of the linker sequence 141–144 (HBc149/Δ141–144) was less effective in restoring capsid assembly, compared to deletion of the entire linker (in HBc140) or 144–149 (in HBc143). This suggests that N- and C-terminal sequences of the linker are not equivalent in modulating capsid assembly and the C-terminal part of the linker (144–149) may have a more detrimental effect on NTD assembly than the N-terminal part of the linker (141–144) when the CTD is absent.
How the linker may influence capsid assembly, in a host cell- and CTD-dependent manner, is one of the intriguing questions brought up by our studies here that warrants further studies. The linker may interfere with NTD assembly in human cells, in the absence of CTD, by affecting the conformation of NTD, or by interacting with a host factor(s) to inhibit assembly (Fig 8). When expressed in bacteria, the high protein concentration achieved may somehow overcome the inhibitory effects of the linker on NTD assembly, or host cell-specific factors may alleviate the linker effect. As HBc149-4R, in contrast to HBc149, assembled efficiently in human cells, a role for electrostatic interactions between the highly basic CTD and a negatively charged ligand (e.g., RNA, or acidic residues in the HBc NTD) can be implicated in alleviating the inhibitory effect of the linker on NTD assembly in human cells by the CTD. Moreover, capsid stability, instead of or in addition to assembly, could be affected by the linker, as suggested by the apparent disruption of the HBc/Δ141–149 (with the complete linker deletion, fusing the CTD directly to the NTD) capsid once it was released extracellularly.
The linker, and its specific sequences, are important for capsid assembly in bacteria when both the NTD and CTD are present (i.e., in the context of the full-length HBc) [24], but not in human cells as we have shown here. Other than the differences in HBc subunit concentration and salt/pH conditions, phosphorylation of the HBc CTD, which occurs in human cells but not in bacteria and is furthermore modulated by the linker as we have shown here, is known to modulate capsid assembly [35]. This host cell-dependent and linker-modulated CTD phosphorylation (Fig 8) may be part of the reason why deleting the linker or substituting its sequences interferes with capsid assembly in bacteria but not in human cells. In addition, CTD is known to interact with host factors in mammalian cells, such as I2PP2A and B23 [45], and SRPK [46], which may also contribute to the host cell-dependent effects of the linker mutations. Indeed, we have shown previously that the binding of I2PP2A and B23 to the CTD is modulated by the linker in the case of the duck hepatitis B virus core protein [45], which is thought to be much longer than the HBc linker and located between position 186–230 [47].
It remains possible that deletion of the CTD impaired the production and/or stability of the mutant protein in human cells (but not in bacteria), accounting for the very low expression level of HBc149 in hepatoma cells. However, we believe that the lower expression level of this mutant was mostly due to its defect in efficient assembly in mammalian cells (and consequently, more rapid degradation). First, we have shown recently that this same mutant is expressed at levels equal to or higher than the WT HBc in a mammalian cell extract, the rabbit reticulocyte lysate in vitro translation system; yet, it still fails to assemble, unlike the WT HBc in the same system that assembles efficiently [35]. Second, HBc149 expression and assembly in human cells can both be rescued by co-expression of the WT HBc [35].
Deletion of the entire linker severely impaired pgRNA packaging, and partial deletion of sequences from 145–149 had a more deleterious effect than that of 141–144, suggesting that the linker sequences from 145–149 had a more important role than 141–144 in pgRNA packaging in the presence of the CTD, similar to the non-equivalent role of the two parts of the linker on capsid assembly in the absence of the CTD. On the other hand, none of the linker residues individually was absolutely required for pgRNA packaging as they could be substituted without affecting pgRNA packaging. One potential mechanism for the linker to modulate pgRNA packaging may be via its influence on CTD phosphorylation (Fig 8), which we could demonstrate here. As we proposed recently [35], hyper- or hypo-phosphorylation of HBc CTD can both impair specific pgRNA packaging, by decreasing overall RNA (including the specific pgRNA) binding affinity or failing to block non-specific RNA binding, respectively. Details of the effects of the linker on CTD phosphorylation, in a phosphorylation site- and maturation stage-specific manner will require comprehensive studies in the future. How the linker may affect CTD phosphorylation state also remains to be elucidated. One possibility is that the linker modulates CTD conformation, which in turn affects the accessibility of the CTD phosphorylation sites to host kinases and/or phosphatases. Alternatively, the linker may affect the recruitment of these CTD-modifying host enzymes, either directly by serving as binding sites for these factors, or through an indirect means (Fig 8). Additional effects of the linker, beyond affecting CTD phosphorylation, including its influence on NTD assembly, may also play a role in modulating pgRNA packaging.
Deletion of the entire linker, or partial deletion of the linker sequences from 145–149 abolished viral reverse transcription, whereas deletion of 141–144 had only a modest effect. This result again suggests that the linker sequences from 145–149 had a more important role than 141–144 in reverse transcription, as in pgRNA packaging. However, since the linker is nine-residues long, it was impossible to construct a deletion mutant removing precisely half of the linker (i.e., 4.5 residues). So, it remains possible that HBc/Δ141–144 was more effective than HBc/Δ145–149 in making immature DNA (and packaging pgRNA) simply because it is one residue longer than HBc/Δ145–149. On the other hand, as with pgRNA packaging, it is clear that none of the linker residues individually was required for SS DNA synthesis as they could be substituted with little effect on SS DNA levels. Furthermore, our results here have shown that a linker that is five (instead of nine as in the case of the WT)-residues long is still capable of supporting pgRNA packaging and SS DNA synthesis, at least partially. This is consistent with the observation that the linker is disordered in recombinant capsids assembled in bacteria from HBc149 (i.e., missing the entire CTD) [9] and the notion that the linker may form a flexible, mobile array on the inner surface [24] of the maturing NC to facilitate this stage of viral DNA synthesis. On the other hand, RC DNA synthesis was clearly impaired by two of the three linker substitutions as well as the partial deletion from 141–144, which had little effect on SS DNA synthesis. The remaining linker substitution that was competent for RC DNA synthesis is a very conservative one with almost identical sequence and predicted structure to the WT linker. Thus, for RC DNA production, the linker did not merely function as a flexible spacer but played a specific role. How the linker might facilitate RC DNA synthesis in a sequence-dependent manner is not yet known. As the CTD state of phosphorylation is known to be important for RC DNA synthesis [17,19,20], and the linker sequences could affect CTD phosphorylation, the specific linker sequences could modulate RC DNA synthesis through their effect on CTD phosphorylation (Fig 8), as proposed above for their effects on pgRNA packaging. In addition, the linker may be involved in the conformational changes of the maturing NC that accompany, and may be required for, RC DNA synthesis [48]. In addition, the linker itself may undergo conformational changes, in a sequence-dependent manner, during the viral replication cycle that are modulated by the NTD or CTD. We note also that although no exogenous nuclease digestion was used during viral DNA extraction, our results here can’t exclude the possibility that in those mutants where no RC DNA was detectable, some RC DNA might actually have been made but degraded as soon as it was made in the cell.
Perhaps the most intriguing result we have obtained here regarding the linker functions is its essential role in the secretion of empty virions. Those linker deletion and substitution mutants that impaired RC DNA synthesis were also defective in the secretion of complete (RC DNA-containing) virions. The conservative linker substitution (LC) that remained competent for RC DNA synthesis was also capable of secreting complete virions. This is expected as RC DNA synthesis is required for complete virion formation. A specific effect of these mutants on the secretion of complete virions, however, could not be ascertained from these results (Fig 8). On the other hand, it is clear from our results here that the specific linker sequence is critical for empty virion secretion. All linker mutations, either complete or partial deletions or substitutions, impaired secretion of empty virions. Even the conservative linker substitution (LC), which was fully competent in all other aspects of the viral life cycle tested here including the secretion of complete virions, showed a severe defect (though not as severe as the other linker substitution or deletion mutants) in the secretion of empty virions. The LC linker substitution increased the ratio of complete to empty virions, by ca. 10-fold from ca. 1% to 10% [32], by decreasing the secretion of empty virions without affecting that of complete virions. Thus, we have not only uncovered an essential role of the linker in the secretion of empty virions, but also revealed that the requirements for the secretion of complete vs. empty virions can be separated genetically.
The efficient secretion of HBc149-4R capsids in empty virions further suggests that the linker is not only necessary but may be sufficient to support empty virion formation, although it remains formally possible that both the linker and several R residues from the CTD are required for the secretion of empty virions. As the beginning of the HBc CTD has the sequence 150RRRGR154…, it may be argued that HBc149-4R actually retains a severely truncated “CTD,” i.e., the first three (or five without G153) residues of the CTD. Thus, further studies will be needed to clarify the contribution of the CTD, if any, in the secretion of empty HBV virions. It was recently reported that HBc147 capsids (missing the entire CTD and two C-terminal residues of the linker) failed to be secreted in empty virions [36]. Whereas the authors hypothesized that their result implicated a critical role for the CTD in empty virion secretion, our findings here suggest an alternative interpretation of the same result, i.e., the last two residues of the linker (148 and 149) plays a critical role in supporting empty virions secretion. The CTD state of phosphorylation, which appeared to be affected by the linker mutations, is unlikely to account for the effect of the linker mutations on virion secretion as CTD phosphorylation state, per se, does not play a critical role in virion formation [18]. These results, combined, suggest the intriguing possibility that linker residues interact, directly, with the envelope proteins during virion formation (Fig 8). Whereas the linker is generally thought to be located inside the capsid [24,49] and thus unlikely to interact with the envelope proteins on the capsid surface, it may nevertheless be exposed, at least transiently, on the capsid surface. Some evidence in support of an exterior localization of the linker has indeed been presented; for example, epitopes attached to the linker are accessible to antibody binding in empty HBV capsids [50,51]. Additionally or alternatively, the linker sequence may be involved, perhaps via interactions with host factors, in trafficking of the capsids to the site of budding for their envelopment.
Future studies, including high-resolution structural analysis, will be required to further elucidate the mechanisms of action of the linker functions uncovered here and to determine any structural changes in the capsid, at the various stages of viral replication, that may be modulated by the linker. For example, whereas the linker is known to affect the dichotomy of T = 4 or T = 3 capsids in bacteria, whether this is also the case in human cells remains to be determined. Furthermore, it remains unknown if both size classes of capsids are competent in pgRNA packaging or reverse transcription. On the other hand, both T = 3 and T = 4 capsids are found in extracellular virions [52], indicating that they are both competent for virion formation and so the capsid size is unlikely to be a determinant of virion formation. As uncovered here, the critical roles that the HBc linker plays at multiple stages of HBV replication, which have been thought to involve only the HBc NTD and/or CTD, emphasize the close and dynamic interactions among all three regions of HBc that together carry out the multiple essential functions of HBc in viral replication. As conformational changes are likely to be associated with NC maturation and envelopment [6,30,32,48,53], further structural studies of the HBc linker mutants that affect various stages of viral replication should provide important insights into the effects of the linker on the conformations of the HBc NTD, CTD, and the NC as a whole and how the conformational effects translate to functional effects on NC maturation and envelopment.
The multiple roles of the HBc linker in HBV replication that we uncovered here provide an explanation for the high degree of sequence conservation in this region of HBc. In addition, as the same DNA sequence coding for the HBc linker also codes for the very N-terminal part (residues 5–14) of the viral RT protein, the need to preserve polymerase sequence and functions possibly also has contributed to the conservation of the DNA sequence in this region of the HBV genome. However, as we highlighted recently [54], the N-terminal sequences of the polymerase are actually not highly conserved and mutagenesis work so far indicates that this region of the polymerase is not essential for any known functions of the polymerase although it does contribute, to some degree, to the polymerase functions in pgRNA packaging and protein-primed initiation of reverse transcription. Thus, it is likely that the preservation of the HBc linker sequence and functions has played a more important role in the DNA sequence conservation of this region of the viral genome. On the other hand, some variations of the linker sequence have been observed [23]. In light of our findings here, future studies to examine the functional effects of the naturally occurring linker variations are warranted.
HBc has emerged recently as the primary target, after the HBV RT protein, for developing effective antiviral strategies to clear HBV infection. Almost all agents in development so far are targeted to the NTD [5,55,56]. Our results here indicate that sequences outside the NTD, including the CTD as well as the linker, could represent important targets for HBc-directed antiviral development. In fact, a small molecule compound has been reported recently that inhibits HBc assembly and functions in a manner that is dependent on sequences in the CTD [57]. Similarly, it may be possible to identify compounds that target the conserved HBc linker region to inhibit multiple steps of HBV replication. Our discovery here of the multiple critical functions of the HBc linker in HBV replication also has broad implications. Thus, linkers connecting protein domains are common occurrences including those in other viral capsid proteins [58]. For the human immunodeficiency virus type 1 (HIV-1), the linker in its capsid protein has been shown to regulate capsid stability and reverse transcription [59].
pCI-HBc and -HBc149 expressing the full-length and CTD-deleted HBc have been described before [35]. pCI-HBc149-4R is identical to pCI-HBc149, except four R residues are added after HBc position 149 (Fig 1). pCI-HBc140, -HBc143, -HBc149/Δ141–144, -HBc/Δ141–149, -HBc/Δ141–144, -HBc/Δ145–149 were derived from pCI-HBc through PCR-mediated mutagenesis for the expression of CTD and/or linker deletion mutants (Fig 1). Three linker substitution mutants of HBc were also constructed via PCR mutagenesis. The C-terminal seven residues of the linker were randomized in sequence in the mutant LR, or replaced with the seven N-terminal residues of HBc in LN as described before [24]. In the third substitution mutant, LC, the entire linker was replaced with a nine-residue segment from a cellular protein (cellobiose dehydrogenase) similar in sequence and predicted structure to the linker [24] (Fig 1). pSV-HBV1.5/C- expresses a HBc-defective HBV genome [30], which is capable of supporting viral replication upon complementation with HBc. pCMV-HBV expresses the HBV pgRNA from the heterologous cytomegalovirus (CMV) immediate early promoter and the HBV surface mRNAs from the endogenous HBV promoter, leading the production of all viral RNAs and proteins required for replication and virion secretion [60,61].
A mouse monoclonal antibody (mAb), clone T2221, against the HBc NTD [39] was purchased from Tokyo Future Style (Cat no. 2AHC24). The mAb 10E11 against HBc NTD (residues 2–10) [40] was purchased from Abcam (Cat no. ab8639). The mAb, anti-WHc, specific for the WHc NTD (likely the first 8 residues), is cross-reactive with HBc due to the identity of the very N-terminal HBc and WHc sequences, as reported before [32,41]. The HBc CTD-specific mAbs, 25–7 and B701, have been described recently [18,35]. The rabbit polyclonal antibody against HBc were purchased from Dako. The rabbit anti-HBs polyclonal antibody was purchased from Virostat [18]. The anti-preS2 mAb (Arigo Biolaboratories) detect the preS2 region that is shared by both the L and M (but absent from the S) HBV envelope proteins.
HBc expression constructs and/or HBV genomic constructs were transfected into the human hepatoma cell line HepG2 or Huh7 cells (kindly provided by Christoph Seeger, Fox Chase Cancer Center) as previously described [42,62,63]. Briefly, HepG2 cells in 60-mm dishes were transfected with 4 μg of plasmid using FuGENE6 (Roche). Huh7 cells seeded in 60-mm dishes were transfected with 10 μg of plasmid using CalPhos Mammalian Transfection Kit (Clontech). Cells and culture supernatant were harvested on day 7 post-transfection. Cells were lysed with NP40 and HBc proteins in the cytoplasmic lysate were resolved by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE), transferred to polyvinylidene difluoride (PVDF) membrane, and detected by the indicated antibodies as described previously [19,32]. Core DNA from NCs was isolated from the cytoplasmic lysate without nuclease digestion and analyzed by Southern blot analysis as described previously [42]. A genome-length, 32P-labeled HBV DNA probe was used to detect the viral DNA replicative intermediates by Southern blot analysis.
Native agarose gel electrophoresis of intact NCs from the cytoplasmic lysate, or extracellular viral particles obtained after DNase I digestion of polyethylene glycol (PEG) precipitated cell culture supernatant [32] were carried out by using previously reported procedures [18,19,31,32]. Briefly, following transfer to nitrocellulose membrane, viral DNA associated with the particles was detected using a full-length HBV DNA probe, or pgRNA packaged into NCs by using a minus-sense riboprobe. The same membrane was subsequently probed with the indicated HBc or surface specific antibody to detect HBc or surface proteins. The signals from the 32P-labeled RNA probe were quantified using a phosphor imaging system (GE Healthcare). The chemiluminescent signals representing the capsid protein were quantified using the ChemiDoc MP system and BioLab software, as previously described [64]. Densitometry using appropriately exposed films was also used in some cases to quantify the RNA and protein signals. All quantifications were repeated with at least three separate transfection experiments.
A TnT-coupled rabbit reticulocyte lysate (RRL) in vitro translation system (Promega) was used to express the WT HBc or linker deletion/substitution mutants, as described previously [35]. In vitro-translated proteins were analyzed by SDS-PAGE and western blot using the indicated anti-HBc antibodies.
|
10.1371/journal.pgen.1004281 | Genome Sequencing and Comparative Genomics of the Broad Host-Range Pathogen Rhizoctonia solani AG8 | Rhizoctonia solani is a soil-borne basidiomycete fungus with a necrotrophic lifestyle which is classified into fourteen reproductively incompatible anastomosis groups (AGs). One of these, AG8, is a devastating pathogen causing bare patch of cereals, brassicas and legumes. R. solani is a multinucleate heterokaryon containing significant heterozygosity within a single cell. This complexity posed significant challenges for the assembly of its genome. We present a high quality genome assembly of R. solani AG8 and a manually curated set of 13,964 genes supported by RNA-seq. The AG8 genome assembly used novel methods to produce a haploid representation of its heterokaryotic state. The whole-genomes of AG8, the rice pathogen AG1-IA and the potato pathogen AG3 were observed to be syntenic and co-linear. Genes and functions putatively relevant to pathogenicity were highlighted by comparing AG8 to known pathogenicity genes, orthology databases spanning 197 phytopathogenic taxa and AG1-IA. We also observed SNP-level “hypermutation” of CpG dinucleotides to TpG between AG8 nuclei, with similarities to repeat-induced point mutation (RIP). Interestingly, gene-coding regions were widely affected along with repetitive DNA, which has not been previously observed for RIP in mononuclear fungi of the Pezizomycotina. The rate of heterozygous SNP mutations within this single isolate of AG8 was observed to be higher than SNP mutation rates observed across populations of most fungal species compared. Comparative analyses were combined to predict biological processes relevant to AG8 and 308 proteins with effector-like characteristics, forming a valuable resource for further study of this pathosystem. Predicted effector-like proteins had elevated levels of non-synonymous point mutations relative to synonymous mutations (dN/dS), suggesting that they may be under diversifying selection pressures. In addition, the distant relationship to sequenced necrotrophs of the Ascomycota suggests the R. solani genome sequence may prove to be a useful resource in future comparative analysis of plant pathogens.
| The fungus Rhizoctonia solani is divided into several sub-species which cause disease in a range of plant species that includes most major agriculture, forestry and bioenergy species. This study focuses on sub-species AG8 which causes disease of cereals, canola and legumes, and compares its genome to other R. solani sub-species and a wide range of fungal and non-fungal species. R. solani is unusual in that it can possess more than one nucleus per cell. The multiple nuclei and sequence mutations between them made assembly of its genome challenging, and required novel techniques. We observed signs that DNA sequences originating from multiple nuclei in AG8 exhibit a high frequency of single nucleotide polymorphisms (SNPs) and more SNP diversity than most fungal populations. These SNP mutations also have similarities to repeat-induced point mutations (RIP). Moreover in AG8, RIP-like SNPs are not restricted to intergenic regions but are also widely observed in gene-coding regions. This is novel as RIP has previously only been reported in repetitive DNA of distantly-related fungi that have only a single nucleus per cell. We generated a list of 308 genes with similar properties to known plant-disease proteins, in which we found higher rates of non-synonymous mutations than normal.
| Rhizoctonia solani (formerly, teleomorph: Thanetophorus cucumeris) is a globally-distributed, soil-borne fungal phytopathogen employing a necrotrophic lifestyle. Collectively, the host-range of the R. solani species spans numerous plant species vital to the agriculture, forestry and bioenergy industries, including but not limited to: wheat, rice, barley, canola, soybean, corn, potato and sugar beet [1]. Chemical control methods may not be feasible nor economical for the control of many soil-borne pathogens [2]. Hence, agronomic controls such as crop-rotation are heavily relied upon to fight this disease, though the polyphagous habit of some isolates can include commonly rotated crop species. For example, cereal and legume rotations are susceptible to AG8 [1], [3]; and corn, canola and soybean rotations are susceptible to AG1 and AG2 [4]–[5]. Susceptible crop species possess at best, low to moderate levels of genetic resistance which are of limited use to conventional breeding strategies [6]–[8]. The impact of R. solani has been observed to increase in incidence and severity with increased adoption of conservation (no-till) farming techniques [2].The combinations of these factors places R. solani as a significant threat to global food security and other agro-forestry industries.
The R. solani species complex is comprised of fourteen anastomosis groups (AGs), most of which are reproductively incompatible with each other and are numbered AG-1 through AG-13. The ‘bridging isolate’ AG-BI is the exception, being compatible with multiple AGs [1], [9]. Despite an apparently low level of phylogenetic divergence between AGs [10] they exhibit diverse phenotypic variation, particularly with respect to the host-ranges of phytopathogenic AGs (Supporting Table S1A). Less frequently, certain AGs have been observed to have a predominantly saprophytic or mycorrhizal life-cycle.
Our study presents a comprehensive genome assembly and functional analysis of R. solani AG8, causative agent of bare patch of wheat, barley and legume species [3], [11]–[12]. Of the AGs that infect wheat, AG8 is the most damaging. In Australia, the impact of R. solani on wheat and barley production is estimated upwards of $77 million per annum and bare patch also remains a major problem for the production of wheat and other crops in the US [13]. The host-range of the sequenced isolate WAC10335 (zymogram group ZG1-1 [14]) also extends to legume species of agricultural and scientific importance: Lupinus spp. (lupin) [15] and Medicago truncatula (barrel medic) [16], but not to the non-legume Arabidopsis [17]. As a basidiomycete, the plant pathogens most closely related to R. solani with genome sequences available are the biotrophic smuts [18]–[20], rusts [21]–[22] and the tree-pathogenic Moniliophthora spp. [23], which possess vastly different lifestyles. Thus, the information gained from R. solani is expected to be of importance in filling gaps in our knowledge of plant pathogen biology, which apart from rusts and smuts, is skewed towards the ascomycetes.
Significant genomic resources for other AGs of R. solani have also recently become publicly available, formerly being limited to EST libraries of AG1-IA [24] and AG4 [25]. The recent generation of whole genome sequences of R. solani AGs presents new opportunities for comparative genomics between R. solani anastomosis groups. The most comprehensive whole-genome study to date has been that of the rice pathogen AG1-IA [26] [GenBank: AFRT00000000]. The genome assembly of the closely related AG1-IB was published recently [27] [GenBank: CAOJ00000000], however full scaffold sequences were not in the public domain at the time of writing and thus AG1-IB data has not been used for synteny comparisons in this study. The mitochondrial genome sequence of the potato pathogen AG3 strain Rhs1AP and its comparison to that of AG1-IB has been published recently [28]. A draft nuclear genome for AG3 is also available (http://www.rsolani.org with kind permission from Cubeta et al.), however a nuclear gene dataset and genome survey have not yet been published [29].
R. solani AG8 exists as a multi-nuclear heterokaryon in which individual R. solani cells may carry multiple nuclei and copy number can vary between cells. An average of 8 nuclei per cell has previously been observed in AG8, but numbers commonly ranged from 6 to 15 [1]. While reduction of nuclear complexity via protoplast isolation has been reported for R. solani [30]–[32], we chose to assemble a representative haploid assembly of all AG8 nuclei in an agriculturally-relevant isolate and investigate mechanisms and type of sequence variations between nuclei in this largely asexual isolate. We report evidence of SNP-level diversity between heterokaryotic nuclei of a complex fungal genome, which has not previously featured extensively in genome studies of fungal phytopathogens. The heterozygosity between nuclei of AG8 compounded the complexity of its de novo genome assembly [available from GenBank: AVOZ00000000] and we also describe novel bioinformatic approaches used to overcome these challenges. This study also compares whole-genome synteny between R. solani anastomosis groups (AG8, AG1-1A and AG3) and uses comparative genomics techniques to highlight genes and functions unique to AG8 and AG1-1A. Predicted properties of AG8 proteins have been leveraged to generate a list of 308 ‘effector-like’ genes that may be related to plant-pathogenicity. These collective resources will be important for further experimentation in this pathosystem.
The heterokaryotic nature of the R. solani genome posed considerable challenges for genome assembly. To overcome these challenges we developed a novel genome assembly pipeline (Figure 1). The assembly process, including software and parameters, is described in the Materials and Methods section with additional information in Supporting Text S1. Preliminary de novo assemblies exhibited high levels of sequence redundancy and heterozygosity across gene-encoding regions. We confirmed that multiple nuclei were present in variable numbers within cells of the sequenced isolate (Figure 2A). In order to reduce sequence redundancies caused by the assembly of heterozygous homeologs, the process used to assemble the AG8 genome included a step to merge haplotype contigs prior to scaffolding. This step was followed by generation of a haploid ‘majority consensus’ sequence from alignments of genomic sequence reads to merged scaffolds. However prior to this study, the extent of sequence variation between homeologous chromosomes originating from different nuclei was unknown. Alignment of genomic deep-sequencing reads to the genome assembly indicated an abundance of heterozygous SNP mutations throughout the AG8 assembly (Figure 2B). As many as 74% of heterozygous SNP alleles were transition mutations between cytosine and thymine (or their complementary bases guanine and adenine) (Figure 2C, Supporting Table S2A). Nucleotides flanking these C→T ‘hypermutations’ exhibited a moderate bias of approximately 40% for a G at the 3′ base (i.e. CpG→TpG) (Supporting Table S2B). These cytosine and CpG hypermutations were widespread across the AG8 genome and occurred within protein-coding genes and repetitive DNA regions at similar levels (Figure 2D), with only a slight reduction in CpG frequency in genes relative to repeats. One of the consequences of C→T mutation is the introduction of stop codons into protein-coding open-reading frames (ORFs) [33]. We reason that it is possible for ORFs to be inactivated by nonsense mutations in the majority of nuclei, yet still produce functionally active, full length proteins from a low number of non-mutated nuclei in R. solani AG8. Thus the assembly process also included a step which reverted heterozygous mutations between C and T to cytosine, regardless of allele frequencies. The final R. solani AG8 draft assembly comprises 861 scaffolds, has a total length of 39.8 Mbp which is consistent with previous haploid cytogenetic estimates of 37 to 46 Mbp [34], an N50 of 65 and an N50 length of 160.5 kbp (Table 1).
The AG8 genome assembly statistics compared favorably with those of other R. solani isolates AG1-1A, AG1-1B and AG3 as shown in Table 1. Sequence comparisons between the whole genome assemblies of R. solani AG8, AG1-1A and AG3 exhibited widespread co-linearity or macrosynteny [35] (Figure 3, Supporting Table S3). No conclusive evidence for dispensable chromosomes, as reported for F. oxysporum [36], was observed.
A single scaffold (Scaffold_77) of ∼140 kbp in length was predicted to represent the mitochondrial genome. The ends of the mitochondrial scaffold sequence were confirmed to be physically joined in a circular configuration by PCR (Supporting Text S2). The mitochondrial scaffold contained the expected set of fungal mitochondrial genes (atp6, cytb, cox1-3, nad1-5 & nad4L, rps5, rns & rnl) and was abundant with LAGLIDADG and GIY-YIG intronic endonucleases. This is consistent with recent reports for the mitochondrial genomes of AG3 and AG1-IB, which are of similarly large sizes (235.8 kbp and 162.8 kbp respectively) and possess high abundances of endonucleases [28].
Within the nuclear genome, repetitive DNA sequences (Supporting Table S4A) represented just over 10% of its total length. Gypsy retrotransposons were the most abundant repeat type and represented 4% of the nuclear genome. Protein-coding gene-based tri-nucleotide simple sequence repeats, WD40-like and tetratrichopeptide repeats, represented approximately 1%. Comparing the repetitive content of AG8 with available repeat data for AG1-1A, we observed more repetitive DNA in the assembly of AG8 (10.03% of the assembly) compared to that of AG1-1A (5.27%) [26]. It should be noted that critical differences in assembly, de novo repeat prediction and repeat classification methods may limit the comparability of these two datasets, however the proportions of the most dominant repetitive elements was strikingly similar. The most dominant transposable elements in both AG8 and AG1-1A were LTR retrotransposons: the most common being the Gypsy/Dirs1 family at 3.98% and 3.43% respectively; followed by the Ty1/Copia family at 0.14% and 0.60% respectively. This pattern of Gypsy being more numerous than Copia retroelements, appears to be typical of most fungal genomes [37]. Non-coding RNA (ncRNA) genes were predicted in silico (Supporting Table S5A), which overall made up less than 0.007% (26.5 kbp) of the total genome length.
To enable discovery and accurate annotation of protein-coding genes present in the AG8 assembly, particularly those expressed in the presence of plant tissues, three high-coverage Illumina RNA-seq libraries were aligned to the genome to delineate gene exon boundaries. To obtain transcript data for as many genes as possible, the libraries included one library of AG8 undergoing vegetative growth in culture and two “infection-mimicking” libraries. These libraries were derived from AG8 grown on water agar containing wheat (Triticum aestivum) or Medicago truncatula seedlings separated by a permeable nitrocellulose membrane. This enabled collection of fungal tissue whilst reducing plant tissue contamination to negligible amounts. Alignment of RNA-seq data and proteins from related fungal species and pathogenicity gene databases were combined with in silico gene predictions to automatically predict gene structure annotations, which were then manually curated.
The density of gene-coding regions was relatively even throughout the assembled genomic scaffolds (Figure 2Eii), with reduced density at some scaffold termini with high levels of repeats (Figure 2Eiii). A total of 13,964 protein-coding AG8 genes that can serve as a reference for R. solani comparative genomics were predicted after RNA-seq-assisted manual gene annotation. Of these, 8,449 proteins had a BLASTP match to the NCBI NR protein database (Supporting Figure S1, Supporting Table S6). The taxonomic distribution of lowest-common ancestor taxa for these BLASTP matches indicated wide conservation of 83% (7016/8449) of R. solani AG8 with fungal proteins, 52.5% (4436/8449) specifically conserved within the Basidiomycota (Supporting Figure S1) and 17.9% conserved within the class Agaricomycetes. The extracellular secreted component of these proteins was predicted using a combination of SignalP [38], WolfPsort [39] and Phobius [40] (Figure 4). A total of 1,959 proteins (14.0% of all proteins) were predicted to be secreted by one or more methods and 608 (4.4%) were predicted to be secreted by all three methods. For comparative purposes, SignalP predictions were applied to R. solani AG8 and across 86 fungal species (Supporting Table S7). There were 911 secreted proteins predicted by SignalP for AG8, which was similar to the numbers predicted for closely-related plant-pathogenic species of the class Agaricomycetes. The secretome counts across biotrophic Basidiomycetes of other classes were relatively variable, e.g. Puccinia striformis (1,264), P. graminis f. sp. tritici (2,012) and Ustilago maydis (595). However AG8 was within a similar range to the average predicted secretome count across all fungi (1,052), which was predominantly comprised of necrotrophs.
To surmise the biological processes important to R. solani AG8 in the infection process, we predicted the functions of its 13,964 genes by comparison to the CAZy (Carbohydrate-Active enZyme) and Pfam (Protein family) databases. In total, we assigned CAZy annotations to 1,137 genes (Supporting Table S8B,C) and Pfam annotations to 6,099 genes (44.5%) (Supporting Table S9A).
Analysis of CAZymes present in the R. solani AG8 genome (Figure 5) revealed a dual bias for the degradation of the structures of plant cells and modification of the fungal cell wall for growth or protection from host-defences (Supporting Table S8C). The most abundant CAZy families are described here. The most prevalent glycoside hydrolase (GH) CAZyme class (GH18) represented chitinases, followed in frequency by classes representing cellulases (GH5), polygalacturonases (GH28) and beta-glucanases (GH16), which degrade major components of plant cell walls. The most abundant glycosyltransferase (GT) classes were strongly geared towards cellulose (GT2, GT41), hemicellulose (GT77, GT4, GT34) and chitin (GT2) degradation. The most common carbohydrate esterase (CE) class contained choline esterases (CE10). Polysaccharide lyase (PL) CAZymes were strongly biased towards pectin-degradation, with the two most dominant classes (PL1 and PL3) both representing pectate lyases. The three most abundant carbohydrate-binding (CBM) class CAZymes were lectin-like proteins. Two of these (CBM13 and CBM57) are predicted to bind cellulose and hemicelluloses and include ricinB-like lectins. The third (CBM18) contains sialic-acid-binding lectins, which may play a role in protection from plant host-defenses by ‘shielding’ sugars protruding from the fungal cell wall [41]. The fourth most frequent CBM class (CBM1) binds chitin and cellulose and appears to be conserved exclusively within fungal species.
Pfam domains in R. solani AG8 were compared to Pfam annotations assigned to a panel of 50 pathogenic and non-pathogenic fungal species (obtained from the JGI Integrated Microbial Genomes database) (Supporting Figure S2, Supporting Table S9A) [42]. R. solani AG8 exhibited high abundance of tyrosine protein kinase signalling, membrane transport, protein-protein binding, reduction-oxidation, DNA methylation and a bias among cell-wall degrading enzymes towards pectin and peptidase degradation. Pfam domains with protein-protein binding functions were dominated by various classes of tetratrichopeptide repeats, but also included other domains involved in protein binding interactions: (WD40-like) PD40 beta-propeller [Pfam: PF07676]; Ankyrin [Pfam: PF13606] and leucine-rich repeats [Pfam: PF00560]. The most abundant peptidase domain was the CHAT (Caspsase HetF-Associated with TPRs) domain [Pfam: PF12770] which may be involved in programmed cell death. In summary, R. solani AG8 possesses a number of gene families whose members have a broad range of potential biological roles, for example those encoding caspases or protein-binding functions. Further study would be required to determine their relevance to plant pathogenicity or other lifestyle characteristics. These findings do however indicate that R. solani AG8 possesses a large number of carbohydrate-binding lectins of unknown function as well as a battery of CAZymes suitable for consumption of carbohydrates commonly found in cereal hosts, but also is geared towards the degradation of pectin.
Publicly-available protein data for AG1-IA [26] was also used to generate functional annotations for AG1-IA. Statistical comparisons between functions predicted in AG8 and AG1-IA were performed using Fisher's exact test (p≤0.05) (Supporting Table S10A). R. solani AG8 and AG1-IA primarily infect two different hosts - wheat and rice respectively. Differences between them in their relative abundances of functionally-annotated genes may reveal important differences in their infection strategies. Overall, fewer Pfam domains were found to be significantly higher in AG1-IA than in AG8. In AG1-IA (Supporting Table S10B), the Pfams that were significantly more abundant and may be related to pathogenicity included several types of transmembrane transporter domain and formin-like proteins that may be involved in cytokinesis. Many more functions were found to be increased in AG8 relative to AG1-IA (Supporting Table S10C), however most of these were of too broad or unknown function to infer their biological roles. Nevertheless, several functions stood out as potentially important for plant pathogenicity in AG8, including CAZymes, peptidases, membrane transporters, transcription factors and toxin-like proteins. Peptidases abundant in AG8 included the CHAT and C14 domain caspases as well as fungalysin-like peptidases. The CAZyme functions that were significantly more numerous in AG8 were predominantly glycosyl-hydrolases (polygalacturonases, β-galactosidases), pectate lyases and carbohydrate binding proteins (ricin-like and jacalin lectins and fungal-specific CBM1 proteins). Fungal pathogens of dicots generally possess higher numbers of pectin-degrading enzymes than monocot pathogens [43]. Though an important pathogen of monocot cereals, most notably wheat, the sequenced isolate of R. solani AG8 was isolated from the dicot lupin and is also an important pathogen of other leguminous dicots. The abundance of pectate lyases in AG8 relative to AG1-IA is likely to reflect the broad host range of the sequenced AG8 isolate.
Interestingly, AG8 had more members of two Pfams similar to ricinB lectins [44] and delta endotoxins [45], highly toxic proteins commonly associated with defence against insect predators which have been prioritised for further study. In contrast to AG1-IA which had none, AG8 possessed 3 delta-endotoxin-like proteins (RSAG8_06697, RSAG8_07821 and RSAG8_07820) with the Pfam domain Bac_thur_toxin [Pfam: PF01338]. This domain was originally defined based on the insecticidal delta endotoxins of Bacillus thuringiensis. Pfam matches and orthology analysis suggested the presence of orthologous delta endotoxin-like proteins in other phytopathogenic species including Fusarium graminearum (Fusarium head blight of wheat and barley) and the bacteria Dickeya dadantii (syn. Erwinia chrysanthemi, soft-rot, wilt and blight on a range of plant hosts and septicaemia of pea aphid) [46] (Supporting Table S9A, Supporting Table S11). Whether these ricinB and delta-endotoxin homologs confer an advantage against competitors or predators or may instead be toxic to the plant host remains to be determined.
Effector proteins have been observed to be secreted by several microbial pathogens [47] and cause disease on their respective hosts. A set of characteristics common to plant pathogenicity effectors from fungi that would allow reliable bioinformatic predictions has not yet been accurately defined. However experimentally validated effectors tend to be low molecular weight, secreted, cysteine-rich proteins which may contain certain conserved amino-acid motifs near the N-terminus [47]–[48] (Supporting Table S12). Effector-like proteins were predicted in AG8, requiring: complete annotation from translation start to stop with <3 consecutive unknown (‘X’) amino acids; predicted molecular weight ≤30 kDa; predicted as secreted with 0–1 predicted transmembrane domains; and with ≥4 cysteine residues. A total of 308 AG8 proteins matched all of these criteria. These candidates were searched for known motifs previously associated with plant pathogenicity, however the occurrence of these motif matches was not significant relative to the complete protein dataset. As an example the RxLR-like motif (Kale et al., 2011), though found in 73% of the predicted effector candidates, was also found in 77% of the whole R. solani AG8 proteome (Supporting Table S13) indicating this permissive motif may not be useful for effector candidate prediction in R. solani AG8. We were also unable to identify any novel N-terminal-associated motifs that were highly conserved among these 308 proteins (Supporting Text S3). However, we observed the ratio of non-synonymous to synonymous mutations (dN/dS) within these 308 candidate genes to be 0.97 compared to 0.77 across all genes. Our understanding of the roles of these 308 effector candidates will benefit from the addition of further experimental data, resulting in a more succinct list of candidates with a potential direct role in disease on one or more of the many plant hosts of R. solani AG8. Unfortunately, no method for the stable transformation of R. solani AG8 is presently available and thus functional testing of candidate pathogenicity genes will be challenging.
To gain further support for an association with pathogenicity, approximately 10% (29) of the 308 predicted ‘effector-like’ genes were randomly selected and their mRNA expression relative to a set of 7 constitutively expressed genes was compared between R. solani AG8 sampled at 7 days post-infection of wheat and 7 day-old AG8 mycelia grown on media. Of these 29 genes, 25 (85%) had a positive fold-change and 17 had a significantly higher relative expression in-planta (Student's t-test; p≤0.05, log2 fold change ≥1) (Supporting Table S14B). This dataset highlights several plant-pathogenicity candidates, but other genes also important for pathogenicity may not be changing in abundance during infection relative to in-vitro growth.
Repeat-induced point mutations (RIP) are fungal-specific SNP mutations previously reported to be restricted to the filamentous Ascomycota (Pezizomycotina) [49]. RIP in the Pezizomycotina involves transition mutations converting cytosine to thymine, with a moderate bias for CpA dinucleotides [49]. Other features of RIP include targeted mutation of repetitive DNA, with single-copy DNA regions being largely unaffected. An important exception to this is where RIP mutations ‘leak’ into single-copy DNA regions from neighbouring repetitive DNA which occurs more frequently closer to repeats [50].
The small number of studies looking for RIP-like mutations in the Basidiomycota do not exhibit the characteristic CpA mutation bias observed in the Pezizomycotina [49], however two studies have reported a CpG dinucleotide bias between repetitive DNA sequences within the Basidiomycota and a TpCpG trinucleotide bias specific to the subphylum Pucciniomycotina [51]–[52]. As an Agaricomycete, we expect R. solani to exhibit a bias towards CpG but not TpCpG. However, it should also be noted that hypermutations of CpG may also be caused by widely conserved processes involving the methylation of cytosine to 5-methylcytosine (5mC) and subsequent deamination which converts 5mC to thymine [53]. Importantly, conversion of cytosine to thyimine via methylation and deamination does not actively target repetitive DNA or ‘leak’ in the same manner as RIP.
Analysis of nucleotides immediately flanking heterozygous C↔T SNP sites in AG8 exhibited a CpG dinucleotide bias consistent with previous observations of ‘RIP-like’ cytosine hypermutations in the Basidiomycota [52] (Figure 2D). The distribution of these RIP-like mutations in AG8 was observed to occur across repetitive and gene-encoding regions alike at a relatively constant ratio versus non-RIP-like mutations, where heterozygous C↔T alleles comprised ∼70–80% of all SNP mutations (Figure 2Eiv-v) and in turn CpG dinucleotides comprised ∼40–50% of heterozygous C↔T alleles. In mononuclear fungal genomes, RIP has previously only been observed to act upon repetitive DNA or to ‘leak’ into adjacent non-repetitive sequences [50]. Due to the novel genome assembly process for AG8 which involved merging of redundant haplotypes, a survey of SNP mutations in its annotated repetitive DNA would likely lead to incorrectly inflated counts of RIP-like mutations. Therefore we instead looked at the frequency of CpG↔TpG mutations versus their distance from the nearest repeat, which indicated that CpG mutations were more frequent closer to repeats (Figure 6). Furthermore, although the ratios of (C↔T/all SNPs) and (CpG↔TpG/C↔T) were relatively similar between genes and other regions of the genome, the frequency of mutations in gene regions were lower than in the genome as a whole, suggesting strong selection pressures to retain protein function. The ratio of CpG/CpH (where H = not G) was slightly lower in repeats (0.3) than in genes (0.4) (Table 2) and we speculate that this likely to be due to complete (i.e. homozygous) conversion of C→T occasionally occuring across all copies of a repeat, as they are under no selection pressure to retain their pre-RIP sequences. Thus there would be fewer sites that can be detected as heterozygous SNPs by aligning genomic reads to the genome assembly.
Regardless of whether the underlying process is similar to RIP or not, CpG-biased hypermutation is likely to play an important role in the evolution of the AG8 genome. RIP has been recently proposed to have the potential to randomly introduce nonsense mutations, converting longer secreted proteins into small, secreted proteins thus making them gradually more effector-like [54]. Stop-codon frequency across the 12,771 annotated AG8 genes possessing stop codons is highest for TGA (40%) compared to TAA (31%) and TAG (29%). As TGA stop codons would be the primary nonsense product of CpG-biased hypermutation, similar evolutionary processes may also occur in AG8. Furthermore, the presence of multiple nuclei in AG8 could potentially compensate for loss of gene function due to hypermutation in one or more nuclei, allowing for a higher tolerance for the accumulation of mutations in gene-coding regions. Analysis of total SNP, and CpN dinucleotide frequencies (expressed in Table 2 as average distance in bp between mutations), showed that a SNP mutation occurred on average every 70 bp, cytosine hypermutations occurred every 89 bp and that there was a 40% bias towards CpG mutations occurring every 307 bp. Within the 308 predicted ‘effector-like’ genes, SNP mutations occurred on average every 55 bp, cytosine hypermutations every 81 bp and CpG mutations occurred every 265 bp. However, the ratios of (C↔T/all SNPs) and (CpG↔TpG/C↔T) were not significantly different between the complete set of 13,964 AG8 genes and the 308 effector-like genes. Interestingly, despite apparently similar mutation ratios, the ratio of non-synonymous to synonymous SNP mutations (dN/dS) was 0.97 in ‘effector-like’ candidates compared to 0.77 across all genes. This may suggest that the increased mutation rate conferred by CpG-biased hypermutation is advantageous for accelerating the adaptation of pathogenicity genes which, if being actively counter-acted by plant defences, are likely to be under diversifying selection.
The density of heterozygous SNP mutations within AG8 was compared to SNP densities between the genome assemblies of AG8 and AG1-IA, AG1-IB and AG3 (Table 3). SNP density in AG8 was highest within intronic regions (19.6 SNPs/kbp), moderate in coding exons and genes (14.5–15.9 SNPs/kbp) and lowest in intergenic regions (11.5 SNPs/kbp). Comparisons of SNP mutations between AG8 and alternate AGs exhibited an approximately ten-fold increase in SNP density compared to the rate of heterozygous SNPs within AG8. The corresponding values within for comparisons between AG8 and AG1-IA ranged from 162.8–228.2 SNPs/kbp, AG1-IA from 141.6–200.3 SNPs/kbp and AG3 from 98.5–145.3 SNPs/kbp. We note however that in these comparisons between the genome assemblies of AG8 and other AGs, it was not possible to ascertain whether these SNPs (or homologous bases) were homozygous or heterozygous in the alternate AG. Nevertheless a higher SNP density between the AG8 genome and those of the other three AGs, relative to heterozygous AG8 SNPs, was consistent in all three comparisons.
Comparisons between individual genomes and fungal population genetics studies were also used to place the SNP diversity within R. solani AG8 into a wider context. Similar to AG8, the Basidiomycete stripe rust fungus Puccinia striformis is heterokaryotic but exhibits a lower SNP density within its genome assembly of 5.98 SNPs/kbp [21]. It may be significant that P. striformis is binucleate and therefore only possesses 2 nuclei per cell as opposed to the 6–15 nuclei that have been observed within cells of R. solani AG8 [1]. Similarly, SNP variation across a population of shiitake mushroom (Lentinula edodes) was reported to be 4.6 SNPs/kbp (186,0789 SNPs in 40.2 Mbp) [55]. In barley powdery mildew (Blumeria graminis), the SNP rate observed between pairs of isolates was lower at 1 SNP/kbp [56]. Across isolates of the multinucleate endomycorrhizal Glomeromycete Rhizophagus irregularis [57] and the beetle-symbiont Leptographium longiclavatum [58], even lower SNP densities of 0.2 SNPs/kbp (28,872 SNPs in 140.9 Mb) and 0.6 SNPs/kbp (17,266 in 28.9 Mbp) respectively, were observed. In contrast, a population study of the multinucleate human pathogens Coccidioides immitis and C. posadasii reported a rate of 23.7 SNPs/kbp relative to the C. immitus RS reference genome assembly (687,250 SNPs in 28.95 Mb) [59], which though slightly higher is within a similar range to R. solani AG8 (Table 3). In conclusion, the SNP diversity in R. solani AG8 appears to be higher than that observed thus far within individual isolates of binucleate rusts, between isolates of the same pathogenic species and across non-pathogen populations. Furthermore, diversity within R. solani AG8 is comparable to a population of another multinucleate pathogen (C. immitus) and much higher than that observed within a population of a multinucleate non-pathogen (R. irregularis). We speculate that the combination of multinuclearity and selection pressures relating to pathogenicity may be driving the accumulation of widespread heterozygous SNP diversity in R. solani AG8.
In this study, we present a novel bioinformatics pipeline for the accurate and comprehensive assembly of a complex fungal genome, the heterozygous and multinucleate pathogen Rhizoctonia solani AG8 (Figure 1). The combination of genome and transcriptome sequencing allowed for data-driven gene prediction and comparative genomics with other publically available genomes of alternate anastomosis groups and other fungal species. Using a combination of novel genome assembly methods, RNA-seq, manual gene curation and comparative genomic techniques, a list of 308 ‘effector-like’ plant-pathogenicity gene candidates has been predicted. Analysis of mRNA expression for a subset of candidate pathogenicity genes during infection of wheat has highlighted several candidates for further study. Additionally, comparisons to available data for alternate AGs of R. solani have highlighted important differences, which may be related to differing host ranges, host tissue preference and environmental stress tolerance. The resources presented here should provide powerful tools for the identification of host-specialised mechanisms for fungal-plant interactions and pathogenicity for this important group of fungal pathogens.
CpG-biased hypermutations were observed between nuclei of AG8, within genes and repeat sequences alike and have some similarities with repeat-induced point mutation (RIP). Previous observations of RIP in haploid fungal genomes have only reported its activity upon repetitive sequences [49], [52] or non-repetitive regions within a finite distance of a repeat [50]. Although we observed hypermutation within genes, intriguingly these mutations were more numerous with increasing proximity to repeats, suggesting that repeats are mutated more frequently than genes and that a process similar to ‘RIP-leakage’ may occur. Furthermore, the molecular mechanisms of RIP have not yet been fully characterised [60] and the consequences of combining RIP-like hypermutation and multinuclearity in R. solani are unknown. In the basidiomycete human pathogen Cryptococcus neoformans, increases in ploidy and the accumulation of mutations have been implicated as mechanisms for its adaptation to immune- and drug-related selection pressures [61]. Also of note is that across isolates of the human-pathogenic and multinucleate ascomycete Coccidiodies immitus, higher relative frequencies of repeat-associated CpG mutation have also been observed [59] (unusual for species of the Pezizomycotina which typically exhibit a bias towards mutation of CpA [49]). We speculate that RIP-like SNP mutations accumulating in multiple nuclei may similarly be a means by which R. solani is also able to rapidly generate allelic diversity despite being predominantly clonally propagated [1]. Loss-of-heterozygosity and copy-number variation analyses to confirm this hypothesis would require further study using a sequencing platform which can produce longer read lengths and higher base-call accuracies than those used in this study. However, if this is the case, this mechanism may be a factor contributing to the relatively mild effectiveness of fungicide treatment against this pathogen [62] and its adaptation to a broad range of plant hosts.
R. solani AG8 isolate WAC10335 was isolated from lupin and provided by the Department of Agriculture and Food of Western Australia (DAFWA). Anastomosis group was confirmed by ribosomal ITS sequences and host-range was confirmed by inoculation assays on wheat, lupin and Medicago truncatula [3]. R. solani does not readily produce sexual or asexual spores thus single spore isolation was not possible, therefore a single rapidly growing hyphal tip was excised from a colony growing on PDA and transferred to water agar containing 250 µg/ml cefotaxime. Pathogenicity of the resulting culture was confirmed as equivalent to the original. A pure in vitro culture of R. solani was produced by incubation in PDB at 25°C with gentle shaking for 7 days. Hyphae were filtered from the culture through sterile Miracloth and rinsed with sterile water. DNA was purified by grinding hyphal tissue in liquid nitrogen and suspension in DNA extraction buffer (2% (w/v) CTAB, 1.4 M NaCl, 0.2% (v/v) β-mercaptoethanol, 20 mM EDTA and 100 mM Tris-HCl) and mixing at 60°C. Following two rounds of chloroform/isoamylalcohol extraction, the aqueous supernatant was treated with RNase I (Invitrogen) at 20 µg/ml. The DNA was purified through an additional two rounds of chloroform/isoamylalcohol extraction and precipitated by adding 0.1 volumes of 3M NaOAc (pH 5.2) and 0.6 volumes isopropanol. The resulting DNA pellet was resuspended in 10 mM Tris-HCl (pH 8.0) buffer and quantitated by Qubit (Invitrogen) and BioAnalyser prior to sequencing.
Two Illumina paired-end libraries of genomic DNA were sequenced, with 75 and 100 bp read lengths and median insert lengths of 250 bp and 300 bp respectively. Three Illumina genomic mate-pair libraries with insert lengths of 2 kbp, 5 kbp and 10 kbp were also obtained. Paired-end libraries were combined and trimmed for sequencing adapter/primer sequences, low-quality (<Q30), and low-complexity sequences via CutAdapt v1.1 [63] filtered for adapter sequences from the Truseq RNA and DNA sample preparation kits versions 1 and 2. Pairs with one or more reads ≤50 bp after trimming were discarded. Where possible, overlapping 3′ ends between pairs were merged into long singleton reads via FLASH v 1.2.2 [64]. FLASH was also applied to the mate-paired libraries, to remove paired-end contamination of incorrect insert length and pair orientation which would complicate genome assembly (Supporting Text S1).
For the purpose of gene annotation, Illumina paired-end libraries of 100 bp read lengths were obtained from 3 mRNA libraries derived from AG8 grown under: vegetative conditions (7 days at 25°C in PDB with gentle shaking) (non-infection) and; Medicago or wheat infection-mimicking conditions. Under infection-mimicking conditions, AG8 was cultured on a film of nitrocellulose overlaid on water agar containing young sterilised Medicago truncatula or wheat seedlings. After seven days incubation at 25°C the film and hyphae were removed, ensuring negligible plant contamination in subsequent RNA extractions with TRIzol (Sigma-Aldrich, St. Louis, MO). Two sequencing libraries were generated per mRNA library, with 200 bp and 500 bp insert sizes. Transcript libraries were trimmed for contaminant sequences via Cutadapt v1.1 as per genomic reads.
Complex genome structure caused by heterozygosity and multinuclearity prevented the use of commonly employed de novo assembly methods. To this end, a novel pipeline was developed for AG8 (Supporting Text S1). Paired-end libraries were assembled with SOAPdenovo v1.0.5 (k-mer length = 61) [65]. This assembly was scaffolded with SSPACE 2.0 using the parameters (end extension, min size 500 bp) [66] and subject to 5 rounds of Gapcloser2 [65] using paired-end and 3′ end merged single-end reads. Mate-paired reads were used for scaffolding but excluded from gap-closing to avoid introducing inversion errors (Supporting Text S1). Haplotype redundancy was reduced using HaploMerger v20111230 [67] (batchD: filterAli = 0; minlength = 10 bp; maxInternal = 10000000; mincoverage = 0). Tandem duplication assembly errors (common to polyploid assemblies) were corrected by a twofold approach (Supporting Text S1). The first method involved intra-scaffold re-assembly between rounds of scaffolding and gap-closing, where gaps were broken and tested for overlap via CAP3 [68]. The second method involved self-alignment via BLASTN [69], applied after scaffolding, gap-closing and N-breakage rounds had completed. Alignments occurring in tandem on the same sequence were identified, and the sequences of the second repeat plus the intermediate region were removed from the assembly if repeats were ≤500 bp apart or ≥30% polyN in intermediate region. Introduction of errors by these processes was corrected by re-alignment of raw genome reads with bowtie2 v 2.0.5 [70] followed by local-realignment at indels, variant-calling and variant-consensus generation via GATK v1.6.11 [71]. Variant Call Format (VCF v4.0) tables of SNP and indel variation between the paired-end, 3′-end merged (long single-end reads), 2 kbp mate-paired, 5 kbp mate-paired and 10 kbp mate-paired sequence libraries relative to the genome assembly sequences, were merged with VCFtools v0.1.6 [72] where variants agreed between at least 2 out of the 5 libraries. The most frequent alleles in the merged VCF were incorporated into the consensus sequence of the final assembly, with the exception of sites where cytosine (C) to thymine (T) (reverse complement: G to A) polymorphisms were observed at which the assembly was reverted to the C (or G) allele regardless of allelic frequency. The genomic distribution of SNP mutations was calculated using BEDTools v0.1.7 [73].
Genome assembly sequences of AG8, AG1-IA and AG3 were compared using MUMmer 3.0 [74] using both nucmer and promer (parameters: –maxmatch). Summary statistics were derived from coordinate outputs.
Repetitive sequences were predicted via RepeatScout v1.0.5 [75], requiring consensus sequences ≥50 bp and ≥5 copies. Full-length repeats were reconstructed from RepeatScout outputs with CAP3 (v10.15.7, -h100 -p80 -z1) [68], manually curated and mapped to the genome assembly via RepeatMasker v3.2.9 (parameters: -e crossmatch -s) [76]. Repeat types were characterised using a combination of BLASTn vs NCBI Nucleotide, BLASTx vs NCBI Protein [69], CENSOR vs REPBASE v17.11 [77] and TEClass [78]. Repeat regions were also predicted with TransposonPSI v08222010 [79] and RepeatMasker vs REPBASE v17.11 (species = “Eukaryota”) [80]. All repeat data, excluding repeats corresponding to protein-coding genes (Supporting Table S4B), were used as negative support for gene annotation.
Exon splice-junctions were determined by aligning six RNA-seq libraries to the AG8 assembly via TopHat 2.0.4 (minimum intron size 20 bp, maximum intron 5000 bp, no coverage search, 2 splice mismatches, microexon search enabled, very sensitive, 20 read mismatches, 3 segment mismatches, max insertion length 12 bp, max deletion length 12 bp, report discordant and secondary alignments) [81]. Transcriptome de novo assembly was performed via Trinity r2012-03-17 (k-mer trimming with JellyFish 1.1.4, jaccard clipping, minimum contig length 150 bp, min k-mer coverage 5×, minimum glue 5, minimum percent read mapping 70%) [82]–[83]. Combined and individual library-specific assemblies were used for manual gene annotation.
The library-specific and the combined transcriptome assemblies were used to determine exon structure using PASA r2012-06-25 (minimum percent aligned 75%, maximum intron length 5000 bp). The output of PASA was passed to the EvidenceModeller r2012-06-25 auto-annotation pipeline [84], which also incorporated the following supporting data: splice-junctions determined from RNA-seq alignment to the genome assembly via Tophat 2.0.5; in silico gene prediction via GeneMark-ES v3.2.3 [85]; ESTs and proteins of previously sequenced fungi and PHIbase v3.4 [86] aligned to the genome assembly with AAT r03052011 [87]; predicted repetitive DNA (see above) and; non-coding locus predictions via tRNAscan-SE v1.23 (genomic, COVE only) [88] and Infernal v1.0 [89]. Gene annotations were evaluated by EvidenceModeller, visualised in Apollo v1.11.6 [90] and manually curated. Predicted protein translations were compared to the NCBI NR Protein database by BLASTP [69] (BLAST v2.2.26, e-value≤1e−3, top 20 hits) and the taxonomic distribution of their corresponding lowest-common ancestor taxa was summarised with MEGAN5 (LCA parameters: minimum support 1, minimum score 40, max expected 1e−3, top percentage 100) [91].
Conserved protein domains in AG8 and AG1-IA were predicted with HMMER 3.0 [92] versus Pfam(A) v26.0 [88] with gathering cutoffs. Carbohydrate-active enzymes (CAZymes) were predicted with CAT v1.8 [93]. Multiple alignments of the most abundant CAZyme families were generated with MAFFT L-INS-i v7.130 [94] (Supporting Table S15). Comparison of Pfams were performed between R. solani AG8 and other sequenced fungi from JGI IMG v4 [42]. Orthology comparisons between AG8 predicted proteins and protein datasets from 197 fungal, oomycete, prokaryotic, insect and nematode species and included a range of pathogens with different host ranges and non-pathogens (Supporting Table S11) was performed via ProteinOrtho v4.26 (BLASTP v2.2.26, E-value = 1e-05, alg.-conn. = 0.1, coverage = 0.5, percent_identity = 25, adaptive_similarity = 0.95, inc_pairs = 1, inc_singles = 1, selfblast = 1, unambiguous = 0) [95]. Predicted secretome comparisons were performed using SignalP 4.1 [38] between R. solani AG8 and 86 other fungi (Supporting Table S7).
Candidate ‘effector-like’ pathogenicity genes were classified by: complete annotation with translation start and stop codons and ≤3 consecutive unknown ‘X’ amino acids; predicted to be secreted by at least one method; 0–1 predicted transmembrane domains (single domains can be mis-predicted within secretion signal peptides); predicted molecular weight ≤30 kDa; and ≥4 cysteine amino acids. Molecular weights and amino-acid compositions were predicted with Bio::Tools::SeqStats (BioPerl) [96]. Sub-cellular localisation, secretion status and transmembrane domains were predicted with Phobius v1.01 [40], SignalP v4.1 [38] and WolfPSort v0.2 [39]. Matches to motifs previously associated with plant pathogenicity effectors (Supporting Table S13) were searched with PREG [97] (Supporting Table S13). We also attempted to identify high frequency novel motifs within the ‘effector-like’ candidates with MEME v4.9.1 (model = ANR, minsites = 2, maxsites = 300, nmotifs = 50, minwidth = 5, maxwidth = 50) [98] (Supporting Text S3).
Heterozygous SNP mutations derived from genomic read alignment to the final genome assembly, as described above, within all genes and predicted ‘effector-like’ genes were tested for: 1) stop-codon bias; 2) gene structure location bias with SNPeff [99]; 3) non-synonymous vs synonymous SNP ratio (dN/dS) via SNPeff [99]; 4) frequency and density via BEDtools coverageBed [73].
Gene expression of selected genes (Supporting Table S14A) was tested via quantitative polymerase chain reaction (qPCR) in wheat roots at 7 days post-infection and in 7 day old in vitro grown PDB culture. Wheat samples were inoculated with millet seeds pre-infect with WAC10335 and grown in pots of vermiculite for 7 days at 24°C. Wheat seeds were surface-sterilised and germinated on moist filter paper at 4°C for 4 days, then planted into pre-infected vermiculite pots and covered by a layer of fresh fine vermiculite. The pots were transferred to a growth room at 16°C and 12 hours light/day (150 µmol.m−2.s−1) for 7 days. Plants were harvested and root and above ground tissues separated. RNA was extracted from root tissue using Trizol reagent (Sigma) according to the manufacturer's instructions and cDNA produced using Superscript III (Invitrogen) following the manufacturer's instructions. Quantitative PCRs used SsoFast EvaGreen Supermix (BioRad).
A total of 29 out of 308 predicted ‘effector-like’ pathogenicity genes were selected for testing based on their assigned functional annotations. Seven control genes were also selected based on stable expression, averaging ≥70 FPKM and ≤0.1× fold change between libraries, across the three RNA-seq libraries discussed in this study and/or for putative functions suggesting stable expression patterns (e.g. actin and tubulin). Primer pairs were designed from coding-exon sequences (CDS) using primer3 [100] (opt. amplicon 200 bp, primer 18–25 bp, opt. Tm 60°C, max. ΔTm 1°C, min. GC clamp 2 bp, max. homopolymer 3 bp). In silico PCR screening via e-PCR [101] required ≤1 amplicon (10 bp to 10 kbp) versus genome assembly and CDS sequences. Quantitative PCR was performed with 2 technical replicates and 3 biological replicates. Log2 fold-changes between in-vitro and infection samples were calculated by the ΔΔCT method in accordance with Anderson et al. [102], relative to the mean of 7 controls. A two-tailed Student's T-test was applied to relative abundances between in planta and in vitro samples (equal variance, p-value≤0.05).
|
10.1371/journal.ppat.1005405 | A Viral microRNA Cluster Regulates the Expression of PTEN, p27 and of a bcl-2 Homolog | The Epstein-Barr virus (EBV) infects and transforms B-lymphocytes with high efficiency. This process requires expression of the viral latent proteins and of the 3 miR-BHRF1 microRNAs. Here we show that B-cells infected by a virus that lacks these non-coding RNAs (Δ123) grew more slowly between day 5 and day 20, relative to wild type controls. This effect could be ascribed to a reduced S phase entry combined with a moderately increased apoptosis rate. Whilst the first phenotypic trait was consistent with an enhanced PTEN expression in B-cells infected with Δ123, the second could be explained by very low BHRF1 protein and RNA levels in the same cells. Indeed, B-cells infected either by a recombinant virus that lacks the BHRF1 protein, a viral bcl-2 homolog, or by Δ123 underwent a similar degree of apoptosis, whereas knockouts of both BHRF1 microRNAs and protein proved transformation-incompetent. We find that that the miR-BHRF1-3 seed regions, and to a lesser extent those of miR-BHRF1-2 mediate these stimulatory effects. After this critical period, B-cells infected with the Δ123 mutant recovered a normal growth rate and became more resistant to provoked apoptosis. This resulted from an enhanced BHRF1 protein expression relative to cells infected with wild type viruses and correlated with decreased p27 expression, two pro-oncogenic events. The upregulation of BHRF1 can be explained by the observation that large BHRF1 mRNAs are the source of BHRF1 protein but are destroyed following BHRF1 microRNA processing, in particular of miR-BHRF1-2. The BHRF1 microRNAs are unlikely to directly target p27 but their absence may facilitate the selection of B-cells that express low levels of this protein. Thus, the BHRF1 microRNAs allowed a time-restricted expression of the BHRF1 protein to innocuously expand the virus B-cell reservoir during the first weeks post-infection without increasing long-term immune pressure.
| This paper explains some of the molecular mechanisms used by the Epstein-Barr virus (EBV) BHRF1 microRNA cluster to enhance transformation of B-cells after infection. We find that B-cells exposed to a virus that lacks the BHRF1 microRNAs (Δ123) undergo more apoptosis and grow more slowly between the second and the fourth weeks after infection than cells infected by an intact virus. These effects are partly mediated by the viral protein BHRF1, a homolog of the anti-apoptotic bcl-2 protein. The viral microRNAs allow abundant expression of BHRF1 early after infection and its down-regulation when transformation has been established. The first effect is mediated by the seed regions of miR-BHRF1-2 and -3, whereas the second is dependent on RNA cleavage mediated by processing of miR-BHRF1-2. Furthermore, we found that the ability of the BHRF1 microRNAs to increase cell cycle entry is related to their ability to downregulate PTEN, a crucial negative regulator of the cell cycle. We also study the consequences of the absence of the microRNAs for the infected cells. B-cells infected with Δ123 become more resistant to apoptosis and express lower levels of p27, two events that facilitate the development of genome instability. Thus, the viral microRNAs allow rapid and innocuous expansion of infected B-cells, their long-term reservoir, thereby facilitating the life-long coexistence between the virus and its host.
| The Epstein-Barr virus (EBV) is the first discovered tumor human virus and is etiologically associated with approximately 2% of all tumors worldwide [1, 2]. These tumors are largely diverse in terms of lineage and include multiple types of lymphomas and carcinomas [3]. Immune deficiency, e.g. caused by immunosuppressive regimen is a strong risk factor for the development of EBV-associated lymphomas [2]. These tumors are thought, at least to some extent, to reflect EBV’s ability to transform primary B-cells [2]. This process can be easily observed in vitro as it leads to the establishment of lymphoblastoid cell lines (LCLs) and requires the simultaneous expression of some members of the viral latent gene family [2]. In recent years, it has become clear that the BHRF1 microRNAs (miRNAs) encoded by the virus markedly potentiate this process. Recombinant viruses that lack one or several of these three miR-BHRF1s are less transforming than their wild type counterparts and the effect is cumulative [4–6]. One study has ascribed this property to the ability of the BHRF1 miRNAs to prevent massive apoptosis in the first days of infection [4]. Furthermore, viruses that lack the three BHRF1 miRNAs (Δ123) grow more slowly and display abnormalities of the cell cycle [4, 5]. Humanized NSG mice infected by Δ123 eventually develop B-cell proliferations that are indistinguishable from those caused by wild type infection, but cell growth induced by the mutant is delayed by several weeks, confirming that the BHRF1 miRNAs are particularly required in the early phases of infection [7]. The BHRF1 protein, around which the BHRF1 microRNAs are located, has also been implicated in EBV-mediated B-cell transformation, although its role appears to be more difficult to define. BHRF1 is a bcl-2 homolog that shares its anti-apoptotic properties [8, 9]. Although its expression level is hardly detectable in LCLs, it is strongly expressed in the Wp-restricted Burkitt’s lymphoma (BL) cells, a subset of Burkitt’s lymphomas that are infected by EBVs that carry a deletion of the EBNA2 gene and whose latent genes are driven by the Wp promoter [10, 11]. A recombinant virus that lacks the BHRF1 protein retains full transformation abilities, suggesting that this protein is dispensable for transformation [12]. However, its enhanced expression in Wp-restricted BLs leads to a markedly enhanced resistance to apoptosis induced by ionomycin [11, 13]. Thus, both the BHRF1 protein and the BHRF1 miRNAs have been implicated in the regulation of apoptosis.
The prominent role played by some of the EBV latent genes in B-cell transformation raises the question of a possible interaction of the BHRF1 miRNAs with the latent genes. Although these have not been directly identified in a search for the miR-BHRF1 targets, we have clearly identified the EBNA-LP latent gene as a, probably indirect, target of the BHRF1 miRNAs [5]. The expression of this protein is usually downregulated in LCLs after several weeks of growth in culture but this process is largely delayed after infection with Δ123 [5]. The other latent genes were also upregulated in LCLs infected by Δ123 relative to wild type counterparts but the effect was much weaker and inconstant. This raised the question whether the BHRF1 open reading frame is also a target of the BHRF1 miRNAs but its transcription level was not affected in cells infected by the mutant [4, 5].
In this paper we examine the role played by the complete BHRF1 locus during EBV infection. We found that the BHRF1 miRNA cluster controls the temporal expression of the BHRF1 protein but also downregulates PTEN. The deletion of this cluster also led to the frequent emergence of transformed B-cells with a downregulation of p27.
EBV infection of B-cells induces permanent cell division that gives rise to the establishment of lymphoblastoid cell lines (LCLs). Therefore, we began our investigations by monitoring cell growth and cell vitality over the first four weeks after infection with the Δ123 mutant or with wild type EBV controls. This was achieved by directly counting mitoses in the samples or staining cells with phospho-histone H3, a marker of cells undergoing mitosis. Both methods showed no evidence of cell division before day 3, as expected [14]. Cells infected with the wild type control then began dividing, reaching a peak at around day 10 and then maintained a constant mitotic rate between 1 and 2% (Fig 1a and 1b). The same B-cells infected in parallel with Δ123 differed from wild type controls in that their mitotic rate was 2–3 fold lower. However, after 25 days, both mutants and controls hardly showed any difference. These data suggested transient differences in cell cycle regulation between both groups of cells. Therefore, we performed a BrdU incorporation assay at day 13 post-infection (p.i.). This experiment showed a decrease in the fraction of cells that entered the S phase, as well as a relative increase in the number of cells present in G2/M in Δ123-infected cells (Fig 1c). This resulted in a statistically significant difference in the G2/M to S ratio between B-cells infected with Δ123 or with wild type virus. We then stained the same infected cells with an antibody specific to cleaved pro-caspase 3 that detects the form of the protein activated during apoptosis, coupled to a TUNEL assay that detects double strand DNA breaks. These assays are well suited for the detection of apoptosis at the single cell level and the results are summarized in Fig 1d and 1e. They showed that between day 1 and day 5, B-cells infected with wild type viruses or with Δ123 behaved identically. However, from day 8 on, the apoptosis rate grew larger in the cells infected with the miRNA triple mutant and became twice as high at day 15. The apoptotic rate decreased in these cells after day 18 to reach those evinced by wild type LCLs at day 34. Altogether, we conclude that cells infected by the Δ123 mutant do not enter the cell cycle as efficiently and undergo more apoptosis than the controls after initiation of cell division between day 8 and 20 after infection. As described in the sequel, transformation of additional B-cell samples revealed that the amplitude of the difference between B-cells infected by wild type or mutant viruses in terms of apoptotic rate and mitotic growth can vary. However, the general picture remained similar.
The BHRF1 miRNA cluster is located around the BHRF1 gene, whose protein product is endowed with anti-apoptotic properties [9, 11]. Therefore, we monitored BHRF1 protein expression by western blot around the critical period, between 1 and 20 days p.i. We used an EBV-negative clone of the Burkitt’s lymphoma cell line Elijah and the Wp-restricted cell line Oku as a negative and a positive control, respectively. This assay, shown in Fig 2a, revealed that the BHRF1 protein is transiently produced in cells infected by the wild type controls at levels in the range of those observed in Oku. Oku is known to express BHRF1 at much higher levels than established LCLs [11]. Expression began at day 1, became fully visible at day 3 and reached a peak at day 5, after which it decreased again to nearly disappear at day 18. In stark contrast, B-cells infected with Δ123 produced hardly detectable levels of the protein. We then performed northern blots at the peak of BHRF1 protein expression at day 5 in B-cells infected with EBV wild type or Δ123 (Fig 2b). This assay revealed that B-cells infected with wild type viruses produce multiple and abundant transcripts ranging from 1.3 to larger than 10 kb. In comparison, cells infected with Δ123 showed only large transcripts that were altogether much less abundant than in the wild type-infected LCLs. Thus, the reduced BHRF1 protein expression correlates with reduced BHRF1-specific transcription. We left the infected cells grow for another 55 days and repeated the experiment. The blots revealed that by that time the 1.3 kb band had become prominent in the LCL infected by wild type viruses with larger bands becoming much fainter. Interestingly, the LCL infected with Δ123 retained large BHRF1 transcripts, albeit slightly smaller than at day 5. Consequently, the large BHRF1 transcripts were more abundant in LCLs infected with Δ123 than in those generated with wild type controls. These results indicate that the wild type BHRF1 locus is transcribed at a different rate at an early and late stage of infection. However, whilst transcription markedly diminished with time in the wild type LCL, it remained nearly constant in cells infected with the mutant. We attempted to confirm these investigations by qPCR. We interrogated several BHRF1 transcripts, as indicated in Fig 2c, as well as those driven by the Wp promoter that dominates transcription at day 5. We found that LCLs infected with the wild type virus express much higher levels of BHRF1 transcripts at day 5 than at day 12. The same pattern was visible for W2-BHRF1 spliced transcripts and for Wp-driven transcripts. In the LCL infected with the Δ123 mutant, the BHRF1 transcripts at day 5 were expressed at a much lower level (10 to 20% of wild type level). They then increased to become more abundant than in the wild type controls, to fall again at approximately wild type levels after one month of infection. Thus, the usual pattern of Wp transcription, i.e. high levels shortly after infection that rapidly decrease after a few days, is translated to the right and reduced in its initial peak in cells infected with the mutant. We conclude from these findings that the BHRF1 locus undergoes very dynamic changes in terms of transcription over time and that BHRF1 transcription and translation is abnormally low at an early time point in LCLs infected by Δ123.
The latter results also suggested that the low abundance of the BHRF1 protein might be responsible for the observed increased apoptosis rate. We addressed this issue by studying the phenotype of a virus that carries an inactivating point mutation of the BHRF1 start codon (ΔBHRF1, see S1 Fig) and of a virus that lacks both the BHRF1 miRNAs and the BHRF1 open reading frame (Δ123ΔBHRF1). B-cells exposed to these viruses or to a wild type control showed little differences in apoptosis and cell growth until day 5 (Fig 3a, 3b and 3c). From day 6 onwards, apoptosis remained low in B-cells infected with wild type EBV but increased steadily in B-cells infected with the Δ123ΔBHRF1 double mutant to reach 100% between day 15 and day 30, depending on the blood samples tested. B-cells transformed by wild type viruses retained a mitotic rate of about 1.5% of the total cells after staining with PH3. In contrast, LCLs generated with the Δ123ΔBHRF1 double mutant exhibited a much lower mitotic rate that never exceeded 0.5%. LCLs transformed with ΔBHRF1 or Δ123 showed an intermediate profile between these 2 extreme phenotypes. In some of the 5 studied cases, the ΔBHRF1 LCLs and the Δ123-infected LCLs did not differ markedly from wild type-infected LCLs, with a good mitotic rate and a low level of apoptosis. In others, their behavior was closer to those of cells infected with the double knockout virus. In all cases, the LCLs generated with the ΔBHRF1 virus displayed higher mitotic rates than their Δ123 counterparts. We addressed this issue in more detail and performed a BrdU incorporation assay early after infection with wild type EBV, Δ123 and ΔBHRF1. The results of this experiment are depicted in Fig 3d and show that the percentage of cells in S phase is lowest in B-cells transformed with Δ123, highest in those transformed with wild type EBV and intermediate in cells transformed with ΔBHRF1. These data suggest that the mild increase in apoptosis and some of the cell cycle abnormalities observed in B-cells infected with Δ123 could be largely explained by the reduction in BHRF1 protein levels and that the low BHRF1 expression level, the only difference between B-cells infected with Δ123 and those infected with Δ123ΔBHRF1, becomes indispensable for survival of cells infected by a Δ123 virus.
We then performed transformation assays at low cell density and low MOI on feeder cells in 96-well cluster plates (Fig 3e). Feeder cells have previously been shown to reduce apoptosis in EBV-infected B-cells [15]. This assay showed that the B-cell transformation rate was highest in cells infected with wild type EBV, whilst B-cells infected with the Δ123ΔBHRF1 double knockout did not show any signs of outgrowth. The transformation assays performed with ΔBHRF1 or Δ123 showed again intermediate results. However, the transformation rate was much higher after infection with ΔBHRF1 than after infection with Δ123. Thus, the absence of BHRF1 but not those of the BHRF1 miRNAs can be largely compensated by feeder cells. Therefore, the Δ123 phenotype is not limited to a reduction in BHRF1 levels. This experiment also shows that ΔBHRF1 viruses are mildly less transforming than wild type viruses, an observation consistent with the moderate reduction in S phase entry observed previously. We tested whether it is possible at all to generate LCLs with the Δ123ΔBHRF1 double mutant. To this end, we infected B-cells from 4 different donors at high cell density (104 cells per well of a 96-well cluster plate) at an MOI of 10 infectious units per cell and kept the cells on a feeder cell layer for 1.5 months. This led to the establishment of LCLs in 2 out of 4 cases.
The results obtained with the Δ123 virus showed that the BHRF1 miRNAs are involved in the control of the BHRF1 protein production. We determined which of these miRNAs is implicated in this process by infecting primary B-cells with viruses that lack one of the BHRF1 miRNAs. Western blot analysis with a BHRF1-specific antibody revealed that infection of primary B-cells with a virus that lacks miR-BHRF1-1 or with the wild type control gave rise to the same level of BHRF1 protein production at day 5 (Fig 4a). In contrast, BHRF1 expression in B-cells infected with single and double miRNA mutants that lack miR-BHRF1-2, miR-BHRF1-3 or both (Δ2, Δ3, Δ23) was markedly reduced relative to B-cells infected with wild type viruses (Fig 4b and S1 Fig). We wished to confirm these findings by infecting B-cells with a virus that carries a seed mutation in miR-BHRF1-3 (3SM) (S1 Fig). This mutant expresses miR-BHRF1-1 and miR-BHRF1-2 at wild type levels (S2 Fig). Primary B-cells were infected in parallel with wild type virus, Δ123 and 3SM and harvested at day 5 post-infection. We performed an immunoblot with an anti-BHRF1 antibody that confirmed a clearly decreased BHRF1 protein expression after infection with 3SM, relative to wild type levels, although the amplitude of the effect was not as pronounced as after infection with the Δ3 virus (Fig 4c). This might point towards a role for miR-BHRF1-3* in this process. However, the low expression of this miRNA in LCLs argues against its role in the regulation of BHRF1 protein expression [16, 17]. We also performed this experiment with a mutant that carries mutations of both seed regions encoded by pre-miR-BHRF1-2 (2/2*DSM) (S1 Fig). Indeed, we previously showed that the Δ2 virus also evinces a reduced miR-BHRF1-3 expression [6]. 2/2*DSM expresses normal miR-BHRF1-3 levels and is therefore suitable to study the contribution of miR-BHRF1-2 to the regulation of BHRF1 protein expression (S2 Fig). B-cells infected with this mutant displayed no altered phenotype and expressed the BHRF1 protein at day 5 post-infection at approximately 60% of the levels observed in cells infected with wild type viruses (Fig 4d). Altogether, this set of experiments identify miR-BHRF1-3, and to a lesser extent miR-BHRF1-2, seed regions as positive modulators of BHRF1 protein expression early after infection. We then gauged the expression of the BHRF1 transcripts at the same early time point using qPCR (Fig 4e). This assay confirmed the reduced Wp-driven transcription in B-cells infected with Δ123, but also revealed that this effect was also visible in B-cells infected with Δ3, 2/2*DSM and 3SM. The BHRF1 transcripts were also less abundant in all these samples, although the Δ3 mutant showed more drastic effects than the 2 seed mutants, thereby confirming the data gathered at the protein level.
The BHRF1 protein can be produced either from a lytic or from a latent promoter. To determine which of these forms is produced at an early stage of infection we infected B-cells with a virus that lacks the BZLF1 gene (ΔZ) that encodes a transactivator indispensable for the onset of lytic replication in B-cells (Fig 4f). This experiment showed that BHRF1 protein production is only slightly reduced at day 5 in B-cells infected with the ΔZ mutant, relative to wild type virus. Thus, it is the latent form of the BHRF1 protein that is predominantly produced at an early time point as previously suggested [11].
The data gathered so far confirmed that the B-cells infected with Δ123 have a reduction in cell cycle entry, even on feeder cells under conditions in which apoptosis is limited. We knew from previous work that a virus that lacks miR-BHRF1-3 displays similar, if less pronounced, abnormalities [6]. We looked for a gene implicated in the cell cycle control that would be regulated by miR-BHRF1-3. PTEN has previously been identified by a PAR-CLIP method as a potential target of miR-BHRF1-3 [17] (Fig 5a). Therefore, we tested expression of this protein at different time points after infection with Δ123 and wild type controls and found that it increased in intensity regularly from day 1 to day 5. After day 5 it became obvious that B-cells infected with the Δ123 virus expressed more PTEN than wild type counterparts (Fig 5b). We wished to confirm that this effect was due to the absence of miR-BHRF1-3 and assessed expression of PTEN in cells infected with the Δ3 or the 3SM virus. This experiment confirmed that cells infected with either of these single miRNA mutants evinced a stronger PTEN expression relative to wild type (Fig 5c and 5d). We also quantified PTEN expression in LCLs generated with ΔBHRF1 (S3 Fig). This assay could not reveal any difference in PTEN expression in these cells, relative to wild type controls.
We went on to perform a luciferase reporter assay in which the luciferase gene is fused with part of the 3’UTR of PTEN that contains the putative miR-BHRF1-3 binding site. We also included a negative control in which the putative miR-BHRF1-3 binding site had been mutated (seed-match mutant). Cotransfection of either of these constructs together with a miR-BHRF1-3 expression plasmid revealed a modest but statistically significant decrease in relative luciferase activity in the wild type PTEN 3’UTR fusion that was not visible in the seed-match mutant control (Fig 5e). This weak effect can at least in part be ascribed to the intrinsic low miR-BHRF1-3 expression level [6].
We then treated 2 14-days old EBV wild type-infected LCLs with wortmannin, an inhibitor of PI3K that mimics an activation of PTEN [18]. We found that the treatment of these cells with wortmannin decreased entry in S phase by one third and increased the G2/M to S phase ratio to a value close to the one we observed in the LCLs infected by Δ123 (Fig 5f). Thus, wortmannin treatment reproduced the cell cycle abnormalities observed after excision of the BHRF1 miRNAs. This supports the idea that the relative excess of PTEN seen in LCLs generated with Δ123 is responsible, partly or entirely for the observed abnormalities in cell cycle entry.
We then turned our attention to infected cells that survived the critical day 5 to day 20 period and measured the expression of BHRF1 in cells established for more than 20 days. We found that, in accord with previous observations [11], established cell lines generated with wild type controls, hardly express BHRF1 (Fig 6a). In contrast, LCLs generated with Δ123 showed a clear expression of the protein. We repeated this experiment for 3 additional B-cell donors and obtained similar results that are given in Fig 6a. These results suggested that the BHRF1 mRNA that is used for translation of the protein could also be used as a template for miRNA processing. Therefore, we performed a northern blot on polyadenylated RNAs with a probe specific for the 3’ end of the BHRF1 gene (Fig 6b). This assay showed that the 0.5 kb transcript, that can only be generated by processing of the polyadenylated BHRF1 mRNA transcript at miR-BHRF1-2 or miR-BHRF1-3 (Fig 6c), was present in B-cells infected with wild type EBV but not in those infected with Δ123, confirming that the BHRF1 mRNA is cut by miRNA processing. To find out which of the BHRF1 miRNAs is responsible for this process, we infected B-cells with the Δ1, Δ2 and Δ3 mutants as well as with double mutants and measured BHRF1 expression in these LCLs. We found that only the LCLs infected by a virus that lacks miR-BHRF1-2 expressed higher levels of BHRF1 protein, demonstrating that processing of miR-BHRF1-2 cuts the potentially translated BHRF1 mRNA (Fig 6d). We then performed a northern blot with the same cells and could confirm that the BHRF1 3’UTR is not cut in the absence of miR-BHRF1-2. Thus, this miRNA plays an important role in the control of BHRF1 protein expression (Fig 6e). In the absence of miR-BHRF1-3, the 3’UTR was cleaved at miR-BHRF1-2. This resulted in a slightly larger signal in the northern blot. Thus, miR-BHRF1-2 and miR-BHRF1-3 are sequentially processed resulting in a cleavage of the primary BHRF1 transcript. This is consistent with our previous observation that efficient miR-BHRF1-3 processing requires the presence of miR-BHRF1-2 [6].
The set of BHRF1 miRNA mutants gave us the opportunity to investigate the mechanisms used by the virus to express the BHRF1 protein and more generally the transcription of the BHRF1 locus in more detail. To this end, we used again northern blots to assess the nature of the transcripts that entail the BHRF1 gene with probes specific to the BHRF1 open reading frame, its intron, or to the origin of lytic replication (oriLyt) which is located directly 5’ of the BHRF1 locus (S4 Fig). This analysis led us to propose a model in which large unspliced polyadenylated transcripts encompassing the oriLyt region and the BHRF1 ORF are a source of BHRF1 protein. These transcripts also yield BHRF1 miRNAs, whose processing give rise to smaller and smaller RNA products.
Hybridization of total RNA from wild type-infected LCLs with the BHRF1 ORF probe revealed a prominent 1.3 kb transcript that contains the BHRF1 intron but neither oriLyt not the BHRF1 3’UTR. It also evidenced the existence of larger transcripts of very faint intensity, some of which are larger than 8 kb. This pattern reproduces what we previously observed with other wild type LCLs at 60 days and suggests that the difference in the pattern of BHRF1 transcripts observed at day 5 post-infection might be related to the fact that BHRF1 miRNAs are not fully expressed at that time, or at least not proportionally to total BHRF1 transcription (Fig 2b). Indeed, in mutants that lack miR-BHRF1-1 or miR-BHRF1-2, the 1.3 kb transcript shifted to transcripts that were larger than 8 kb in size and of variable intensity, depending on the mutants. In mutants that lack miR-BHRF1-2, there was in addition a band at approximately 2.3 kb. Importantly, miR-BHRF1-3 is not required for the generation of the 1.3 kb transcript. Thus, the 1.3 kb transcript is generated from larger fragments through the processing of miR-BHRF1-1 and miR-BHRF1-2, entails the BHRF1 intron but not oriLyt, and has the size of the BHRF1 RNA fragment between these 2 miRNAs. In all probability, it is identical to the transcript generated by miRNA processing that was previously identified [19, 20]. This 1.3 kb transcript was already visible at day 5, but, in contrast to what we saw in established cell lines, larger signals of equal or even stronger intensity were also present (Fig 2b). However, it is important to note that the total BHRF1 transcription at 1.5 to 2 months post-infection is much reduced compared to the first days post-infection, suggesting that processing of the BHRF1 miRNAs leads to a disappearance of high molecular weight BHRF1 RNAs, except if the BHRF1 transcripts are in massive excess.
The investigation of poly A+ mRNAs revealed that LCLs produced additional RNAs. LCLs infected with wild type controls contain one clear signal at 2.2 kb as well as larger, much fainter signals. The 2.2 kb fragment contains the BHRF1 intron, the BHRF1 ORF but not the oriLyt fragment. It is absent in LCLs infected with mutants that lack miR-BHRF1-1 or -3 but is very abundant and slightly larger in cells infected with mutants that lack miR-BHRF1-2. The LCLs infected by viruses that lacked miR-BHRF1-1 or -2, including the Δ123 mutant, also displayed an accumulation of transcripts ranging from 4 to larger than 10 kb, in line with the assumption that large BHRF1 transcripts serve as a source of these miRNAs. The dominant role of miR-BHRF1-2 processing was also visible in the LCLs generated with Δ13 in that they also produced the large transcripts, but at a lower expression level than those infected with viruses that lack this miRNA. These large transcripts contain the BHRF1 intron, as well as the oriLyt sequence. We conclude that the 2.2kb transcript is generated from large transcripts that encompass oriLyt, contains the BHRF1 intron and open reading frame and the BHRF1 3’UTR, requires the presence of miR-BHRF1-1 for its generation and accumulates in the absence of miR-BHRF1-2. Taking into account that the distance between miR-BHRF1-1 and the end of the BHRF1 polyA tail is 2.2 kb, we conclude that the 2.2 kb RNA represents a polyadenylated transcript that begins at miR-BHRF1-1 and runs through to the BHRF1 polyA. It is important to note that this transcript runs slightly higher in the LCLs infected by Δ23 and even higher in those infected with Δ2. This fits with the fact that the BHRF1 transcript is 46 and 126 ribonucleotides longer in the Δ23 and the Δ2 mutants, respectively, than in the wild type BHRF1 gene. Altogether, we conclude that the 2.2 kb polyadenylated transcript is very likely to represent a precursor form of the 1.3 kb transcript identified in the total RNA blot that is itself generated upon miR-BHRF1-2’s processing. The abundance of the larger BHRF1 transcripts was inversely related to the existence of the 2.2 kb transcripts and their existence was strongly dependent on the absence of miR-BHRF1-2. In the absence of miR-BHRF1-1, there were also some larger but less strongly expressed high molecular weight BHRF1 transcripts. This points to the central role of miR-BHRF1-2 in the processing of these large transcripts. We noticed that there is an inverse correlation between the expression of the BHRF1 protein and the presence of shorter BHRF1 RNA forms. This suggests that the large polyadenylated mRNAs that encompass BHRF1 are the source of the BHRF1 protein and that they are destroyed during BHRF1 miRNA processing, in particular though processing of miR-BHRF1-2. LCLs generated with Δ3 express high levels of miR-BHRF1-2 [6] and this fits with the observation that this cell line does not express any of the BHRF1 intermediates but only their final product, the 1.3 kb RNA fragment.
We then went on to assess the functional consequences of BHRF1 overexpression. To this end, we provoked apoptosis in LCLs established with B-cells from multiple donors and generated by infection with Δ123 or with wild type controls. This was achieved by incubating the cells with etoposide, staurosporine or simvastatin for a variable length of time and staining these cells with Annexin-V and 7AAD (Fig 7a, 7b and 7c). Treatment with these drugs gave rise to massive cell death in all samples, although the killing efficiency was significantly higher in the wild type LCLs than in LCLs generated with Δ123. We confirmed these data with a PARP cleavage assay (Fig 7d). Whilst some cells infected with Δ123 retained some intact PARP after treatment with etoposide or staurosporine, this was not the case for cells infected with wild type viruses. We then performed similar experiments with B-cells infected with ΔBHRF1 and Δ123ΔBHRF1. We found that LCLs generated with ΔBHRF1 or with wild type viruses do not significantly differ in their response to apoptotic stimuli (Fig 7e, 7f and 7g). This fits with the observation that BHRF1 is expressed at very low levels in established LCLs [11]. Fig 7h, 7i and 7j show that B-cells infected with Δ123ΔBHRF1were much more sensitive to provoked apoptosis than Δ123 LCLs, and even showed more cell death upon induction than the wild type controls. This suggests that in the absence of the BHRF1 protein, infected B-cells rely on the BHRF1 miRNAs to withstand apoptotic stimuli. These results concur with our observation that at an early time point after infection, B-cells infected with a virus such as Δ123 that expresses low BHRF1 protein levels are crucially dependent on this residual activity as seen by infection with the Δ123ΔBHRF1 mutant (Fig 3).
We assessed the cell cycle characteristics of LCLs that had been established for more than 35 days using a BrdU incorporation assay (Fig 8a). Although the percentage of cells in S phase remained lower in LCLs infected by Δ123 relative to those generated with wild type viruses, the ratio between these numbers increased from 0.42 (Fig 1c 5.7/13.5) to 0.65 (Fig 8a 14.7/22.6). We also generated growth curves with LCLs transformed by Δ123 and wild type virus and found that both types of LCLs grow at very similar rates (Fig 8b). However, as seen in Fig 8c, PTEN levels remained higher in cells infected with the Δ123 mutant relative to wild type. Therefore, LCLs infected with Δ123 must have acquired additional changes in their cell cycle regulation after several weeks in culture. We screened the expression of key regulators of the cell cycle, including p53, Rb and the CDKN family in the different types of LCLs by western blot. This analysis revealed that CDKN1/p27 is frequently expressed at lower levels in established LCLs infected by the Δ123 mutant relative to wild type controls after more than one month in culture (Fig 8d). We monitored expression of p27 during the day 5 to day 18 crucial period and found that the protein levels of p27 decreased rapidly after infection to nearly disappear at day 13 p.i. However, this expression re-increased after day 18, albeit less strongly in B-cells infected with the Δ123 mutant (Fig 8e). These data argue against a direct impact of the BHRF1 miRNAs on p27 and establish a parallel between the decrease in p27 and the recovery of the cell growth rate in B-cells infected with Δ123.
The clarification of the molecular mechanisms fine tuned by miRNAs is frequently a difficult undertaking and the BHRF1 miRNAs are no exception to that rule. CLIP-based strategies have identified potential targets of these miRNAs but the identity of the crucial proteins that they modulate remains enshrouded in mystery [17]. We have evaluated in detail the role served by the BHRF1 miRNAs during the first weeks of EBV-mediated B-cell transformation and found that cells infected with a virus that lacks the three BHRF1 miRNAs undergo on average twice as much apoptosis than cells infected with wild type controls between day 5 and day 20. Our results are only partially concordant with an earlier report that described a massive apoptosis in cells infected with a virus devoid of the BHRF1 miRNAs during the first five days of infection, followed by a quick recovery of cell numbers 10 days after infection [4]. These data are difficult to reconcile with the fact that the BHRF1 miRNAs are fully produced only 5 to 8 days after the onset of infection [21, 22] and we indeed did not see any differences between cells infected with the Δ123 mutant and cells infected with wild type controls during the first 5 days of infection. One difference between the studies lies in that we investigated the infected B-cells at the single cell level and used a larger panel of markers to characterize apoptotic cells. Indeed, positive staining for Annexin-V used in the study by Seto et al. is not specific for apoptosis and is found in many other modes of cell death [23].
We found that the expression of the viral bcl-2 homolog BHRF1 is modulated by the BHRF1 miRNAs in an unexpectedly complex manner. During the first days of infection, the BHRF1 protein has previously been reported to be expressed at relatively high levels and we could confirm this observation [11]. However, B-cells infected with the Δ123 virus hardly express BHRF1, suggesting that the observed increased apoptosis after infection with Δ123 is due to a reduction in BHRF1 expression. Indeed, a virus that lacks the BHRF1 protein and a virus that lacks the BHRF1 miRNAs induce similar phenotypes in infected B-cells. Interestingly, cells infected with ΔBHRF1 not only exhibited an increased apoptotic rate but also a reduced entry into cell cycle. At the present stage of the work, it is unclear whether the cell cycle abnormalities are a consequence of increased apoptosis, or whether this reflects a specific so far unrecognized, PTEN-independent, property of the BHRF1 protein. Our results are discordant with those obtained by Altmann et al. who found that only a virus lacking both BHRF1 and BALF1 has a reduced transforming ability as a consequence of enhanced apoptosis upon infection of B-cells [12]. However, in this study, the authors showed growth curves of B-cells infected with a virus that lacks the BHRF1 protein that are reminiscent of our own results and clearly differ from those obtained with wild type virus. Moreover, these authors also discussed the possibility that a virus that lacks BHRF1 only might also induce an abnormal phenotype without reaching definite conclusions.
Importantly, B-cells infected with a virus that lacks both the BHRF1 miRNAs and the BHRF1 protein undergo a strong degree of apoptosis starting at day 5 post-infection and increasing until day 20 at which time infected cells die. Thus, the phenotype induced by Δ123ΔBHRF1 is much more severe than the one induced by Δ123. Importantly, BHRF1 is expressed, although at low levels, shortly after infection with Δ123. This suggests that the very low amounts of BHRF1 are sufficient to limit the apoptosis in cells infected by Δ123. However, it is important to note that the construction of the Δ123ΔBHRF1 mutant led to a complete deletion of this gene. Therefore, although it is unlikely, we cannot exclude that so far unidentified genetic elements might be altered in this mutant. Moreover, we found that the low BHRF1 expression level in established LCLs infected with wild type EBV has no significant impact on drug-induced apoptosis, as a LCL infected with a virus that lacks the BHRF1 protein but expresses the BHRF1 miRNAs is not more sensitive to provoked apoptosis than wild type controls. This underlines again the role of the BHRF1 miRNAs in the regulation of the apoptotic status in the absence of the BHRF1 protein.
The observation that the phenotype induced by Δ123ΔBHRF1 is more severe than the one observed in B-cells infected with ΔBHRF1 also indicates that the function of the BHRF1 miRNAs is not subsumed by the modulation on BHRF1 protein levels. Therefore, we looked for additional targets of the BHRF1 miRNAs and the recognition that B-cells infected by Δ123 display a reduced entry in cell cycle oriented the search. We also knew that a recombinant virus lacking miR-BHRF1-3 displays some cell cycle abnormalities and therefore looked for potential miR-BHRF1-3 targets identified in the PAR-CLIP assays [6, 17]. This led to the identification of PTEN as a protein whose expression is downregulated by miR-BHRF1-3. This effect was visible as early as day 5 post-infection with Δ123 and persisted in established LCLs. Treatment of wild type LCLs with the PI3K inhibitor wortmannin gave rise to a decrease in cell cycle entry, reproducing the cell cycle alterations observed in B-cells infected with Δ123 and suggesting that the PI3K pathway is active in EBV-transformed B-cells, as previously suggested [24]. Brennan et al. found that inhibition of PI3K leads to a downregulation of cyclin D2 and cyclin D3, combined with an increase in p27 that causes growth arrest [24]. PTEN is an inhibitor of PI3K, and its increased expression in LCLs is likely to reduce the PI3K activity in these cells. Therefore, downregulation of PTEN through miR-BHRF1-3 is expected to facilitate cell division in LCLs generated with wild type EBV. PTEN is also targeted by cellular miRNAs such as the miR-17~92 cluster, that is expressed in LCLs [17, 25, 26]. Thus, viral and cellular miRNAs seem to collaborate to downregulate its expression. Moreover, PTEN, through its repression of the PI3K/Akt pathway, also negatively modulates the apoptotic status of the cell [27]. We found that miR-BHRF1-3 is able to weakly downregulate a luciferase-PTEN 3’UTR reporter gene. However, the difficulties in expressing miR-BHRF1-3 render interpretation of these results difficult. Altogether, these results, in combination with those of the PAR-CLIP assay suggest that miR-BHRF1-3 directly targets PTEN, but other mechanisms cannot be excluded.
We found that the BHRF1 miRNAs influence the dynamic expression profile of Wp-driven and BHRF1-specific transcripts. In their absence, the peak of transcription shortly after infection was delayed by about one week, after which transcription remained higher than in the wild type counterparts. We could show that BHRF1 protein expression depends on miR-BHRF1-3 and miR-BHRF1-2 seed regions and identification of their targets should shed light on their exact mechanism of action.
After 30 days, the infected cells recovered a normal growth rate, as assessed by growth curves and PH3 staining. The cell cycle profile of B-cells infected with Δ123 substantially improved at this point, with an increased proportion of cells that enter the S phase, although it did not reach wild type levels. These LCLs were also more resistant to drug-induced apoptosis.
Accordingly, the BHRF1 protein expression profile differed between established LCLs and freshly infected B-cells. LCLs established with Δ123 expressed more BHRF1 protein than their wild type counterparts and this explains the increased resistance to apoptosis, as LCLs infected with Δ123ΔBHRF1 or ΔBHRF1 have lost this property. We could identify miR-BHRF1-2 as the miRNA that is mainly responsible for the downregulation of the BHRF1 protein after establishment of the LCLs. The sequential processing of miR-BHRF1-2 and miR-BHRF1-3 that results in the cleavage of the 3’ end of the BHRF1 mRNA suggests that miR-BHRF1-2 does not solely act through the classical RISC-mediated target mRNA regulation, but highlights another mechanism through which a miRNA regulates protein expression. Therefore, we have shown that the latent BHRF1 mRNAs can act as a template for BHRF1 translation but can also be used by the Microprocessor machinery to generate the BHRF1 miRNAs and that both processes are in competition for access to the mRNA. The alternative use between protein translation and miRNA processing was previously reported for the cellular gene follistatin and miR-198 that are coded on the same mRNA [28]. Northern blot analyses have shown that infected B- cells produce abundant probably Wp-initiated high molecular weight mRNAs that contain the BHRF1 ORF. The exact identity of the BHRF1 transcripts that are translated remains unknown but we know that they contain the BHRF1 intron. Furthermore, analysis of established LCLs has also shown that high molecular weight BHRF1 RNAs are polyadenylated and probably also contain oriLyt-specific sequences. BHRF1 transcripts initiated at oriLyt have been previously identified and could potentially be the source of the protein [29–31]. Moreover, large polyadenylated RNAs containing intronic BamHI W sequences linked to the EBNA-LP, EBNA2, and BHRF1 polyA sequences have been also detected in EBV-infected cells [29].
The observation that the same miRNAs can successively activate and repress BHRF1 protein expression is puzzling but we could identify clear differences between these 2 stages. The BHRF1 transcription rate is much higher at an early time point than after establishment of the LCL. At this time, Wp-driven transcription dominates and it has previously been found that this promoter drives the expression of BHRF1 [11]. Furthermore, we also found that whilst the stimulatory effects of miR-BHRF1-2 and -3 depended on their seed regions, the repressive effects of the BHRF1 cluster was mainly mediated by miR-BHRF1-2 processing. We hypothesize that, at the beginning of infection the strong Wp-driven BHRF1 transcription, boosted by miR-BHRF1-2 and -3 favors BHRF1 translation. This would imply that high transcription rates disadvantages miRNA processing or that Wp-driven transcripts are less accessible to the Microprocessor machinery. At this stage, the ratio between transcripts processed or not by the Microprocessor machinery allows BHRF1 protein synthesis. When Wp-mediated transcription ceases, BHRF1 transcription is reduced, the BHRF1 miRNAs are efficiently processed and the BHRF1 translation is reduced to a minimum. Why would infected cells first express BHRF1 and then downregulate its expression? Our data show that BHRF1 expression protects infected B-cells against apoptosis at the beginning of the infection. The kinetic of expression of BHRF1 and of the viral latent membrane protein 1 (LMP1), two proteins endowed with anti-apoptotic properties [9, 32], is strikingly opposite. LMP1 is expressed at low levels shortly after infection and reaches its plateau of expression only 21 days after infection, which is exactly the period at which the increased apoptotic rate in cells infected with Δ123 normalizes [33]. Thus, it is possible that BHRF1 assumes LMP1 anti-apoptotic functions until this protein reaches its optimal expression level. However, virus-infected cells need to protect themselves from the immune response. As BHRF1 is efficiently targeted by the T cell response, its expression needs to be downregulated to a minimum when it is not anymore needed [34].
In contrast to BHRF1, PTEN levels remained high in cells infected by Δ123 and this is likely to explain the persisting reduced entry in S phase in these cells. This fits with the concept that this protein is targeted by miR-BHRF1-3, whose levels reach a plateau after day 5 to 8 in infected cells [21, 22]. Nevertheless, the improvement in S phase entry in established LCLs prompted us to screen the expression of cell cycle regulators in these cells and we identified p27 as a protein whose expression is reduced after 30 days in culture. The expression of p27 varies markedly over time in LCLs and follows a complex pattern. As expected for non-dividing cells, the expression of this protein was high one day after infection. It decreased then to become hardly visible 2 weeks after infection. However, it then started to increase again but remained lower from this stage in cells infected with Δ123 viruses.
The reduced expression of p27 suggests that infected cells counteract the effects of PTEN overexpression. Indeed, PTEN blocks the repressive effects of Akt on p27 and thus increases p27 expression [27]. At this stage, we cannot distinguish between a selection process that facilitates the growth of cells with low p27 in the context of increased PTEN expression and an indirect effect of the BHRF1 miRNAs. We favor the first possibility because the expression of p27 re-increases in wild type LCLs, even in the presence of the BHRF1 miRNAs and thus seems to be independent of them.
In conclusion, we have identified BHRF1, PTEN and p27 as direct or indirect targets of the BHRF1 miRNA cluster. These 3 proteins regulate the apoptotic status and entry into S phase, two essential cell functions. Although it is likely that other important targets of the BHRF1 miRNAs remain to be discovered, many of the phenotypic traits evinced by the Δ123 can be explained by the modulation of these 3 proteins. The expression of the BHRF1 miRNAs increases BHRF1 protein production and reduces PTEN production after 5 days post-infection to facilitate cell division. At a later time point, the BHRF1 miRNAs reduce expression of the BHRF1 anti-apoptotic protein and indirectly increase expression of p27, two events associated with a reduced propensity to neoplastic cell transformation [35, 36] that is beneficial for long-term persistence in the host. BHRF1 has gained increasing attention in recent years for its important function in LCLs and in Burkitt’s lymphomas. BHRF1 also represents a potentially very attractive therapeutic target [37]. It is therefore not surprising that its expression is tightly modulated within the infected cell. It is interesting to note that the appearance of a miRNA cluster within the BHRF1 locus is relatively new during evolution and is restricted to gammaherpesviruses with a tropism for B-cells and may be related to the need to restrict efficient MHC class I and class II presentation from B-cells [34, 38, 39].
Cells were kept in RPMI-1640 (Life Technologies) supplemented with 10% FBS (Sigma) at 37°C in a 5% CO2-buffered humid atmosphere. Primary B-cells infected with EBV were kept in RPMI/20%FBS until establishment of the cell line and 100μg/ml Hygromycin B (Calbiochem) was added to HEK 293 producer cells to induce selection pressure in cells stably carrying the EBV-BAC. HEK 293 cells are neuro-endocrine cells obtained by transformation of embryonic epithelial kidney cells with adenovirus (ATCC: CRL-1573). Primary B-cells were isolated from adult human blood buffy coats by Ficoll (GE healthcare) density gradient centrifugation and the CD19+ B-cell population was purified using CD19 PanB Dynabeads (Life technologies) and DETACHaBEAD CD19 (Life technologies). Elijah-5E5 is an EBV-negative subclone of the EBV positive Burkitt’s lymphoma cell line Elijah (kindly provided by A.B. Rickinson). BJAB is an EBV-negative Burkitt’s lymphoma cell line (kindly provided by A.B. Rickinson) [40]. Oku-BL is a Wp-restricted Burkitt’s lymphoma cell line (kindly provided by A.B. Rickinson) [10]. WI38 are primary human lung fibroblasts (ATCC: CCL-75 [41]).
All human primary B-cells used in this study were isolated from anonymous buffy coats purchased from the Blood Bank of the University of Heidelberg. No ethical approval is required.
The recombinant EBV wild type BAC (B95-8 wt/2089) was constructed by introducing the bacterial F-factor, GFP, a chloramphenicol resistance gene and a hygromycin selection marker in the EBV strain B95-8 [42]. The construction of the miR-BHRF1 deletion mutant and revertant [5], of the miR-BHRF1 single mutants [6] and of the BZLF1 deficient virus ΔZ [43] were reported previously. The Δ123ΔBHRF1 mutant was obtained by exchanging the complete BHRF1 locus of the B95-8 wild type BAC (EBV coordinates 53758:55278 (V01555.2)) with a kanamycin resistance cassette by homologous recombination [44]. This cassette was amplified from pCP15 using the primers 1422 (ATGTGGGGGT GGAAATATGA GCAAGAATAA GGACGGCTCC AACAGCTATG ACCATGATTA CGCC) and 683 (ATTTTAACGA AGAGCGTGAA GCACCGCTTG CAAATTACGT CCAGTCACGA CGTTGTAAAA CGAC). We used En Passant Mutagenesis [45] to construct a BHRF1 deficient recombinant virus (ΔBHRF1) in which the BHRF1 ATG start codon (54376:54378 (V01555.2)) was replaced by ATTAG in GS1783, a bacterial strain that contains the B95-8 wild type BAC. To amplify the kanamycin resistance gene from the pepKanS plasmid, the primers 1855 (CCTCTTAATT ACATTTGTGC CAGATCTTGT AGAGCAAGAT TAAGTAGGGAT AACAGGGTAA TCG) and 1856 (TATACACAGG GCTAACAGTA TCTCCCTTGT TGAATAGGCC TAATCTTGCT CTACAAGATC TGGCACAAAT GTAATGCCAG TGTTACAACC AATTAACC) were used.
To generate the 3SM seed mutated recombinant virus, three point mutations were introduced in the seed region of miR-BHRF1-3 (AACGGGA converted to AACGTTG). To this end, the nucleotides in the mature seed as well as the complementary miR-BHRF1-3* strand (55261:55263 and 55307:55309 (V01555.2)) were mutated using En Passant mutagenesis. The kanamycin resistance gene from the pepKanS plasmid was amplified using the primers 2125 (CAATTGGGTG TCCTAGGTGG GATATACGCC TGTGGTGTTC TAACGTTGAG TGTGTAAGCA CACACGTAAT TTGCAAGCGG ATAAGTAGGG ATAACAGGGT AATCG) and 2126 (CTCAGTTATT TCTTTAGTAT CTTGTCCTTG TGTTATTTTA ACGCCAAGCG TGAAGCACCG CTTGCAAATT ACGTGTGTGC TTACACACTC AACGTTAGCC AGTGTTACAA CCAATTAACC) and introduced into B95-8 Δ3 in GS1783.
Successful construction of all clones was verified by sequencing of the mutated region and integrity of the complete genome was confirmed by restriction enzyme digestion of the BAC clones and of the stably transfected HEK 293 producer cells.
Induction of HEK 293 producer cells to generate virus supernatants for B-cell infection and the quantification of virus titers were performed as previously described [5].
106 primary B-cells were infected using a MOI of 10 EBV genome equivalents per cell for 2 hrs at RT and cultured in RPMI/20%FBS. Four to five days after infection with EBV, B-cells initiate permanent growth that gives rise to the establishment of cell lines termed lymphoblastoid cell lines (LCL). For transformation assays, 5×102 cells were infected with a MOI of 0.01, that is enough viruses to induce GFP expression in 1% of infected cells [5], for 2 hrs at RT and plated on 96-well plates coated with 50Gy irradiated WI38 fibroblasts. 48 wells per donor and virus supernatant were seeded and the number of transformed wells was determined 30 days post-infection.
Cells were washed once in 1xPBS, spread on glass slides (Medco) and air-dried. For phospho-histone 3 staining (PH3; 1:100; Cell Signaling), cells were fixed in 4% PFA for 20 min at RT, washed 5 min in 1xPBS, permeabilized by immersion in 1xPBS/0.5% Triton-X for 2 min and washed again for 5 min in 1xPBS. Cells were incubated with the primary antibody diluted in PBS/10%HINGS (heat inactivated; Gibco) for 30 min at 37°C in a humid chamber and washed 3 times 5 min in 1xPBS. Secondary goat anti-rabbit Cy3 conjugated antibody (1:1200; Dianova) was applied for 30 min at 37°C in a humidity chamber, cells were washed 3 times for 5 min in 1xPBS and the DNA was counterstained using Hoechst 33258 for 2 min at RT. The cells were embedded in 90% glycerol and analyzed by fluorescence microscopy (Leica DM5000 B).
A triple staining of the mitotic spindle (α-tubulin), the centromeres (centrin-2) and the DNA (DAPI) was used on cells flattened by cytospin to analyze mitosis. To this end, cells were harvested, washed twice in 1xPBS/3% FBS and 5×104 cells were spun on slides (Tharmac, Cytoträger) in 100μl PBS/3%FBS using the cytospin 4 (Thermo; EZ single cytofunnel, Thermo) for 10 min at 2000rpm, maximum acceleration. Cells were air-dried, fixed in PFA and permeabilized using PBS/Triton-X as described above. Unspecific protein binding was blocked for 45 min in PBS/3%BSA at room temperature in a humid chamber. Cells were stained with rabbit α-centrin-2 (1:100; Santa Cruz) and mouse α-α–tubulin (1:4000; Sigma) in PBS/3%BSA for 2 hrs at 37°C in a humidity chamber, washed 5 times in 1xPBS, followed by incubation for 2 hrs in the secondary antibodies diluted in PBS/3%BSA (goat α-mouse IgG-Alexa488, 1:300, Invitrogen; goat α-rabbit Cy3, 1:1200, Dianova). After washing 5 times in 1xPBS, the cells were embedded in ProLong Gold antifade reagent with DAPI (Life technologies) and analyzed at a magnification of 630x.
For BrdU incorporation assays, cells were adjusted to 5×105 cells per ml one day prior to cell cycle analysis. 5×105 cells were pulsed with 10μM BrdU for 35 min at 37°C and stained using the APC BrdU Flow Kit (BD Pharmingen) according to the manufacturers instructions. The cell cycle profile was determined with a FACSCalibur flow cytometer. This assay allowed recognition of cells in G1/G0 (BrdU negative, 7AAD single DNA content), S phase (BrdU positive, shifting from single to double DNA content (7AAD)) and the G2/M phase (BrdU negative, double DNA content (7AAD)).
The growth rate of established LCLs was determined using the trypan blue (Sigma) dye exclusion method. The number of viable cells was determined 24 hrs and 48 hrs after adjusting the cultures to a density of 3x105 cells per ml.
Cells were harvested at indicated time points, washed once in 1xPBS, spread on glass slides, fixed and stained for cleaved caspase 3 (Casp3; 1:400; Cell Signaling) as described above for PH3 and single cells were analyzed by fluorescence microscopy. For detection of apoptosis using the TUNEL technology (terminal deoxynucleotidyl transferase (TdT)-mediated dUTP nick end labeling), cells were washed once in 1xPBS, spread on glass slides (Medco), the DNA double strand nicks were enzymatically labeled with the In Situ Cell Death Detection Kit, TMR red (Roche) according to the manufacturers instructions and analyzed by fluorescence microscopy.
Following induction of apoptosis, cell death was determined using Annexin-V-Alexa647 (Roche). To this end, 106 cells were pelleted, washed once in 1xPBS, incubated for 15 min at room temperature in 100μl Annexin-V-binding buffer (Annexin-V (3μl/106 cells), 7AAD (eBiosciences; 4μl/106 cells) in 10mM Hepes pH 7.4, 140mM NaCl, 5mM CaCl2). Annexin-V positive cells were quantified using a FACSCalibur flow cytometer.
Apoptosis was induced in LCLs of different B-cell donors 2–3 months after infection. Cells were treated with etoposide (4μg/ml, Sigma), staurosporine (4μg/ml, Sigma) or a DMSO solvent control for 20 hrs. We also used simvastatin (2mM, Calbiochem) or an ethanol solvent control for 5 days [46]. After treatment, cells were subjected to western blotting (PARP) or Annexin-V viability staining. We used a 10nM concentration of 17β-hydroxy wortmannin (Cayman Chemical) to inhibit the PI3K in EBV wild type-infected primary B lymphocytes 14 days post-infection. Cells were treated for 17 hrs and the cell cycle profile was determined using a BrdU incorporation assay as described above.
Cells were harvested, washed once in 1xPBS, lysed in RIPA buffer supplemented with a protease inhibitor cocktail (1:1000, Sigma) and sonicated. 50μg of protein were separated on a 7.5% (PARP1) or 15% (p27, BHRF1, PTEN, Actin) SDS-polyacrylamide gel and electroblotted on a protran membrane (Amersham, 0.45 NC) by wet blotting. Incubation with primary and secondary antibodies was performed as described previously [5] using primary antibodies specific for PTEN (1:8000, Abcam), BHRF1 (1:100, kindly provided by J-Y. Chen [47]), p27 (1:1000, Santa-Cruz), PARP1 (1:1000, Cell Signaling) and Actin (1:10000, Dianova).
Total RNA was isolated using TRIzol (Life technologies, 1ml per 107 B-cells) according to the manufacturers protocol. polyA+ RNA was isolated by hybridization of the total TRIzol purified RNA to oligo-dT-coupled latex beads using the nucleotrap mRNA mini isolation kit (Machery-Nagel) according to the manufacturers instructions.
Northern blots of polyA+ RNA and total RNA were performed by separating 1.5μg or 7.5μg RNA, respectively, alongside with 5μl RNA marker (0.5-10kb RNA marker, Life Technologies) on a denaturing 1% agarose/1xMOPS gel containing 2.2M formaldehyde for 5 hrs at 100V. RNA was transferred to a Hybond-XL membrane (Amersham) by capillary blotting over night in 10xSSC and cross-linked to the membrane by baking for 2 hrs at 80°C. The blot was hybridized with 50ng of a [32P]-α-dCTP (Perkin Elmer) radiolabeled DNA probe specific for the 3’UTR of BHRF1 (coordinates 55389:55567 in the EBV reference genome V01555.2), the BHRF1 open reading frame (coordinates 54360:54853 (V01555.2)), the BHRF1 intron (coordinates 53953:54359 (V01555.2)), or the left origin of lytic replication (coordinates 53351:53756 (V01555.2)). Labeling was obtained by using the random primed DNA labeling kit (Roche). The blot was hybridized over night at 65°C in Church buffer, washed 4 times in 0.1%SDS/1xSSC and exposed to Hyperfilm-MP (Amersham) at -80°C as indicated in the figure legends. For microRNA northern blots, 20μg of total RNA per sample was separated on 15% Mini-Protean TBE-urea acrylamide gels (BioRad) at 80V for 2 hrs in 0.5xTBE (Ambion). RNA was transferred on a Hybond N+ membrane (Amersham) by semi-dry blotting for 2,5 hrs at 250mA and UV-crosslinked to the membrane (1200μJ). The blot was hybridized with 20pmol of an [32P]-γ-ATP (Perkin Elmer) labeled oligonucleotide (MWG Eurofins) complementary to the seed-mutated mature miR-BHRF1-3 (TGTGCTTACACACTCAACGTTA) or the seed-mutated mature miR-BHRF1-2* of the 2/2*DSM recombinant (GCAAACGGCTGCAACAACGTTT) at 37°C for 1h in ExpressHyb solution (ClonTech). Blots were washed twice at 37°C in 2xSSC/0.05%SDS, and twice in 0.1xSSC/0.1% SDS at room temperature and exposed to Hyperfilm-MP (Amersham) at -80°C for 7 days.
MicroRNAs were quantified using stem-loop RT qPCR [48] and miR-BHRF1 specific primers as described previously [49]. 110ng total RNA was reverse transcribed with the TaqMan miRNA reverse transcription kit (Applied Biosystems) using 12.5μM of each RT primer. Per sample, 10ng of template were mixed with 1.5μM forward primer, 0.7μM reverse primer, 0.2μM probe and TaqMan universal PCR mastermix (Applied Biosystems). The samples were incubated at 50°C for 2 min, 95°C for 10 min, followed by 40 cycles of 95°C for 15 sec and 56°C for 1 min using a StepOnePlus real-time PCR system (Applied Biosystems). All samples were measured in duplicate and the RNU48 TaqMan microRNA control assay (Applied Biosystems) was used as internal control for normalization of all samples.
To quantify different BHRF1 containing transcritps and the Wp promoter activity, 400ng total RNA was reverse transcribed using the AMV reverse transcriptase (Roche) with a mix of RT primers specific for GAPDH, combined with primers specific for the W2W1 exon junction or BHRF1 [11, 50]. qPCR was performed on 20ng of template using the TaqMan universal PCR mastermix (Applied Biosystems) and incubated at 50°C for 2 min, 95°C for 10 min, followed by 40 cycles of 95°C for 15 sec and 60°C for 1 min on a StepOnePlus real-time PCR system (Applied Biosystems). The activity of the Wp promoter and the expression of W2-BHRF1 spliced transcripts were quantified as described earlier [11, 50]. For the quantification of all BHRF1 transcripts, RNA was reverse transcribed using the universal BHRF1 RT primer [11], and amplified using 0.3μM forward primer (CCCTCTTAAT TACATTTGTG CCAGAT (54337:54362 (V01555.2)), 0.3μM reverse primer [11], and 0.2μM of the Fam-labeled probe (TAGAGCAAGA TGGCCTATTC AACAAGGGAG A (54367:54397 (V01555.2)). The VIC-labeled human GAPD (GAPDH) endogenous control (Applied Biosystems) was used as internal reference for normalization, All samples were measured in duplicate.
Oligonucleotide primers (MWG Eurofins) encoding part of the wild type 3’UTR of PTEN (coordinates 102288:102331 (NG_007466.2)) or a seed-matched mutant PTEN 3’UTR (coordinates 102288:102331 (NG_007466.2) in which the position 102311:102318 (NG_007466.2) was replaced for a stretch of eight adenines were annealed and introduced in the 3’UTR of the firefly luciferase reporter plasmid pGL4.5 (Promega), which had been modified to contain an EcoR1 and Xho1 cutting site 3’ of the luc2 open reading frame. Constructs were confirmed by sequencing.
HEK 293 cells were seeded at a density of 7*104 cells per well in a 24-well cluster plate. The following day, 210ng of the wild type PTEN 3’UTR firefly luciferase fusion or of the seed-matched mutant control plasmids, and 840ng of miR-BHRF1-3 (55198:55395 (V01555.2) in pcDNA3.1(+)) or of the empty vector control pcDNA3.1(+) (Invitrogen), and 210ng of a pRL-SV40 plasmid (Promega) encoding the renilla luciferase to control for differences in cell numbers and transfection efficiency were cotransfected using 3μl metafectene per μg of plasmid DNA. The activity of the firefly and renilla luciferase were determined 2 days after transfection using the dual-luciferase reporter assay system (Promega) according to the manufactures instructions and measured using a Fluoroskan Ascent FL luminometer (Thermo Scientific) in triplicate measurements for each sample.
GraphPad Prism 6 was used to conduct all statistical analysis. The error bars represent the standard deviation of the data sets. Statistical significance was determined using the student's t-test and all data that were derived from LCLs generated from the same blood donor were analyzed as paired samples.
|
10.1371/journal.pgen.1003510 | Filamin and Phospholipase C-ε Are Required for Calcium Signaling in the Caenorhabditis elegans Spermatheca | The Caenorhabditis elegans spermatheca is a myoepithelial tube that stores sperm and undergoes cycles of stretching and constriction as oocytes enter, are fertilized, and exit into the uterus. FLN-1/filamin, a stretch-sensitive structural and signaling scaffold, and PLC-1/phospholipase C-ε, an enzyme that generates the second messenger IP3, are required for embryos to exit normally after fertilization. Using GCaMP, a genetically encoded calcium indicator, we show that entry of an oocyte into the spermatheca initiates a distinctive series of IP3-dependent calcium oscillations that propagate across the tissue via gap junctions and lead to constriction of the spermatheca. PLC-1 is required for the calcium release mechanism triggered by oocyte entry, and FLN-1 is required for timely initiation of the calcium oscillations. INX-12, a gap junction subunit, coordinates propagation of the calcium transients across the spermatheca. Gain-of-function mutations in ITR-1/IP3R, an IP3-dependent calcium channel, and loss-of-function mutations in LFE-2, a negative regulator of IP3 signaling, increase calcium release and suppress the exit defect in filamin-deficient animals. We further demonstrate that a regulatory cassette consisting of MEL-11/myosin phosphatase and NMY-1/non-muscle myosin is required for coordinated contraction of the spermatheca. In summary, this study answers long-standing questions concerning calcium signaling dynamics in the C. elegans spermatheca and suggests FLN-1 is needed in response to oocyte entry to trigger calcium release and coordinated contraction of the spermathecal tissue.
| During organism development and normal physiological function cells sense, integrate, and respond to a variety of cues or signals including biochemical and mechanical stimuli. In this study we used Caenorhabditis elegans, a small transparent worm, to study filamin (FLN-1), a structural protein that may act as a molecular strain gauge. The C. elegans spermatheca is a contractile tube that is stretched during normal function, making it an ideal candidate for study of how cells respond to stretch. Oocytes are ovulated into the spermatheca, fertilized, and then pushed into the uterus by constriction of the spermatheca. The ability of the spermatheca to constrict depends on inositol 1,4,5-triphosphate (IP3), a signaling molecule produced by the enzyme phospholipase C (PLC-1) that triggers calcium release within cells. In animals with mutated FLN-1 or PLC-1 the spermathecal cells fail to constrict. Using genetic analysis and a calcium-sensitive fluorescent protein, we show that FLN-1 functions with PLC-1 to regulate IP3 production, calcium release, and contraction of the spermatheca. Filamin may function to sense stretch caused by entering oocytes and to trigger constriction. These findings establish a link between filamin and calcium signaling that may apply to similar signaling pathways in other systems.
| Mechanotransduction, the conversion of physical forces into biochemical signals, is a critical component of cell signaling [1], [2]. Force sensation is essential during organism development and guides cell migration, differentiation, and morphogenesis [3], [4]. Mechanotransduction is essential for normal physiological functioning of the cardiovascular, musculoskeletal, and digestive systems; for example, blood vessel diameter decreases in response to pressure increases to maintain consistent blood volume flow [5], [6]. Cytoskeletal proteins are interconnected in a dynamic, cell-wide structure, and are optimally positioned to sense and transduce changing mechanical parameters [7]. In addition to acting as primary mechanosensors, cytoskeletal proteins are required to anchor and organize stretch-activated membrane ion channels [1].
Filamins are large cytoskeletal scaffolding proteins that consist of an N-terminal actin-binding domain (ABD), followed by a species- and isoform-dependent number of immunoglobulin-like repeats (IgFLN) [8], [9]. Human filamins contain 24 IgFLN domains, and mutations are associated with a broad spectrum of human diseases, including periventricular heterotopia (FLNA), boomerang dysplasia (FLNB), and various myopathies (FLNC) [10]–[13]. Filamin knockout mice show a similar spectrum of phenotypes [14]–[16]. The Drosophila filamin cheerio is composed of 20 IgFLN domains and is important for formation and function of ring canals and follicle cell motility during oogenesis [17]–[19].
Filamins organize the actin cytoskeleton into an elastic three-dimensional network that is responsible for maintaining cell structure and resisting mechanical forces [20], [21]. Experimental evidence suggests that filamin may act as a force-sensitive molecular scaffold [8], [9], [22]. Filamin contains cryptic binding sites that are obscured by adjacent domains [23], [24]. Atomic force microscopy data reveal that filamin molecules unfold when stretched and refold when tension is relieved [25]. Stretching of the filamin rod domain exposes binding sites for integrins and other proteins and leads to strengthening of focal adhesions [23], [26], and a force-dependent filamin-integrin interaction has been observed in cell culture [3]. In addition to acting as a direct mechanosensor, filamin also acts to connect stretch-activated ion channels, such as polycystins, with the cytoskeleton [6].
Our previous work focused on the initial characterization of a well-conserved C. elegans filamin ortholog FLN-1 [27], [28]. FLN-1 consists of an N-terminal ABD followed by 20 IgFLN domains, and is expressed in the spermatheca and uterus, where it colocalizes with F-actin [27], [28]. The allele fln-1(tm545) deletes a portion of the ABD, and a frameshift disrupts translation of all 20 of the IgFLN domains. Therefore, tm545 is a likely strong hypomorph or a null allele [27]. The most striking phenotype of the fln-1(tm545) animals is the failure of fertilized embryos to exit from the spermatheca [27]. The spermatheca, as part of the C. elegans gonad, serves as the site of sperm storage and fertilization [29]. It consists of three distinct regions: the distal constriction, a central accordion-like bag, and the spermatheca-uterine (sp-ut) valve [30]–[33]. The spermatheca, along with a large portion of the oviduct, is enveloped by a single layer of myoepithelial cells [33]. As oocytes are ovulated over the lifetime of the animal, the spermatheca undergoes numerous cycles of stretching, constriction, and relaxation, making it an ideal system to study the role of FLN-1 in cell response to stretch and coordination of cell signaling in an intact tissue.
During each ovulatory cycle, the most proximal oocyte is ovulated into the spermatheca, fertilized, and then released into the uterus [29]. Oocyte entry depends on complex signaling between the oocytes, the sperm, and the sheath cells. LIN-3/EGF is secreted by the proximal oocyte and acts on the proximal sheath cells [34]. In the sheath, LET-23/EGF receptor is predicted to activate PLC-3/phospholipase C-γ, which generates inositol 1,4,5-triphosphate (IP3) and diacylglycerol (DAG) [34], [35]. The IP3 signal is negatively regulated by IPP-5/inositol 5′-phosphatase and LFE-2/IP3 3′-kinase [34], [36]. ITR-1/IP3 receptor releases calcium from the endoplasmic reticulum when stimulated by IP3 [34]. Calcium likely controls the sheath cell myosin contraction by regulating troponin and tropomyosin [37]. The contraction of the proximal sheath cells propels the oocyte into the spermatheca, where fertilization immediately occurs [29].
After fertilization, directional constriction of the spermatheca propels the embryo into the uterus [29]. The molecular mechanism responsible for initiating and regulating spermathecal constriction is poorly understood, but PLC-1/phospholipase C-ε [38] is required for this process. Loss of PLC-1 results in trapping of embryos in the spermatheca [38]. PLC-1, like PLC-3, generates IP3 and DAG from PIP2; however, PLC-ε enzymes are regulated by small GTPases, while PLC-γ enzymes are regulated by receptor tyrosine kinases [39]–[41]. The specific usage of PLC-3 and PLC-1 in the sheath and the spermatheca suggests that the pathways may be differentially regulated. An increase in cytosolic IP3 likely activates ITR-1/IP3 receptor (IP3R), a tetrameric complex in the endoplasmic reticulum membrane [34], [42]. IP3 binding to IP3R is insufficient for full activation, which requires concomitant calcium binding [42], [43]. Activation of one IP3R in turn activates neighboring IP3Rs by elevating the local calcium concentration [42]. Although intermediate calcium concentration stimulates IP3R, high calcium concentration has an inhibitory effect [42]. This biphasic response of IP3R to calcium can create regenerating calcium waves in the presence of constant IP3 levels [42].
We propose that the release of calcium by the IP3R in the spermathecal cells induces constriction of the spermatheca by activating myosin contraction. In contrast to the sheath cells, the spermatheca does not appear to express a muscle myosin [37]. Smooth muscle regulatory components LET-502/Rho-activated kinase(ROCK) and MEL-11/myosin light chain phosphatase subunit are required for spermathecal function [44], suggesting that the spermathecal myosin belongs to the non-muscle myosin class [45], [46]. NMY-1/non-muscle myosin II is expressed in the spermatheca [39], and here we show NMY-1 is required for spermathecal constriction.
In this study we use GCaMP, a genetically-encoded calcium indicator, to show that oocyte entry stretches the spermathecal cells and triggers coordinated pulses of calcium transients. The calcium transients originate in the distal constriction and appear to propagate proximally across the spermathecal bag. FLN-1 and PLC-1 are required to trigger calcium signaling, and ITR-1 is required downstream of PLC-1 to produce the calcium oscillations. The signal is coordinated across the tissue via gap junctions to produce a directional wave. The directional wave of calcium results in contraction of the actomyosin network and expulsion of the embryo into the uterus. Given the modular protein structure, sub-cellular localization, genetic interaction data, and known mechanosensory functions of filamin, we postulate that FLN-1 is required to convert physical information about the presence of the oocyte into a calcium signal that controls the directional constriction of the spermatheca.
Previous studies have revealed an important role for phosphatidylinositol signaling during ovulation and spermathecal transit in C. elegans, and demonstrated that PLC-1/phospholipase C-ε is required for spermathecal transit [34], [35], [38]. PLC-1 generates IP3, which is thought to trigger calcium release from the endoplasmic reticulum via the IP3 receptor ITR-1 [34], [38]. We have shown previously that fln-1(tm545) and plc-1(rx1) single and double mutants show a very similar phenotype, in which embryos are retained in the spermatheca [27]. Because PLC-1 and FLN-1 are both required in the spermatheca for transit of fertilized oocytes, we explored the possibility that FLN-1 functions in the IP3 signaling pathway to regulate spermathecal function.
Strong itr-1 loss-of-function alleles result in ovulation entry defects, complicating observation of spermathecal transit. To circumvent this problem, we used itr-1 gain-of-function alleles [34], [35]. The itr-1 gain-of-function alleles are located in or near the IP3-binding domain of ITR-1 and are thought to increase the affinity of ITR-1 for IP3 [34], [47], [48]. Using brood size assays, we tested five putative itr-1 gain-of-function alleles for suppression of the fln-1(tm545) brood size defect [27]. itr-1 alleles sy327gf, sy328gf, and sy290gf suppressed the brood size defect (Figure 1), while sy291gf and sy331gf did not (unpublished data).
We chose to focus on the itr-1(sy290gf) allele for subsequent experiments because it affects a well-characterized residue critical for IP3 binding [49]. Biochemical characterization suggests that the sy290 recombinant protein has a two-fold increase in IP3 binding affinity [48]. The itr-1(sy290gf) animals do not exhibit overt ovulation or spermathecal transit defects; however, the sheath cell contractions are more frequent in these animals and they show a reduced brood size (173±17 SD, n = 6) compared to wildtype animals (301±26 SD, n = 14) [35]. To control for possible effects of the marker phenotypes we used itr-1(sy290gf) strains marked with dpy-20(e1282) (dumpy) or unc-24(e138) (kinker). No brood size differences were observed between the differentially marked strains. Similarly, the brood size of fln-1(tm545) unc-24(e138) (20±4 SD, n = 12) animals is not significantly (p = 0.08, Student's t-test) different from fln-1(tm545) animals. fln-1(tm545) itr-1(sy290) unc-24(e138) and fln-1(tm545) itr-1(sy290) dpy-20(e1282) animals have a significantly (p<0.0001, Student's t-test) higher average brood size (46±22 SD, n = 18, and 46±12 SD, n = 15, respectively) than fln-1(tm545) animals (Figure 1).
LFE-2 negatively regulates IP3 signaling by phosphorylating IP3 into inositol 1,3,4,5-tetrakisphosphate (IP4) [34]. Therefore, lfe-2(sy326), a loss-of-function allele, is predicted to have longer duration or intensity of IP3 signals [34]. Like itr-1(sy290gf), lfe-2(sy326) does not result in obvious ovulation or spermathecal transit defects, but does exhibit a reduced brood size (174±90 SD, n = 6) compared to wildtype animals (301±26 SD, n = 14). We speculated that increased IP3 levels in the lfe-2(sy326) background would also suppress the fln-1(tm545) spermathecal transit defect. We found that, indeed, fln-1(tm545); lfe-2(sy326lf) animals have a significantly (p<0.0001, Student's t-test) increased average brood size (43±27, n = 32) compared to fln-1(tm545) animals (Figure 1).
Because we observed a genetic interaction between fln-1 and components of the phosphatidylinositol signaling pathway, we next examined whether calcium signaling plays a role during spermathecal transit. It has been hypothesized that IP3R-regulated calcium release results in sheath cell contractions, however, IP3R-regulated calcium release has not been observed directly in the sheath or the spermatheca. To monitor calcium levels during spermathecal transit we used GCaMPv3 (GFP-Calmodulin-M13 Peptide, version 3), a genetically-encoded calcium indicator [50]. GCaMP has been previously used in C. elegans neurons and hypodermal cells to image calcium transients [50], [51].
We created transgenic nematodes expressing GCaMP under the control of the fln-1 promoter (xbIs1101[fln-1p::GCaMP]), imaged immobilized animals using wide-field epifluorescence with standardized acquisition parameters, and quantified the GCaMP signal by calculating mean pixel intensity (pixel intensity/area) and normalizing to the initial fluorescence (Figure 2A). Importantly, animals expressing GCaMP exhibit wildtype oocyte entry, ovulation transit times, and normal brood sizes (282±51 SD, n = 6; compare Video S1 and Video S2). Animals expressing regular calcium-insensitive fln-1p::GFP were used as controls to determine whether spermathecal shape changes or photobleaching would affect the GCaMP signal (Figure 2C, Video S2). No alteration in GFP fluorescence signal due to spermathecal shape changes was observed, and photobleaching was not detected over many hours of imaging (n = 6) (Figure 2C, Video S2). As an additional control for possible effects of changing cell shapes on measured intensities, we expressed tdTomato and GCaMP in the spermatheca and performed ratiometric imaging. Because the tdTomato signal remained constant throughout the duration of imaging, normalization of GCaMP to tdTomato signal did not result in any significant differences in signal compared to non-ratiometric imaging (n = 3) (Figure S1).
Time-lapse imaging of GCaMP reveals that oocyte entry into the spermatheca initiates a characteristic and reproducible sequence of calcium oscillations (n = 26) (Figure 2, Video S1). Out of necessity we focused our analysis of calcium signaling on the first ovulation; however, in wild type animals, generally similar calcium transients are observed in subsequent ovulations (Figure S2). Oocyte entry into the spermatheca consistently triggers a single pulse of calcium in the sp-ut valve (Figure 2A′, Video S1). The single pulse of calcium in the sp-ut valve may serve to constrict the valve to prevent premature exit of the oocyte. Neither fertilization nor egg shell formation are required to initiate spermathecal calcium signaling (Figure S3). Following complete entry of the oocyte into the spermatheca, the calcium transients increase in intensity as the oocyte progresses through the spermatheca (Figure 2A and B, Video S1). Embryo exit is concomitant with the strongest calcium pulses, suggesting that the calcium pulses trigger spermathecal constriction (Figure 2A). The final pulse of calcium coincides with constriction of the sp-ut valve following embryo exit (Figure 2A′).
Quantitative analysis of the time-lapse image sequences shows that the calcium transients appear to initiate in the distal spermatheca and propagate proximally (Figure 2A′–2B′). To determine if the calcium transients are directional we measured the fluorescence intensity in the distal and proximal spermatheca (Figure 2B). Calcium pulses are first detected in the distal spermatheca, and then in the proximal spermatheca several seconds later (Figure 2A′–2B′). The calcium transients occur in the direction of oocyte movement, suggesting that directional calcium pulses may control spermathecal constriction. The distal spermathecal cells may act as a pacemaker to trigger and synchronize calcium release in other spermathecal cells.
We predicted that the observed distal to proximal spread of the calcium signal would require cell-cell communication and tissue level coordination. To test this idea, we used RNAi to sequentially deplete the 25 gap junction subunits [52], [53], and determined that loss of the innexin INX-12 results in spermathecal transit defects. In inx-12(RNAi) animals, oocytes enter the spermatheca normally, but variably change direction several times before returning into the ovary or proceeding into the uterus (Figure 3, Video S3). Calcium imaging of the inx-12(RNAi) (n = 4) animals revealed that each spermathecal cell is capable of producing calcium pulses, but that the resulting calcium waves are asynchronous and non-directional (Figure 3A′, Video S3). The random calcium pulses likely result in the observed uncoordinated spermathecal constriction. These results suggest that a small molecule, such as calcium or IP3, propagates through the spermatheca via gap junctions to produce synchronous and directional calcium transients.
To investigate whether FLN-1 is required for normal calcium signaling during spermathecal transit, we introduced the fln-1p::GCaMP transgene into fln-1(tm545) animals. Although the initial entry pulse of calcium within the sp-ut valve occurs normally (Figure 4A and 4A′, Video S4), filamin-deficient animals fail to initiate calcium transients in the spermatheca itself, with few calcium signals observed during the time normally required for oocyte transit (n = 19) (Figure 4A, Video S4). Abnormal and highly variable transients are observed in fln-1(tm545) animals approximately 15 minutes after oocyte entry—well after the time a wildtype zygote would have exited (Figure 4A). These delayed calcium pulses fail to produce significant constriction of the spermatheca (Figure 4A′, Video S4), suggesting that the contractile mechanism may be compromised in filamin-deficient animals. These data suggest that filamin is required for timely initiation of calcium signaling and for the contractile mechanism, but not for the calcium release mechanism per se.
Homology and genetic interaction data suggest that hydrolysis of PIP2 by PLC-1 generates IP3 in the spermatheca, triggering intracellular calcium release via the IP3R [38], [54]. Consistent with this idea, plc-1(rx1) animals fail to produce calcium signals following oocyte entry into the spermatheca (n = 5) (Figure 4B, Video S5). plc-1(rx1) animals also do not produce the initial entry pulse of calcium in the sp-ut valve. Likewise, a temperature-sensitive reduction-of-function allele of itr-1(sa73ts) results in abnormal calcium signaling at the semi-permissive temperature of 20°C [55]–[57]. Phenotypes observed range from mild perturbations (Figure 5A) to grossly abnormal calcium transients (n = 6) (Figure 5B). The grossly abnormal calcium signaling is of lower intensity, and results in trapping of embryos within the spermatheca. Additionally, itr-1(sa73ts) animals do not produce the initial pulse of calcium in the sp-ut valve during oocyte entry, the timing between pulses is longer, and the calcium release is restricted to the distal spermatheca (Figure 5). These results suggest that the observed calcium transients in the spermatheca require IP3-regulated release of calcium from internal stores, and are consistent with previous findings in the C. elegans intestine [55], [56].
Unlike fln-1(tm545) single mutant animals, which eventually initiate abnormal calcium pulses, fln-1(tm545); plc-1(rx1) double mutant animals behave like plc-1(rx1) single mutants and fail to produce any calcium transients (n = 3) (Figure 4C, Video S6). This suggests that the delayed calcium pulses seen in fln-1(tm545) animals are generated via the canonical phosphatidylinositol signaling pathway. Although FLN-1 is required for timely initiation of calcium pulses upon oocyte entry, it appears that calcium signaling can eventually be activated by a parallel, filamin-independent pathway.
Because gain-of-function mutations that sensitize ITR-1 to IP3 suppress the fln-1(tm545) brood size defects (Figure 1), we next determined the effect of these mutations on calcium signaling in the spermatheca. We speculated that increased sensitivity of ITR-1 to IP3 would trigger increased calcium release. itr-1(sy290gf) only has a moderate effect on GCaMP intensity in the wildtype background (n = 5) (Figure S4A), which is consistent with the lack of a strong phenotype. We do not detect overt changes in itr-1(sy290gf) calcium dynamics, such as increased propagation speed nor increased frequency of calcium release; however, it is possible that there are higher frequency changes not captured by our imaging parameters. Surprisingly, fln-1(tm545) itr-1(sy290gf) double mutants have markedly increased calcium signaling immediately following embryo entry (n = 3) (Figure 6A, Video S7), and display a novel partial exit phenotype with the embryo held in place by a partially closed sp-ut valve (Figure 6A′). Sp-ut valve constriction around the zygote during eggshell formation results in bow tie-shaped embryos (Video S7). lfe-2(sy326) animals, like itr-1(sy290gf), have marginally increased intensity of calcium signaling in the wildtype background (n = 4) (Figure S4B). However, the calcium transients in lfe-2(sy326) animals are variable from animal to animal, which is consistent with our brood size data (Figure 1), and may reflect incomplete penetrance of the sy326 allele. Similar to the results with itr-1(sy290gf), fln-1(tm545); lfe-2(sy326) animals exhibit increased calcium signaling compared to fln-1(tm545) alone (n = 3) (Figure 6B, Video S8), and partial exit of embryos from the spermatheca (Figure 6B′, Video S8). Importantly, these observations indicate that the fln-1(tm545) spermatheca and valve may be structurally capable of constriction if sufficient calcium is present.
Cell contractility is generated by the actomyosin cytoskeleton, which consists of myosin, myosin regulatory proteins, and F-actin. Two non-muscle myosin regulatory proteins, Rho-activated kinase LET-502 and a subunit of myosin light chain phosphatase MEL-11, are known to be required for normal spermathecal function [44]. The non-muscle myosin NMY-1, redundantly expressed with NMY-2 during embryonic elongation, is strongly expressed in the spermatheca [45], [46]. NMY-1, MEL-11, and LET-502 have been extensively studied in the context of embryonic elongation where they are required to fine-tune contractility of the hypodermal cells [44], [45], [58]–[61]. We speculated that this contractile module also functions in the spermatheca and is responsible for constriction of the spermatheca.
Similar to the phenotype seen in fln-1(tm545) animals, depletion of nmy-1 by RNAi results in a poorly contractile spermatheca and an sp-ut valve that fails to constrict and completely expel the embryo (Figure 7). Embryos are pushed out of nmy-1(RNAi) spermathecae through a relaxed sp-ut valve due to back pressure from newly ovulated oocytes. In contrast, in mel-11(sb56) animals, oocytes fail to enter into a hyper-constricted spermatheca [44]. We speculated that mel-11(RNAi) might produce a weaker phenotype than the sb56 allele. Indeed, mel-11(RNAi) animals are able to ovulate, allowing observation of the spermathecal transit process (Video S9). Depletion of mel-11 in the spermatheca results in hyper-constriction of the spermatheca around the zygote, rupture of the spermatheca, and escape of the zygote into the body cavity (Figure 8A, Video S9). The distal constriction and the sp-ut valve also appear to hyper-constrict, forcing the embryo in this unusual direction. The mel-11 spermathecal rupture phenotype is strongly suppressed in the fln-1 (Figure 8B) and plc-1 (Figure 8C) backgrounds, consistent with the idea that FLN-1 and PLC-1 are required for spermathecal contractility.
Our previous work described the C. elegans filamin orthologs, and established that FLN-1 is required for function of the spermatheca, a smooth muscle-like tissue in the C. elegans gonad [27], [28]. Filamin-deficient spermathecae are unable to constrict, and as a result trap fertilized embryos [27]. In this study, we show that oocyte entry into the spermatheca triggers calcium oscillations that are likely instructive for spermathecal constriction. We find that FLN-1, an actin-binding protein and a known mechanosensitive scaffold, is required to trigger timely IP3-dependent calcium release in response to oocyte entry. We identify a gap junction subunit, INX-12, required for signal propagation across the spermatheca and the non-muscle myosin, NMY-1, required for spermathecal constriction. Our working hypothesis is that filamin is required in the spermatheca to transduce the physical presence of an oocyte into a biochemical signal, thereby triggering constriction of the spermatheca and expulsion of the embryo (Figure 9).
Constriction of the spermatheca in response to stretch is reminiscent of myogenic response seen in vascular smooth muscle cells, where increased intraluminal pressure in blood vessels results in vasoconstriction [1], [62]. The myogenic response is triggered via mechanically-sensitive ion channels that stimulate IP3-dependent calcium release. Filamin is required for cytoskeletal anchoring of an inhibitory polycystin channel subunit [6]. Filamin also interacts with other transmembrane channels, such as cystic fibrosis transmembrane regulator (CFTR) [63], Ca(v)1.4 subunit of a voltage-gated L-type channel [64], pacemaker channel HCN1 [65], G protein-coupled calcium-sensing receptor [66], and potassium channels Kir2.1 [67] and Kv4.2 [68]. Filamin therefore appears to be required for normal channel function, and may mechanically couple the channels to the cytoskeleton, as well as controlling their localization [9], [69].
Our calcium signaling and genetic interaction data are consistent with the possibility that FLN-1 might act as a top-level component in the spermatheca signal transduction pathway (Figure 9). However, the molecular details of how filamin might initiate this cascade are unknown. One possibility is that filamin may be a necessary adaptor between stretch-gated ion channels and the cytoskeleton. Strain on the cytoskeleton would then be communicated to the ion channels, allowing a brief influx of calcium ions which could activate PLC-1 either directly via its EF hand domain [70] or indirectly through a calcium/calmodulin-activated protein. In addition to connecting the stretch-gated channels to the cytoskeleton, filamin might also scaffold downstream effectors required to sense or amplify ion channel opening.
Another possibility is that filamin is a direct mechanosensor, and that stretch of the spermathecal cells by oocyte entry directly stretches the filamin molecule, thereby revealing cryptic binding sites. Evidence from biophysical experiments with purified components [22], [24], and study of the mechanosensory role of filamin in the context of focal adhesions in cultured cells support this idea [9], [71]–[73]. Stretch-activated binding sites could localize a RhoGEF, such as Trio [74], to the cortex and increase the level of Rho-GTP, which has been shown to activate PLC-ε [75], [76]. Because PLC-ε also contains a GEF domain, activation of PLC-1 may lead to prolonged activation and increased levels of Rho-GTP in addition to IP3 and DAG [77]–[79]. Activation of ROCK and calcium release would both act to promote contraction. Other downstream pathways could also be activated by DAG and calcium, such as Protein Kinase C (PKC) [80].
Filamin may perform a partially separable signaling and structural role in the spermatheca. We have shown previously that loss of filamin results in a progressive disorganization of filamentous actin in the spermatheca (Figure 9) [27]. F-actin organization is relatively normal initially; however, filamin is required to maintain cytoskeletal structure as the spermatheca is repeatedly stretched by incoming oocytes [27]. Because our calcium measurements are made during the first ovulation, before the actin cytoskeleton becomes grossly abnormal, and because calcium signaling can be strongly disrupted, for example, by loss of PLC-1, in spermathecae with intact actin cytoskeletons, we suspect that the calcium signaling defects in filamin-deficient animals are not simply a consequence of cytoskeletal defects, but we cannot entirely exclude this possibility.
While our results suggest FLN-1 is needed to trigger normal calcium release in the spermatheca, FLN-1 may be acting in parallel with PLC-1 to activate calcium release via ITR-1. We show itr-1(sy290gf) and lfe-2(sy326) ameliorate the effects of fln-1(tm545), suggesting that they function downstream of fln-1. Consistent with this result, over-expression of LFE-2 under a heat-shock promoter results in an exit defect similar to fln-1(tm545) [34]. The brood size defect of plc-1(rx1) animals is not suppressed by itr-1(sy290gf) nor lfe-2(sy326), presumably due to complete absence of IP3 [38], [54]. This suggests that fln-1(tm545) animals possess sufficient IP3 to activate the sensitized ITR-1 receptor. Therefore a key role of FLN-1 might be to regulate some aspect of ITR-1 response to IP3, ultimately resulting in calcium release and spermathecal constriction.
Although IP3 appears to be needed to initiate calcium signaling, we do not know whether IP3 levels oscillate, or whether IP3 simply triggers the first calcium transient. Increased IP3 levels due to inactivation of LFE-2, an IP3 kinase, or a hypersensitive IP3 receptor do not cause abnormal calcium oscillations in the wildtype background, suggesting that precise control of IP3 level is not required. Once triggered, IP3R may be capable of generating self-sustaining calcium oscillations [42], [43], [81], [82]. IP3 binding to IP3R stimulates the release of calcium into the cytosol, which initially stimulates IP3R, but becomes inhibitory at high concentrations (Figure 9) [42], [43]. Endoplasmic reticulum Ca2+-ATPase pumps then remove calcium from the cytosol, returning calcium levels to the stimulatory range. This repeating cycle results in oscillating calcium levels. Sensitivity of the IP3R to IP3 and calcium may be modulated by accessory proteins [43], which may explain the increasing amplitude of calcium release in the spermatheca. With each round of calcium release, ITR-1 may become more activated, leading to larger pulses of calcium. The calcium oscillations may be terminated by an extrinsic signal or a threshold effect of calcium and IP3.
Our observations suggest that calcium pulses are initiated in the distal spermatheca and spread proximally. Although loss of the gap junction subunit INX-12 disrupts the synchronous transients, all spermathecal cells appear to be capable of producing stochastic calcium pulses cell-autonomously. Gap junctions generally permit passive diffusion of small solutes, such as calcium and IP3. Interestingly, the diffusion of calcium within a cell is limited to small domains by the buffering effects of calcium-binding proteins [82], [83], while the diffusion rate of IP3 is much greater, allowing it to act as a global messenger [84]. Given these diffusion rates it seems likely that IP3 is primarily responsible for synchronization of calcium signaling in the spermatheca.
We observed that strengthening and synchronization of the calcium signal coincides with a steady, coordinated contraction of the spermathecal tissue and expulsion of the fertilized embryo. Spermathecal contraction requires the non-muscle myosin NMY-1. Smooth muscle cell contraction is regulated by phosphorylation of the regulatory myosin light chain (rMLC) by MLC kinase (MLCK) [85]. The C. elegans genome lacks an obvious MLCK homolog, suggesting that rMLC phosphorylation is regulated by other kinases, such as ROCK (Figure 9) [86]. In C. elegans loss of LET-502/ROCK leads to loss of contractility of the spermatheca [44], [58], [87]. Contractility is negatively regulated by dephosphorylation of the rMLC by MLCP, and in C. elegans loss of MEL-11, an MLCP subunit, leads to hyper-constriction of the spermatheca. Understanding how this contractile module is regulated by calcium signals is an exciting future area of study.
In summary, in this study we demonstrate that oocyte entry leads to dynamic calcium signaling and coordinated contraction of the tissue, ultimately leading to expulsion of the fertilized embryo. We further demonstrate that FLN-1, PLC-1, ITR-1 and the gap junction component INX-12 are required for normal calcium signaling in the spermatheca. These results establish filamin as a regulator of calcium signaling, and suggest disruption of filamin may result in defective cell response to stretch. This is important because human filamin mutations result in severe myopathies, cardiovascular, and neurological conditions, but it is unclear how loss of filamin results in these diverse pathologies. Our study shows that filamin, in addition to being a cytoskeletal protein, may modulate calcium signaling pathways in spermathecal and other mechanically-sensitive cells. In addition to providing mechanistic insight into how the spermatheca functions, this study helps to establish the spermatheca as an in vivo model system for the study of how cells coordinate tissue-level responses to mechanical input.
C. elegans strains were cultured on NGM agar plates with OP50 Escherichia coli at 20°C. Nematode observations and manipulations were performed at 20°C unless otherwise noted. For a complete list of strains used in this study please see Table S1. Standard genetic techniques were used to manipulate C. elegans genotypes. Point mutations were tracked with marker alleles. unc-24(e138), a weak kinker allele, was used to follow the itr-1 gain-of-function alleles. dpy-20(e1282), a dumpy allele, was also used to follow itr-1(sy290gf) to control for any marker phenotypes. unc-57(ad592), another weak kinker allele was used to follow lfe-2(sy326). Deletion and insertion alleles were genotyped using polymerase chain reaction (PCR).
RNA interference was performed by feeding animals dsRNA-expressing HT115 DE3 E. coli as described [27], [88]. The RNAi bacteria were seeded onto NGM plates supplemented with carbenicillin and isopropyl β-D-1-thiogalactopyranoside (IPTG). Eggs were obtained from gravid hermaphrodites using alkaline hypochlorite solution and placed on the RNAi plates. RNAi targeting constructs for fln-1 [27], nmy-1, and plc-1 were constructed by PCR amplification of wildtype cDNA using engineered restriction sites, and subsequently cloned into pPD129.36 (Fire Vector Kit). Primer sequences used for plasmid construction are shown in Table S2. mel-11 RNAi targeting construct was isolated from an open reading frame RNAi library (Open Biosystems; Huntsville, AL, USA). Empty pPD129.36 vector was used as a negative control in RNAi experiments.
We used RNA interference to individually deplete the 25 predicted gap junction genes [52], [53] in xbIs1101[fln-1p::GCaMP] animals. RNAi experiments were performed as described above. RNAi targeting constructs were obtained from an open reading frame RNAi library (Open Biosystems; Huntsville, AL, USA) or constructed by PCR amplification of wildtype cDNA and subsequent cloning into pPD129.36 (Fire Vector Kit). Primer sequences are provided in Table S2. We used a low-magnification screen to identify animals with distended spermathecae or embryos present in the ovary as indicators of abnormal spermathecal function. We excluded genes that grossly affected animal development or gonad morphology. Primary screen hits were selected for detailed calcium imaging as described below.
We define brood size as the number of hatched progeny. The total number of hatchlings was determined by segregating L4 animals to individual, freshly seeded plates. Progeny were counted and aspirated beginning two days after the initial transfer, and continuing for two days after end of egg laying. Brood sizes are reported as the mean ± standard deviation. Two-tailed, unpaired t-tests were used to test for statistical significance between relevant genotypes. Statistical analyses were performed using GraphPad Prism 5.
GCaMP3 was obtained from Addgene plasmid 22692 [50]. PCR was used to amplify GCaMP with primers IK189 and IK190 (Table S2) containing engineered restriction endonuclease sites XbaI and XmaI. The XbaI-XmaI fragment was cloned downstream of the fln-1 promoter in pUN85 [27] to generate pUN107. pUN107 contains the fln-1 promoter, GCaMP, and the fln-1 3′ UTR. Transgenic animals were created by microinjecting a DNA solution containing 40 ng/µL of pUN107 and 100 ng/µL of pRF4 (rol-6 marker). Progeny displaying the roller phenotype and green fluorescence in the spermatheca were segregated to establish transgenic lines. A strain (UN1037) with low levels of GCaMP expression and moderate transmission frequency was integrated using UV irradiation to generate strain UN1101 xbIs1101[fln-1p::GCaMP] II. UN1101 was outcrossed ten times to create strain UN1108. Standard genetic crosses were used to introduce xbIs1101 into various genetic backgrounds.
|
10.1371/journal.ppat.1004788 | Hantaan Virus Infection Induces Both Th1 and ThGranzyme B+ Cell Immune Responses That Associated with Viral Control and Clinical Outcome in Humans | Hantaviruses infection causing severe emerging diseases with high mortality rates in humans has become public health concern globally. The potential roles of CD4+T cells in viral control have been extensively studied. However, the contribution of CD4+T cells to the host response against Hantaan virus (HTNV) infection remains unclear. Here, based on the T-cell epitopes mapped on HTNV glycoprotein, we studied the effects and characteristics of CD4+T-cell responses in determining the outcome of hemorrhagic fever with renal syndrome. A total of 79 novel 15-mer T-cell epitopes on the HTNV glycoprotein were identified, among which 20 peptides were dominant target epitopes. Importantly, we showed the presence of both effective Th1 responses with polyfunctional cytokine secretion and ThGranzyme B+ cell responses with cytotoxic mediators production against HTNV infection. The HTNV glycoprotein-specific CD4+T-cell responses inversely correlated with the plasma HTNV RNA load in patients. Individuals with milder disease outcomes showed broader epitopes targeted and stronger CD4+T-cell responses against HTNV glycoproteins compared with more severe patients. The CD4+T cells characterized by broader antigenic repertoire, stronger polyfunctional responses, better expansion capacity and highly differentiated effector memory phenotype(CD27-CD28-CCR7-CD45RA-CD127hi) would elicit greater defense against HTNV infection and lead to much milder outcome of the disease. The host defense mediated by CD4+T cells may through the inducing antiviral condition of the host cells and cytotoxic effect of ThGranzyme B+ cells. Thus, these findings highlight the efforts of CD4+T-cell immunity to HTNV control and provide crucial information to better understand the immune defense against HTNV infection.
| Hantaan virus (HTNV), the prototype of Hantavirus genus, is a rodent-borne pathogen that causes human hemorrhagic fever with renal syndrome, with mortality rate of approximately 15% in Asia. The efforts of our immune system to defend against HTNV are important for clearance of the infection. However, the interaction between CD4+T-cell immunity and HTNV infection in humans is not known. Based on the novel T-cell epitopes we defined on HTNV glycoprotein in Chinese Han population, we confirmed that HTNV glycoprotein could induce vigorous and extensive CD4+T-cell response in humans. For the first time, we showed that both Th1 and ThGranzyme B+ cell responses involved in defense against HTNV infection and inversely correlated with plasma viral load and disease outcome. Additionally, we found that CD4+T cells characterized by broader antigenic repertoire, polyfunctional cytokine secretion, stronger expansion and highly differentiated effector memory phenotype always lead to much milder outcome of the disease, maybe through inducing antiviral condition of host cells and cytotoxic effect of ThGranzyme B+ cells. Our results add weight to the contribution of CD4+T cells in disease control after HTNV infection in humans, which may greatly advance the understanding about how HTNV interact with their host organisms.
| During the past decade, hantaviruses, belonging to the Bunyaviridae family, have gained worldwide attention as widespread emerging zoonotic pathogens [1–2]. Two clinical conditions of human hantavirus infections have been recognized worldwide: 1) hemorrhagic fever with renal syndrome (HFRS) primarily reflecting infections with Hantaan virus (HTNV) in Asia, Dobrava and Puumala (PUUV) viruses in Europe, and Seoul virus worldwide [3–4] and 2) hantavirus pulmonary syndrome (HPS) primarily reflecting infections with Sin Nombre (SNV) and Andes (ANDV) viruses in North and South America, respectively [5]. Globally hantaviruses might cause as many as 200,000 cases of human disease per year, of which more than a half of the disease cases are fulminant HFRS [6]. A total of 100,868 cases were reported during 2005–2012 in mainland China, where HFRS cases, primarily reflecting infections with the prototype member HTNV strain, account for 90% of the total global cases, with a case-fatality rate as high as 15% [7–9]. Moreover, the recent outbreak of HPS in Yosemite National Park in California, USA, showed a higher case-fatality rate of approximately 37%, thereby raising the concerns of the World Health Organization [10]. Because of the high morbidity and mortality, poorly understood disease pathogenesis, and potential use of pathogenic hantaviruses as weapons for bioterrorism, the Biological Weapons Convention has classified these viruses as Category pathogens; therefore, better understanding the immune mechanism against HTNV infection is of priority for global public health and safety.
The antigenicity of hantaviruses largely depends upon two structure proteins, nucleocapsid protein (NP) and envelope glycoprotein. HTNV-NP has been demonstrated to be highly immunogenic and conservative, inducing vigorous cellular and humoral immune responses in humans [11]. HTNV glycoprotein, heterodimers of mature glycoprotein Gn and Gc, is not only responsible for receptor binding and membrane fusion, but also considered as the main source of neutralizing antibody production [12–13]. T-cell immunity is a critical factor for protection from virus infections in humans. The T-cell epitopes on NP of hantaviruses and associated immune responses have been well-characterized [14–17]. We have previously identified eight novel HTNV-NP cytotoxic T cell (CTL) epitopes with different HLA restrictions [18–19] and showed that HTNV-NP-specific T-cell responses might reduce the risk of progression to acute renal failure [18,20]. Indeed, hantavirus Gn/Gc-specific T-cell responses have also been observed. Several CD8+T-cell clones recognizing epitopes on Gn/Gc have been obtained from the blood of HFRS patients infected with PUUV, ANDV or SNV [14,21–23]. Gn specific long-lived effector memory T-cell responses might contribute to protective immunity in ANDV-infected patients [22]. Similarly, protective immunity elicited through infection with recombinant HTNV-Gn/Gc in murine models have also been demonstrated [24]. Therefore, hantavirus Gn/Gc could also serve as a potent immunogen that induces T-cell responses. However, the T-cell epitopes on HTNV-Gn/Gc have not been identified, and the specific responses to these immunogens remain largely unknown.
Recent studies have demonstrated that CD4+T cells are also essential for the effective clearance of viral infections [25–26]. Virus-specific CD4+T cells are important for initiating and maintaining immunity against most viruses through a variety of mechanisms. (1) The rapid production and secretion of cytokines from CD4+T cells during a viral infection, which could induce an antiviral state in the host, indirectly priming the CD8+T-cell response [27] and facilitating antigen-specific antibody production [28]. (2) Subsequently, CD4+T cells are required to maintain and modulate effective CTL responses and neutralizing antibody [29] and develop long-term memory B and CD8+T cells. (3) CD4+T cells can also mediate direct cytotoxicity [29–30] and recruit innate or antigen-specific effector cells to the site of viral replication [31]. Enhancement of antiviral immunity through cytotoxic CD4+T cells has been proved [32]. The cytotoxic CD4+CTLs may represent a distinct subset of effector cells defined by the absence of the master regulator ThPOK [33]. Despite the importance of CD4+T cells during virus infection, CD4+T-cell immunity towards hantaviruses remains limited. To date, little information regarding the CD4+T-cell epitopes on the Gn/Gc of hantaviruses has been reported. Moreover, the role of CD4+T cells in hantavirus clearance in humans remains unclear.
To investigate human CD4+T-cell immunity to HTNV, the T-cell epitopes on HTNV-Gn/Gc and the specific CD4+T-cell responses were evaluated in a large cohort of HFRS patients from Chinese Han population. Herein, we found that HTNV-Gn/Gc could induce vigorous CD4+T-cell responses characterized by a broad antigenic repertoire, enhanced expansion, polyfunctional cytokine secretion and functional phenotypes with both T helper 1 (Th1) and ThGranzyme B+ (ThGzmB+) cell effects. Importantly, HTNV-Gn/Gc-specific CD4+T-cell responses were obviously associated with viral control and clinical outcome. Thus, this study is the first to reveal the properties of a primary, protective CD4+T-cell immune response to HTNV infection in humans, providing a foundation for understanding host immune responses to HTNV infections.
To examine the specific T-cell responses to HTNV-Gn/Gc in HFRS patients, we first screened the T-cell epitopes on HTNV-Gn/Gc. The majority of the participants in this cohort displayed reactive T-cell responses against HTNV-Gn/Gc. Notably, 73.7% (70/95) of the HFRS individuals were responders recognizing at least 1 peptide pool. Among the 70 participants displaying positive responses to HTNV-Gn/Gc, a median of 4 (range 1–11) target peptide pools were detected in each HFRS individual, with a median spot magnitude of 609 (range, 95–4,911) spot-forming cells (SFC)/106 peripheral blood mononuclear cells (PBMCs) against total peptide pools. All peptide pools showed positive responses, among which four peptide pools (P5, P12, P21 and P25) were frequently detected in more than 30% of the subjects, and 7 peptide pools (P17 to P23) contained the most frequently identified HTNV-Gn/Gc reactive T-cell peptides eliciting the strongest responses (S1 Table). Next, we analyzed the positive responses at the single peptide level to identify the T-cell epitopes on HTNV-Gn/Gc. Overall, the single peptide-specific T-cell responses were widely distributed across the primary structure of HTNV-Gn/Gc. Approximately, 155 of the 281 peptides (55.2%) were recognized by at least one person, and among these, 20 peptides were frequently targeted as the immunodominant epitopes in more than 10% of HFRS individuals with diverse HLA backgrounds (Fig 1, Table 1). A median of 6 peptides (range, 1–25) were targeted in a single individual, and the strength of some individual epitope-specific responses reached 1,546 SFC/106 PBMCs. Moreover, to identify a more precise estimation of the epitope-specific T-cell responses, the distribution of ex vivo CD4+ and CD8+T-cell epitopes across the HTNV-Gn/Gc was defined in 25 positive response HFRS patients with sufficient PBMC samples (Fig 2). The detected responses to 28 15-mer peptides were entirely CD4+T-cell dependent, indicating that these peptides were CD4+T-cell epitopes on HTNV-Gc/Gc. Whereas, the CD8+T-cell depletion completely abrogated the interferon (IFN) -γ responses in another 21 15-mer HTNV-Gc/Gc epitopes, confirming CD8+T cells as the source of these epitope-specific responses, and these peptides might contain the 9 or 10-mer HTNV-Gc/Gc CTL epitopes. Moreover, we identified 30 additional peptides that could induce both CD4+ and CD8+T-cell responses against HTNV-Gc/Gc.
After the ex vivo assessment of the T-cell responses against the HTNV-Gn/Gc, we divided the HFRS patients into two groups according to disease severity and compared their HTNV-Gn/Gc-specific immune responses. Among the 90 patients examined, T-cell responses against HTNV-Gn/Gc were detected in 31 of the 35 (88.6%) patients in the mild/moderate group compared with 39 of the 55 (70.9%) in the severe/critical group (Fisher’s exact chi-square test, P = 0.068). Importantly, although the breadth and magnitude of the HTNV-Gn/Gc-specific T-cell responses varied considerably among HFRS individuals, a comparison of the two groups (31 mild/moderate and 39 severe/critical) revealed a total SFC count (the sum of all responses to peptide pools), as an index of the total reactivity against HTNV-Gn/Gc, ranging from 135 to 4,911/106 PBMCs (median 881) for subjects in the mild/moderate group, was significantly higher than that observed for patients in the severe/critical group, with values ranging from 95 to 2,370/106 PBMCs (median 500) (P = 0.0068) (Fig 3A). Although the broad distribution of responses was similar in both groups, higher numbers of recognized HTNV-Gn/Gc peptides were observed in the mild/moderate group (mean 10; range 1–25) compared with severe/critical patients, who showed a relatively narrower repertoire directed against a smaller number of HTNV-Gn/Gc epitopes (mean 5; range 1–18) (P = 0.0010) (Fig 3B). We next analyzed the correlation between magnitude and breadth of the response specific to HTNV-Gn/Gc and found there was a positive association between the number of epitopes recognized and the total SFC/106 PBMCs of the T-cell responses specific to HTNV-Gn/Gc epitope pools in HFRS patients (P<0.0001, r = 0.482) (Fig 3C). Moreover, when dividing the patients into four subgroups according to the total SFC count (/106 PBMCs) including SFC 0–500, SFC 501–1000, SFC 1001–2000 and SFC greater than 2000, the comparison of the recognized epitope number showed a similar tendency that more epitopes are recognized in subjects with mild/moderate disease than that in severe/critical patients in each subgroup, especially in the subgroup with SFC 0–500 (P = 0.007) (Fig 3D). Therefore, this higher reactivity, including broader and stronger T-cell responses, was associated with mild rather than severe disease outcomes of HFRS. Consistent with the total T-cell response pattern against HTNV-Gn/Gc, a comparison of the responses targeting CD4+ or CD8+T-cell epitopes in two groups of 25 HFRS patients (12 mild/moderate and 13 severe/critical) showed similar results. A significant quantitative difference in the magnitude of both CD4+ and CD8+T-cell responses against each HTNV-Gn/Gc peptide was observed between the two groups. The number of SFC/106 cells was higher in the mild/moderate group (median 106, range 39–1,758 for CD4+T-cell responses and median 210, range 70–2,490 for CD8+T-cell responses) than that in severe/critical group (median 88, range 39–253 for CD4+T-cell responses and median 134, range 40–738 for CD8+T-cell responses) (P = 0.027 and P = 0.032 for CD4+ and CD8+T-cell responses, respectively) (Fig 3E-3F). In addition, although the difference was not statistically significant, we observed a wider breadth of both CD4+ and CD8+T-cell epitope responses in the mild/moderate group compared with that in severe/critical patients. To avoid the overlaps between the moderate and severe patients in all above comparisons, we separated the two combined groups into four clinical types as mentioned in Materials and Methods. The comparisons between the mild and critical patients showed similar results with the data between the mild/moderate and severe/critical patients (S1 Fig). The difference in total SFC/106 PBMC of the recognized epitope pools between the two ends of the patients’ spectrum was more marked (P = 0.0012) than that between the two combined groups (P = 0.0068). Collectively, HTNV-Gn/Gc could elicit robust T-cell immunity with extended epitopic breadth of Gn/Gc specificity. However, in contrast to subjects with mild/moderate HFRS, the responses in patients with severe/critical severity showed narrower antigenic repertoire and much weaker responses against HTNV-Gn/Gc, indicating that HTNV-Gn/Gc-specific T-cell responses would be functional against the HTNV infection during HFRS.
Given the importance of specific T-cell responses against the HTNV-Gn/Gc in controlling HTNV infection, we further investigated the cytokine secretion in T cells from HFRS patients after HTNV-Gn/Gc stimulation. At the acute phase of HFRS, HTNV-Gn/Gc-specific CD4+T cells displayed increased cytokine production, predominantly characterized by the Th1 cytokine profile upon recognition of the peptides. IFN-γ was the primary cytokine produced (median 0.486%, range 0.019%-1.950% of total CD4+T cells). A high frequency of tumor necrosis factor (TNF)-α-producing cells was also observed (median 0.372%, range 0.011%-1.800% of total CD4+T cells). A few cells produced interleukin (IL)-2 (median 0.061%, range 0.013%-0.355% of total CD4+T cells), whereas virtually IL-4-producing cells could also be detected (median 1.011%, range 0.031%-3.870% of total CD4+T cells) (Figs 4A, S2, S3). Further analysis revealed dual-cytokine production in HTNV-Gn/Gc-specific CD4+T cells. Subsets of IFN-γ-producing CD4+T cells simultaneously secreted IL-2 (median 0.079% of total CD4+T cells) or TNF-α (median 0.723% of total CD4+T cells), indicating that specific CD4+T cells induced a range of polyfunctional responses (T cells making at least 2 cytokines simultaneously) against HTNV-Gn/Gc stimulation (Figs 4B, S2, S3). Next, we compared the cytokine secretion levels in patients with different HFRS severities. Overall, the functions of the specific CD4+T cells were almost impaired in severe/critical patients, while stronger CD4+T-cell responses were observed in patients with milder HFRS. HTNV-Gn/Gc peptides elicited substantially higher frequencies of single IFN-γ (median 0.746% of total CD4+T cells), IL-2 (median 0.172% of total CD4+T cells) and TNF-α (median 0.868% of total CD4+T cells) or dual-cytokine (median 0.209% for IFN-γ+IL-2+CD4+T cells and median 1.100% for IFN-γ+TNF-α+CD4+T cells) secretion from specific CD4+T cells in mild/ moderate HFRS patients. In contrast, patients with severe/critical HFRS showed less single cytokine IFN-γ (median 0.351% of total CD4+T cells, P<0.001), IL-2 (median 0.048% of total CD4+T cells, P = 0.023) and TNF-α (median 0.150% of total CD4+T cells, P = 0.005) or dual-cytokine (median 0.047% for IFN-γ+IL-2+CD4+T cells, P = 0.015 and median 0.169% for IFN-γ+TNF-α+CD4+T cells, P = 0.049) production. No significant differences in IL-4 production were detected between patients with different severities of infection (Fig 4C).
Notably, some CD4+T cells displayed adequate cytolytic capacity and appeared to be highly functional, secreting cytolytic effectors, including granzyme B (median 2.120%, range 0.275%-13.000% of total CD4+T cells) and perforin (median 0.899%, range 0.005%-5.560% of total CD4+T cells); upregulated CD107a expression (median 0.527%, range 0.098%-6.230% of total CD4+T cells) was observed after recognition of the HTNV-Gn/Gc peptides (Figs 4D, S4, S5). The simultaneous production of granzyme B and CD107a expression (median 0.608% of total CD4+T cells) was also observed in subsets of CD4+T cells (Figs 4D, S4, S5). This important finding is consistent with several previous studies, showing the cytolytic capacity of virus-specific CD4+T cells, defined as a new “ThGzmB+ cell” subset of CD4+T cells. Although the rapid production of IFN-γ or granules by T cells is important in defense against virus infections, there seems to be few double IFN-γ+granzyme B+ or IFN-γ+perforin+ cells were visualized in each CD4+T-cell population analyzed (Fig 4E). Similarly, obvious higher frequency of granzyme B or CD107a expression was observed in the CD4+T cells from mild/moderate patients (median 2.965% for granzyme B and 0.777% for CD107a of total CD4+T cells) compared with severe/critical patients (median 1.795% for granzyme B and 0.469% for CD107a of total CD4+T cells) (P = 0.035 for granzyme B and P = 0.015 for CD107a). We also observed a trend towards higher levels of perforin+CD4+T cells in the mild/moderate group although this difference was not significant (P = 0.353) (Fig 4F).
Next, we measured the relationships between the frequencies of IFN-γ+CTLs and each Th1 cytokine-secreting CD4+T-cell subset. As expected, the frequencies of IL-2+CD4+, IFN-γ+CD4+ or TNF-α+CD4+T-cell subsets were positively correlated with the frequency of IFN-γ+CTLs (P = 0.018, r = 0.523 for IL-2+CD4+T cells; P = 0.001, r = 0.673 for IFN-γ+CD4+T cells; P = 0.007, r = 0.581 for TNF-α+CD4+T cells) (Fig 5A). In addition, the frequency of the granzyme B+CD4+T cells was also positively associated with the frequency of IFN-γ+CTLs (P = 0.041, r = 0.460) (Fig 5A), suggesting that ThGzmB+ cells are potential cytolytic effector cell subsets against HTNV infection.
In a previous study, we showed that upon HTNV infection, the direct onset of viremia was observed and higher levels of HTNV RNA load in the plasma of the HFRS patients was obviously correlated with much more severe HFRS, suggesting that the plasma HTNV RNA load could be used as a prognostic marker and predictor of the disease severity in HFRS [34]. The further analysis revealed highly significant associations between the higher frequencies of IFN-γ+CD4+T cells or granzyme B+CD4+T cells and lower HTNV RNA load in the acute phase of the HFRS (P = 0.022, r = −0.510 for IFN-γ+CD4+T cells and P = 0.044, r = −0.456 for granzyme B+CD4+T cells) (Fig 5B), indicating that higher levels of activated CD4+T cells against HTNV-Gn/Gc during the initial phase of HFRS mediate the better control of HTNV viremia. In addition, because the increased capillary permeability of the acute renal injury and thrombocytopenia are hallmarks in the course of HFRS, the serum creatinine levels and the platelet counts in HFRS patients typically provide a better prediction of disease severity [35]. Notably, we observed inverse correlations between the frequency of IFN-γ or granzyme B-secreting CD4+T cells and the level of serum creatinine (P = 0.023, r = −0.505 for IFN-γ+CD4+T cells and P = 0.034, r = −0.477 for granzyme B+CD4+T cells) (Fig 5C), while positive correlations between the frequency of IFN-γ or granzyme B-secreting CD4+T cells and the number of platelets in HFRS patients (P = 0.036, r = 0.472 for IFN-γ+CD4+T cells and P<0.001, r = 0.742 for granzyme B+CD4+T cells) (Fig 5D).
To better assess the effect of viral control through HTNV-Gn/Gc-specific CD4+T cells in HFRS patients, we carefully examined the kinetics of the frequency of IFN-γ-secreting CD4+T cells, the plasma HTNV RNA load, and the serum creatinine levels during the course of HFRS in patients with different disease severities. Consistent with previous studies, a rapid decrease in the plasma viral load from the febrile to convalescence stage was observed in both mild/moderate and severe/critical patients. This trend was coincident with a reduction in the frequency of IFN-γ secreting CD4+T cells and the levels of serum creatinine (Fig 5E). However, the levels of these three parameters were distinct at the febrile stage in the two HFRS groups. In mild/moderate patients, the frequency of IFN-γ-secreting CD4+T cells was obviously higher than that in severe/critical patients, whereas both the plasma viral load and serum creatinine levels in mild/moderate patients were lower compared with those in severe/critical patients at the initial stage of HFRS (Fig 5E). This observation primarily reflects the potentially function of antiviral effectors of HTNV-Gn/Gc-specific CD4+T cells at the acute stage of the disease, which might be a key determinant of the level of virus replication and the extent of renal injury in HFRS. This finding was further highlighted in the analysis of the viral control of CD4+T cells against HTNV infection.
We next set out to determine whether the granzyme B-secreting CD4+T cells could kill the HTNV-Gn/Gc peptide pools-pulsed target cells. Firstly, ex vivo granzyme B enzyme-linked immunospot (ELISPOT) assay showed that the HTNV-Gn/Gc epitope peptide pools could stimulate CD4+T cells of HFRS patients to produce granzyme B with the average magnitude of 450 SFC/106 cells (range, 21–1,226) (S6 Fig, n = 11 for mild/moderate and severe/critical patients, respectively). The comparison between the two groups showed that the magnitude of granzyme B-producing CD4+T-cell response was stronger in mild/moderate patients than that in severe/critical individuals (median 712, range 215–1,226 SFC/106 cells for mild/moderate group and median 131, range 21–872 SFC/106 cells for severe/critical group, P = 0.0181) (Fig 6A). Second, the biological cytolytic function of CD4+T cells was determined by the cytotoxic assay, in which the CD4+T cells of HFRS patients could lyse HTNV-Gn/Gc peptides-pulsed autologous or MHC class Ⅱ partial matched B lymphoblastic cell lines (B-LCLs) with the average percentage of 22.63 (range, 10.92–45.93) at effector-to-target cell ratio of 200:1. Moreover, we also observed that the percentages of cytotoxicity in mild/moderate group seems to be higher than that in severe/critical group at effector-to-target cell ratio of 200:1(P = 0.0002), 100:1(P = 0.0002) and 50:1 (P = 0.0229) (Fig 6B). The cytotoxic capacity of each patient was showed in S7 Fig (n = 7 for mild/moderate and n = 9 for severe/critical patients). Collectively, these results suggested that ThGzmB+ cell subset might participate in the immune response against HTNV infection through the production of granzyme B and cell-mediated cytolytic function during the early stages of the disease, defining a specific CD4+T cell functional phenotype for HTNV control.
We next assessed whether the antivirus effect of CD4+T cells was dependent on the capacity of expansion. The CFSE staining assay showed that HTNV-Gn/Gc-specific CD4+T-cell expansion was readily detectable in HFRS individuals at the acute phase of the disease (representative in Fig 7A and others in S8 and S9 Figs). However, a significant difference in the CD4+T-cell expansion to HTNV-Gn/Gc between two different groups was observed. The percentage of CFSEdim CD4+T cells in mild/moderate patients was higher than that in severe/critical patients (Fig 7B, P = 0.0013), and the mean fluorescence intensity (MFI) of CFSEdim CD4+T cells was lower in milder patients than that in more severe HFRS individuals (Fig 7C, P = 0.0345). Meanwhile, the MFI of CFSEdim CD4+T cells was inversely associated with the percentage of CFSEdim CD4+T cells at acute phase of HFRS (Fig 7D, P = 0.0199, r = −0.5158). These data indicated that milder disease severity was associated with more vigorous HTNV-Gn/Gc-specific CD4+T-cell expansion, whereas an impaired expansion capacity of HTNV-Gn/Gc-specific CD4+T cells was shown in severe/critical patients. However, when stimulated with the polyclonal activator SBE as control, both CD4+ and CD8+ T cells of the HFRS patients presented proliferative activity with a large proportion of CFSEdim cells (representative in Fig 7A and others in S9 Fig), which was obviously higher than that stimulated with HTNV-Gn/Gc peptides (S10 Fig, P = 0.0049 for CD4+T cell and P = 0.0005 for CD8+T cell), suggesting that the impaired T-cell expansion in severe/critical HFRS patients may be refractory to the specific HTNV-Gn/Gc stimulation, but not susceptible to a systemic cell death after activation.
Because the effective cytokine-secreting CD4+T-cell response plays an important role in the control of HTNV viremia, we reasoned that the ex vivo CD4+T-cell expansion should also correlate with the decreased HTNV RNA load. During the acute phase of the disease, the percentage of CFSEdim CD4+T cells showed a significant inverse correlation with the maximum plasma viral load in each HFRS patient (Fig 7E; P = 0.0435, r = −0.4556), suggesting that the acquisition of proliferative potential and capacity might be an essential character for specific CD4+T cells in protection from HTNV infection. Previous studies have demonstrated a strong linkage between HTNV-specific CD8+T-cell expansion and the control of HTNV infection [18]. Thus, when we analyzed the relationship of the proliferation between HTNV-Gn/Gc-specific CD4+ and CD8+T cells, both the percentage and the MFI of the CFSEdim cells showed an expected positive correlations between CD4+ and CD8+T cells at the initial phase of infection in HFRS patients (Fig 7F-7G; P<0.0001, r = 0.8789 for MFI and P = 0.0002, r = 0.6051 for percentage). In fact, when we comparing the CD8+T cells for the difference of proliferation capacity in the two groups, we found that, similar with CD4+T cells, both the MFI and the percentage of CFSEdim CD8+T cells were higher in milder patients than that in more severe individuals (S11 Fig, P = 0.0087 for percentage and P = 0.0454 for MFI). Taken together, these data demonstrate a significant expansion of HTNV-Gn/Gc specific CD4+T cells during the acute phase of HFRS, associated with the control of HTNV viremia in patients showing a milder degree of disease severity.
To characterize the overall phenotype of HTNV-Gn/Gc-specific CD4+T cells, we first defined the activation extent of CD4+T cells. After stimulation with HTNV-Gn/Gc-overlapping peptide pools, the CD38 expression in circulating blood CD4+T cells was upregulated, peaking at the febrile stage of HFRS (MFI 182) (Fig 8A). The maximal activity of HTNV-Gn/Gc-specific CD4+T cells at the febrile stage might reflect a rapid and prominent CD4+T-cell response against HTNV infection after disease onset, prompting further examination of the phenotype characteristics of efficient CD4+T cells.
We next determined the memory phenotype of HTNV-Gn/Gc-specific CD4+T cells in HFRS patients. To this end, the expression of CCR7, CD45RA and CD127 in responding CD4+T cells was evaluated in patients with different severities. Effector memory (TEM) CD4+T cells (CCR7–CD45RA–) (median 49.2% of CD4+T cells) are typically observed in patients with mild/moderate HFRS, while the individuals with severe/critical disease were enriched for central memory (TCM) CD4+T cells (CCR7+CD45RA–) (median 50.2% of CD4+T cells) (Fig 8B). There was no observable difference between the two groups in the levels of naïve (CCR7+CD45RA+) or transitional (CCR7–CD45RA+) T cell populations. An analysis of the formulation of the memory phenotype in responsive IFN-γ or granzyme B-secreting CD4+T cells for the two patient samples revealed CCR7–CD45RA–CD4+TEM cells as the predominant T cell subset (55.9% TEM in IFN-γ+CD4+T cells and 70.2% TEM in granzyme B+CD4+T cells) (Fig 8C). We also observed an abundant percentage of CD127high CD4+T cells in mild/moderate patients, and these cell numbers were higher than those detected in severe/critical HFRS individuals (Fig 8E). Thus, these findings suggested that HTNV-Gn/Gc-specific CD4+T cells exhibiting a CCR7–CD45RA–CD127high effector memory phenotype might contribute to the milder severity of HFRS.
Next, these cells were examined for the expression of markers associated with T cell differentiation and maturation. Fig 8D illustrates the CD4+ differentiation populations delineated according to the coexpression of CD27 and CD28 maturation markers in the peripheral blood of HFRS patients. Although various proportions of differentiated subsets were observed within the CD4+T-cell population in HFRS patients, the CD27–CD28—highly differentiation subset constituted the majority of the CD4+T-cell population in mild/moderate patients (median 48.30% of CD4+T cells), while these cells were only a minor component of the HTNV-Gn/Gc-specific CD4+T cells in severe/critical patients (median 6.94% of CD4+T cells, P = 0.004). In contrast, patients with severe/critical HFRS exhibited a CD27+CD28+ early differentiation phenotype (median 64.50% of CD4+T cells), and this effect was much higher than that in mild/moderate patients (median 7.06% of CD4+T cells, P = 0.004). Thus, we speculated that patients exhibiting an initial differentiated phenotype of HTNV-Gn/Gc-specific CD4+T cells during the acute phase of the disease would be exposed to higher levels of HTNV replication, resulting in more severe HFRS. In contrast, the CCR7—and CD28–CD27—phenotypes could represent fully differentiated TEM CD4+T cells in mild/moderate HFRS patients, potentially contributing to the effective control of HTNV infections.
Moreover, we analyzed the expression of inhibitory receptors on HTNV-Gn/Gc-specific CD4+T cells, potentially representing another factor involved in CD4+T cell function against HTNV infection. To this end, we assessed the expression of programmed cell death ligand 1 (PD-1) and CD57 in responding CD4+T cells during acute HTNV infection. A substantial fraction of CD4+T cells expressed both PD-1 and CD57 early after symptom onset in the HFRS patients. As shown in Fig 8E, patients with severe/critical HFRS exhibited a significant PD-1high CD57high phenotype (median 5.63% for PD-1 and 2.14% for CD57 in CD4+T cells) compared with the PD-1low CD57low phenotype observed in mild/moderate patients (median 1.36% for PD-1 and 0.44% for CD57 in CD4+T cells) (P = 0.004 for PD-1 and P = 0.032 for CD57), suggesting that the increased expression of inhibitory receptors might be associated with CD4+T-cell dysfunction or downregulation of the effector CD4+T-cell response against HTNV infection, leading to a more serious degree of HFRS.
The host immune response is crucial for the control of viral infection through orchestrated activities of different components of the immune system, among which, cellular immune response greatly contributes to the outcome of virus infectious diseases. Studies of the cellular immunological features of pathogenic HTNV infection provide insight into the understanding of HFRS pathogenesis in humans. In this study, we provided, for the first time, that HTNV could induce both Th1 and ThGzmB+ cell responses involving in the defense against HTNV infection; we showed that HTNV-Gn/Gc-specific CD4+T-cell immunity with broad reactivity, polyfunctionality, expansion capacity, and effector phenotype, inversely correlated with plasma viral load and the HFRS disease outcome; we further provided evidence that HTNV-Gn/Gc-specific CD4+T cells mediated host defense against HTNV infection maybe through the Th1 induced antiviral condition of the host cells and the cytotoxic effect of ThGzmB+ cells.
As only a few CTL epitopes have been defined on HTNV-Gn/Gc in HFRS patients, T-cell epitope-based preventive HTNV vaccine studies could only benefit from a focus on previously defined CTL epitopes on HTNV-NP. This limitation prompted a search for novel protective T-cell epitopes on HTNV. In the present study, we first performed the systematic identification of CD4+ and CD8+T-cell epitopes on HTNV-Gn/Gc. The immunodominant regions on these novel immune epitopes were subsequently analyzed and defined based on frequent observation in a number of HFRS individuals of diverse HLA backgrounds. Notably, immunodominance was precisely defined as the most frequently recognized epitope in a cohort of patients, with a strong magnitude of response relative to the total response. Although these experiments did not provide evidence that dominant HTNV-Gn/Gc-specific T cells kill or suppress virus-infected cells more efficiently than subdominant T cells, the immunodominant region and epitopes might merit special attention for the potential development of a novel vaccines that would generate or boost T-cell responses against HTNV in humans.
Previous studies have suggested that NP are major targets of HTNV-specific CTL recognition in humans [14–15,17,23,36–37]. The present ex vivo study verified that Gn/Gc, as another immunogenic protein of HTNV, could elicit specific CD4+ and CD8+ effector T-cell responses and provide obvious protection against HTNV infection. Importantly, HTNV-Gn/Gc-specific T-cell responses were not detected in 26.3% of HFRS patients. This result potentially reflects a response level below the threshold of assay detection or protection through other immune responses, such as HTNV-specific neutralizing antibody [13,38]. In the 73.7% responding patients, a considerable variation in HTNV-Gn/Gc-specific T-cell epitope recognition patterns was observed between different individuals, including the magnitude and breadth of the responses. Indeed, functional differences in specific T-cell responses for different HTNV-Gn/Gc epitopes were observed, particularly because patients with mild/moderate HFRS always showed stronger T-cell responses compared with severe/critical patients. Moreover, we also observed a general wider breadth of T-cell epitope responses in the mild/moderate group compared with that in severe/critical patients, although there is no statistical difference in the number of recognized epitopes between milder and more severe patients in some subgroups with different total SFC, which may be due to the small sample sizes after subdivision in each subgroup. The differential epitope recognition might reflect host genetic differences and exposure to region-specific or unrelated pathogens [39]. Therefore, we speculated that the successful induction of T-cell responses with high magnitude and increased breadth in response to HTNV-Gn/Gc-specific T-cell epitopes might be crucial for protection against HTNV infection, consistent with the fact that HTNV-NP-specific CTLs are important for controlling viremia and disease progression in HTNV infection [18,20]. Thus, the epitopes identified herein provide the foundation for an evaluation of the role of HTNV-Gn/Gc-specific T cells in protection from HTNV infections in humans.
The main goal of this study was to determine how HTNV-Gn/Gc-specific T-cell responses eliminate virus replication in HFRS patients. However, the lack of patient blood samples prevented a detailed analysis of the 8 to 10-mer CTL epitopes and associated responses to HTNV-Gn/Gc. Therefore, we focused on the potential mechanisms of CD4+T-cell responses in controlling HTNV infections. The data showed that HTNV-Gn/Gc-specific CD4+T cells primarily produce Th1 cytokines and specific cytotoxic mediators, suggesting that the high production of antiviral cytokines is a critical characteristic of effective T-cell immunity for the suppression of virus replication and elimination of virus-infected host cells. Moreover, HTNV-Gn/Gc-specific CD4+T cells also displayed a high degree of polyfunctionality, simultaneously secreting IFN-γ, TNF-α or IL-2. IFN-γ has been implicated in immune regulatory and direct antiviral activities [40]. The rapid production of IFN-γ and other cytokines are important components of the T cell response against viral infections [41]. Therefore, the induction these multifunctional Th1 populations might greatly enhance the anti-HTNV immunity, associated with superior protection efficacy [42–43]. Notably, the production of Th1 cytokines is directly correlated with the IFN-γ-secreting CD8+T-cell response against HTNV-Gn/Gc, strongly suggesting that there are robust cellular responses induced by acute HTNV infection. The activities of Th1 cytokine-producing CD4+T cells may provide further benefits for expansion or function of other effector cells. However, the solid helper function of CD4+T cells needs to be investigated in the future study [29]. In addition to exhibiting adequate Th1 activities, a proportion of CD4+T cells exhibit cytotoxicity in terms of granzyme B and perforin production and the upregulation of CD107a expression upon recognition of HTNV-Gn/Gc. It has been confirmed in several studies that some CD4+T cells with cytotoxic potential were specifically induced at the site of infection during virus infection and implicated in the killing of virus-infected cells [32–33,44–46], and the IL-2 signaling, low antigen dose, as well as OX40 and 4-1BB stimulation may promote the development of CD4+T cells with cytotoxic potential [47–49]. The earlier research on SNV has demonstrated that the establishment of CD4+CTL clone could lyse autologous target cells pulsed with nucleoprotein peptide [14]. Importantly, the cytolytic capacity of HTNV-Gn/Gc-specific granzyme B-producing CD4+T cells in killing the HTNV-Gn/Gc peptides pulsed autologous B-LCLs was also observed in our experiment, with the stronger cytotoxicity percentage in much milder patients, indicating that HTNV-Gn/Gc-specific effector CD4+T cells against HTNV infection in HFRS patients were comprised of polyfunctional Th1 cells and cytotoxic ThGzmB+ cells. Although few IFN-γ+granzyme B+ double positive cells was visualized in each CD4+T-cell population specific to HTNV-Gn/Gn, which was similar with the findings reported in the study of Leishmania [50], many other studies showed the virus-specific IFN-γ +CD4+T cells also expressed granzyme B or CD107a [51–52]. Although some researchers considered the cell-mediated cytotoxic CD4+T cells as a functional distinct “ThCTL” subset [32,53], it still can hardly consider that granzyme B is totally lost from the IFN-γ-secreting cells. In addition, the substantial protective effect of HTNV-Gn/Gc-specific CD4+T cells is also demonstrated through the inverse correlation between cytokine-producing CD4+T cells and the HTNV RNA load during the early phase of the disease or the apparently higher cytokine production in mild/moderate patients compared with severe/critical individuals. Furthermore, we observed CD4+T cell-mediated control of HTNV replication during early, but not late infection, as both IFN-γ secretion from CD4+T cells and peak HTNV viremia were observed during the febrile stage, followed by a gradual decrease throughout the follow-up period of the disease. Thus, the observed viremia kinetics likely reflects the putative direct antiviral effect of cytokine-producing CD4+T cells.
Another important observation of this study is the marked increase in CD4+T-cell activation and proliferation soon after HTNV infection in HFRS patients. First, the expansion of HTNV-Gn/Gc-specific CD4+T cells during early stage HFRS is highly associated with disease outcome. The lower percentage and higher MFI of the CFSEdim cells in both CD4+ and CD8+T cells always observed in the patients with more severe HFRS, indicating there may be deficient proliferative capacity of T cells in severe/critical patients. Second, the levels of proliferating HTNV-Gn/Gc-specific CD4+T cells in circulation are directly correlated with the capacity of HTNV-Gn/Gc-specific CD8+T-cell expansion and inversely correlated with plasma HTNV viremia of the patients, suggesting that the proliferation of specific CD4+T cells in response to HTNV-Gn/Gc exert direct antiviral effects during HFRS. Importantly, the finding that polyclonal antibody stimulation induced great proliferative ability of T cells was stronger than the poor capacity of expansion stimulated with HTNV-Gn/Gc peptides indicated that the decrease of T-cell proliferation from the more ill individuals are refractory to HTNV-Gn/Gc, but not a systematic deficit. We may partially find the reason for this antigen-specific decline in T-cell proliferation from the different expression of inhibitory receptors on T cells during HTNV infection.
Numerous studies have examined the role of PD-1, an ITIM-containing inhibitory receptor expressed on activated T cells, and found that PD-1 on the cell cycle could lead to inhibition of T-cell expansion [54–55]. In the present study, we demonstrated that HTNV infection might differentially affect the expression of PD-1 on CD4+T cells. The upregulation of PD-1 expression on HTNV-Gn/Gc-specific CD4+T cells and the dysfunctional and senescent phenotype (CD127loCD57hi) are more likely presented in patients with much severe HFRS. Notably, CD57 expression on T lymphocytes has been recognized as a marker of in vitro replicative senescence for measuring functional immune deficiency in patients with infectious diseases [56], and the upregulation of inhibitory receptors on T cells is an important mechanism of T-cell dysfunction during chronic viral infections [57]. Therefore, it is hypothesized that the higher PD-1 expressing CD4+T cells may account for the impairment of T-cell proliferative in more severe patients or lack of T-cell assistance during the earliest stages of infection, which determines whether CTL effectors develop into TEM cells conferring immune protection. However, whether the ligands for PD-1 are highly expressed in HTNV-infected cells need to be further studied.
Active lymphocytes in different subsets could express distinct panels of lymphocyte homing receptors, reflecting differences in homing potential [58]. The early immune activation typically stimulates the production of critical TEM cells and maintains the levels of T-cell differentiation and effector compartments [59]. Several studies have shown that naïve T cells and TCM express CCR7, a lymph node homing receptor, for homing to secondary lymphoid organs, in which these cells proliferate and rapidly differentiate into effector cells upon encountering corresponding virus antigen [60]. Whereas, the loss of CCR7 expression on TEM might generate a tissue-homing phenotype, leading to effector cell function against viruses in peripheral tissues [60–61]. Jang et al. recently showed that a predominant expansion of an activated HCV-specific CCR7—CTL could be observed in acute resolving HCV-infected patients, whereas HCV-specific CCR7+ CTLs were consistently observed in patients with persistent infections [62–63]. In particular, HFRS individuals with high plasma viral load during early stage might induce CD4+TCM cells to differentiate into TEM cells to overcome viral replication, leading to decreased viremia, reduced TCM cell numbers, and much milder disease outcomes. In contrast, a partial defects or low differentiation leading to the dominant CD4+TCM subpopulation exhibiting unsatisfactory control of the viral load upon the HTNV infection might be associated with much severe HFRS. As IFN-γ or granzyme B-producing CD4+T cells also showed TEM phenotypes, it is tempting to hypothesize that CD4+TEM cells lack CCR7 expression, which likely migrate to peripheral sites in the body and exert antiviral activity through the production of antiviral cytokines during the early phase of HTNV infection. Thus, the relative proportion of HTNV-Gn/Gc-specific CD4+T cells characterized by the ability to function as fully differentiated effectors (CD28–CD27–), acquisition of an effector tissue-homing membrane phenotype (CD45RA–CCR7–) and expression of high levels of IFN-γ or granzyme B, exhibit great effector performance and might be principally involved in determining the disease outcome of HFRS. These findings suggest that the CD4+TEM cell subset likely serve as major effector cells against HTNV infection.
Another intriguing finding is that the expression of CD127 is downregulated on CD4+T cells in severe or critical patients. CD127 is known highly expressed on T cells and is considered as a marker for memory precursor CD8+T cells during viral infection [64]. Similar with our finding, Shin et al. have reported that the percentage of hepatitis C virus-specific CD127+T cells remained low in chimpanzees with chronically evolving hepatitis, indicating that the early expression of CD127 on T cells could predict the outcome of acute infection [65]. Furthermore, CD127 is the α chain of interleukin 7 receptor (IL-7Rα). IL-7 has been identified as a major homeostatic cytokine for mature T cell and is produced at constitutive levels by stromal cells resident in various organs, as well as by thymic and intestinal epithelial cells and fibroblastic reticular cells at T cell zone of lymph nodes [66–67]. It has been demonstrated that CD127 expression is downregulated from the cell surface upon IL-7 stimulation [68]. For example, in HIV infection, systemic levels of IL-7 are increased, whereas the decreased expression of CD127 is demonstrable on both CD4+ and CD8+T cells [69–70]. Moreover, recent study on HIV infection also showed that both IL-1β and IL-6 could decrease RNA levels and T-cell surface expression of CD127, especially on CD4+ T cells [71]. Based on this finding, we can speculate that the lower expression of CD127 on the CD4+T cells may be caused by the higher levels of proinflammatory cytokines in much more severe HFRS patients, because that the significant higher levels of IL-6 and IL-8 have been observed in more severe patients than in milder HFRS cases [72]. However, whether the decreased expression of CD127 on CD4+T cells in more severe patients is induced by the more IL-7 secretion during HFRS needs to be investigated in our future study.
The presence of T-cell response against HTNV infection in HFRS patients but with different clinical outcomes reveals that there may be a relationship between the role of cellular immune response in controlling HTNV infection and the mechanisms of HTNV to inhibit this response. The host-virus relationship is a dynamic process in which the virus tries to decrease its visibility, whereas the host attempts to prevent and eradicate infection [73]. On the one hand, vigorous CD4+ and CD8+T-cell responses emerge in many acute virus infections. Similar with our findings, the HCV clearance has been observed and correlated with specific CD4+T-cells responses, and the early priming of the CD4+T-cell response is required for viral clearance [74], suggesting that the stronger immune responses were necessary for control of the infection better in the milder patients. On the other hand, virus infection also affects CD4+T-cell responses. Recent findings showed that HCV is able to impair the cellular immune response by developing escape mutations in T-cell epitope recognition sites and inducing specific CTL anergy and deletion [75]. Besides, costimulatory molecules modulation, apoptosis induction and chemokine regulation have also been confirmed as major mechanisms of HCV to inhibit immune control [75]. Based on these, our finding that the higher expression of inhibitory molecule PD-1 on CD4+T cells might be one of the mechanisms mediated by the HTNV to impair the immune responses, which may lead to dysfunction of specific CD4+T-cell response and much more severe disease. In fact, besides the functional CD4+T cells we detected in peripheral blood of HFRS patients, the CD4+T cells are also involved in localized immune responses in tissue [76]. The pathogen activated endothelium could direct localized immune cell adherence [77]. In PUUV-infected patients, the kidney biopsies showed interstitial infiltration of lymphocytes [78], and lung biopsies showed an increase in submucosal CD4+ T cells [79]. Studies of fatal HPS cases also revealed mononuclear cell infiltrates in pulmonary tissue consisting largely of CD4+ and CD8+T cells [80], indicating a local immune response in terms of activated T lymphocytes. Therefore, we could speculate that the much more effective T cells may be attracted into the local tissues in more severe patients with higher viral loads, leading to a decreased number of T cells in peripheral blood to fight against the virus infection, which may partial explain the correlation between the relative lower frequency of peripheral CD4+T cells and the deficit function of the CD4+T cells in severe or critical patients.
In summary, HTNV-Gn/Gc could induce the specific CD4+T-cell response characterized by a broader antigenic repertoire, stronger capacity of cytotoxicity and proliferation, higher frequency of cytokine secretion and fully differentiated effector memory cell phenotype, and would elicit greater defense against HTNV infection, likely through the induction of antiviral conditions in host cells and the cell-mediated cytotoxic effects of ThGzmB+ cells. These findings provide insights into the mechanism of HTNV-Gn/Gc-specific CD4+T cells in developing an efficient anti-HTNV response in HFRS patients, thereby increasing the current understanding of the relationship between HTNV and the host immune system and evoking strong interest for further studies of T-cell immunity in HTNV infection to guide the development of future clinical therapies. However, some data we presented here are limited to the correlative analysis. In future studies, we will try to sample the sites of HTNV infection specifically and use the CD4-depletion animal models to validate the indispensability of CD4+T-cell immunity in protection from HTNV infection. Furthermore, whether CD4+T cells facilitate B cell differentiation and antibody production, particularly neutralizing antibodies against HTNV infection should also be considered next.
Written informed consent was obtained from each HFRS patient or their guardians under a protocol approved by the Institutional Review Board of the Tangdu Hospital and the Fourth Military Medical University. The research involving human materials was also approved by the Ethical Review Board of the University, and the related information was used anonymously.
A total of 95 HFRS patients infected with HTNV were enrolled in the study at Department of Infectious Diseases at the Tangdu Hospital of the Fourth Military Medical University (Xi’an, China). HTNV infection was confirmed via serological testing of immunoglobulin M (IgM) and IgG in serum specimens. According to the diagnostic criteria from the Prevention and Treatment Strategy of HFRS promulgated by the Ministry of Health, People’s Republic of China, the patients were classified into four clinical types [20]. Mild HFRS was defined as mild renal failure without an obvious oliguric stage and moderate for obvious symptoms of uremia, effusion (bulbar conjunctiva), hemorrhage (skin and mucous membrane), and renal failure with a typical oliguric stage. While patients with severe uremia, effusion (bulbar conjunctiva and either peritoneum or pleura), hemorrhage (skin and mucous membrane), and renal failure with oliguria (urine output, 50–500 mL/day) for ≤ 5 days or anuria (urine output, < 50 mL/day) for ≤ 2 days were defined as severe HFRS, and critical patients were considered as those with ≥ 1 of the following symptoms: refractory shock, visceral hemorrhage, heart failure, pulmonary edema, brain edema, severe secondary infection, and severe renal failure with either oliguria (urine output, 50–500 mL/day) for > 5 days, anuria (urine output, < 50 mL/day) for > 2 days, or a blood urea nitrogen level of > 42.84 mmol/L. In this case, the number of patients with a severity degree of mild, moderate, severe, and critical was 11, 24, 24 and 31, respectively. To ensure the sample size in some statistical analyses, we combined the patients according to the disease severity into mild/moderate and severe/critical groups for comparison. The patients with other kidney diseases, diabetes, cardiovascular diseases, hematological diseases, autoimmune diseases, viral hepatitis, and other liver diseases were excluded in this study. According to the clinical observation, the illness could be divided into five sequential stages: febrile, hypotensive, oliguric, diuretic, and convalescent. In general, samples were collected at 3–6 days for the febrile or hypotension stage, 7–12 days for the oliguric stage, 13–18 days for the diuretic stage and after 18 days for the convalescent stage. The phase within 8 days from the fever onset to the early oliguric stage was typically defined as acute or early phase of the disease. Moreover, eight healthy volunteers were enrolled, showing anti-HTNV negative or no HTNV risk factors, constituting the negative control group.
Peripheral blood samples at different stages of HFRS were collected from each patient during hospitalization. The plasma samples of each patient were collected from heparinized whole blood before the PBMCs isolation. PBMCs were isolated using a standard Ficoll-Hypaque (Sigma-Aldrich, MO) density gradient centrifugation, and these cells were used in the following assays. The values for the clinical parameters, including platelet counts and serum creatinine levels, were regularly recorded during the hospitalization of each patient.
The Taqman RT-PCR assay to determine HTNV viral load was performed as previously described [34]. Briefly, viral RNA was extracted from the plasma samples of HFRS patients and the real-time RT-PCR assay was conducted. The primers and probe were designed based on the sequence alignment of the S segment of the HTNV standard strain 76–118 (NC_005218) and two hantaan virus strains, A16 (AF288646.1) and 84FLi (AY017064) obtained from Shaanxi Province. The copies of HTNV viral RNA load were log-transformed to calculate the viral load.
A total of 281 15-mer peptides over-lapping 11 amino acids spanning the entire 1132 amino acid of Gn/Gc sequence of HTNV 76–118 strain (GenBank accession number: P08668.1), referred to as G1 to G281 according to the N- to C-terminal sequence, were synthesized (CL Bio-scientific, Xi’an, China) and clustered into 28 pools of 10 contiguous peptides (pools 1–28, pool 28 contained 11 peptides). Each pool covered a sequence of 51 amino acids, except pool 28, which covered sequences of 55 amino acids. All peptides were synthesized with greater than 90% purity as determined through high-performance liquid chromatography and resuspended in a sterile DMSO phosphate-buffered saline (PBS) solution at 1 mM.
The identification of the HTNV-Gn/Gc specific T-cell epitopes was performed using the IFN-γ ELISPOT assay (Mabtech, Büro Deutschland, Germany) as previously described [20]. Briefly, the 28 peptide pools were firstly used to stimulate the PBMCs of HFRS patients. Subsequently, the positive response peptide pools were divided into single 15-mer peptides and used as stimuli in a second round of ex vivo ELISPOT assays to identify the single-positive responsed 15-mer peptides. The confirmed positive peptide was subsequently characterized through CD8 depletion of PBMCs using anti-CD8-coated Dynal beads (Invitrogen Dynal AS, Oslo, Norway). CD8+T-cell-depleted PBMCs were employed as effector cells for the recognition of CD4+T-cell responsive 15-mer epitopes. For the frequency detection of the HTNV-Gn/Gc-specific granzyme B-secreting CD4+T cells, the CD8+T cell depleted-PBMCs were further removed CD56+NK cells using APC labeled anti-human CD56 (Miltenyi Biotec GmbH, Germany) together with anti-APC microbeads (Miltenyi Biotec GmbH, Germany) and then used in granzyme B ELISPOT assay (Mabtech, Büro Deutschland, Germany). PBMCs or the isolated cells were placed in the ELISPOT plates at 2 × 105 cells/well and stimulated with peptide pools or single peptide at a final concentration of 10 μM. Cells with phytohemagglutinin (PHA, 10 μg/mL, Sigma-Aldrich, St. Louis, MO) or no peptide stimulation served as positive and background controls, respectively. For quantification of ex vivo responses, the assay was duplicated. An automated ELISPOT reader (Cellular Technology Limited, USA) was used to count the spots. Adjusted spot-forming cells (SFC) after subtracting average negative values expressed as SFC/106 PBMCs. The positive response was defined as at least 50 SFC/106 input cells, exceeding 3 times the background response after subtracting the number of spots in the background controls from those in the stimulated samples. The SFC/106 PBMCs in unstimulated control wells never exceeded 5 spots per well.
Two million freshly isolated PBMCs from each HFRS patient were stimulated for 6h with HTNV-Gn/Gc peptide pools (5 μM for each peptide) in the presence of 1 μg/ml costimulatory molecules anti-CD49d (clone 9F10) and anti-CD28 (clone CD28.2) (Biolegend), as previously described [18]. For the detection of degranulation responses to antigen stimulation, the antibody CD107a-PE (BD Pharmingen) was added to the wells during the stimulation. Cells stimulated with phorbol myristate acetate (PMA, 0.1 μg/ml, Sigma-Aldrich, St. Louis, MO)-ionomycin (0.05 μg/ml, Sigma-Aldrich, St. Louis, MO) or medium alone were used as positive or negative controls, respectively. Brefeldin A (10 μg/ml; Sigma-Aldrich, St. Louis, MO) and monensin (Golgistop, 0.7 μl/ml; BD Biosciences, San Jose, CA) were added to the culture and incubated for 4 h. After washing, the PBMCs were stained with antibodies against the surface markers, in different combinations, including CD3-PE,-APC, CD4-FITC,-PE, CD45RA-APC, CCR7-PerCP-Cy5.5, CD127-FITC, CD27-FITC, CD28-PerCP-Cy5.5, CD38-PerCP-Cy5.5, CD57-PE and CD279 (PD-1)- PerCP-Cy5.5 (BD Pharmingen) for 20 minutes at 4°C, followed by washing, fixation and permeabilization with a permeabilization buffer (BD Biosciences, San Jose, CA). The cells were subsequently stained with antibodies against intracellular markers, including IFN-γ-APC or-PE, TNF-α-PE, IL-4-PE, IL-2-FITC, granzyme B-FITC and perforin-APC (BD Pharmingen), for 15 minutes at room temperature, followed by washing, and acquisition using a FACSCalibur flow cytometer (Becton Dickinson). A total of 300,000 events per sample from the lymphocyte gate were collected for each analysis.
The CD4+T cells isolated with anti-CD4-coated Dynal beads (Invitrogen Dynal AS, Oslo, Norway) from the PBMCs of the HFRS patients were used as effector cells, and the Epstein Barr Virus (EBV) transformed autologous B lymphoblastic cell line (B-LCL) or MHC class Ⅱ partial matched B-LCL of each patient pulsed with HTNV-Gn/Gc peptides were used as target cells. A lactate dehy-drogenase (LDH) releasing Cyto Tox 96 nonradioactive cytotoxicity assay (Promega, Madison, WI, USA) was performed, as described in the manufacturer’s protocol. Briefly, CD4+T cells were cocultured with 5×103 B-LCLs in a 96-well U-bottom cell culture plate at declining effector-to-target cell ratios of 200:1, 100:1, 50:1, 20:1, 10:1 and 5:1. After a 4-hour incubation, a 50μL supernatant of each well was transferred to a 96-well flat-bottom plate. Fifty microliters of the substrate mixture was added to each well, followed by incubation at room temperature for 30 minutes in the dark. Subsequently, 50μL of the stop solution was added to each well, and absorbance was measured at 490 nm with ELISA reader (Bio-Rad, iMark, USA). Cytotoxicity was calculated using the following formula:
%Cytotoxicity=[(Experimental−Effector Spontaneous−Target Spontaneous)/(Target Maximum−Target Spontaneous)]×100
The CFSE-labeled proliferation assay was performed as previously described [18]. Briefly, 2 × 107/ml PBMCs were labeled with 10 μM 5, 6-carboxyfluorescein succinimidyl ester (CFSE, Molecular Probes, OR) at 37°C for 15 min, terminated upon the addition of fetal bovine serum and stimulated with HTNV-Gn/Gc peptide pools (5 μM). Staphylococcal enterotoxin B (SEB, 200 ng/ml, Sigma–Aldrich, MO, USA) or anti-human CD3 (Biolegend) stimulation of PBMCs served as positive controls. After 2 days, 10% exogenous IL-2 was added. After 7 days, the cells were harvested and stained with the CD3-PE or—APC (clone HIT3a), CD8-PerCP-Cy5.5 (clone RPA-T8) and CD4-APC or-PE (clone OKT4) mAbs (BD Pharmingen). Approximately 300,000 cells were acquired using a FACSCalibur (BD Immunocytometry Systems, California).
Flow cytometric analysis was performed immediately with FlowJo version 9.2 (TreeStar). FITC-, PE-, PerCP-Cy5.5- and APC-conjugated mouse IgG1, κ were used as isotype controls of the 4-color staining. Lymphocytes were defined as FSC/SSC, and CD4+T cells were defined as CD3+CD4+ events, displayed on an IFN-γ versus other cytokines dot plot. IFN-γ+ or granzyme B+ cells were further analyzed for expression of T-cell memory markers in a CCR7 versus CD45RA. Quadrant gates were set for T cell memory and differentiation phenotype, where naive cells were defined as CCR7+CD45RA+, central memory cells were defined as CCR7+CD45RA–, translational memory cells were defined as CCR7–CD45RA+, and effector memory cells were defined as CCR7–CD45RA–; early, intermediate and highly differentiation cells were defined as CD27+CD28+, CD27–CD28+ and CD27–CD28–, respectively. The cytokine response was considered positive when the percentage of cytokine was greater than 0.1% after background subtraction.
Statistical analyses and graphing were performed using SPSS 16.0 (SPSS Inc., Chicago, IL, USA) and Prism software, version 5.0 (Graphpad; La Jolla, CA). The frequency of the CD4+ T cells and the cytokines secreted was presented as median and range values. The Wilcoxon rank sum test was used for parameter comparison between the two subject groups. For the comparison of paired design groups, the Wilcoxon's signed rank test was applied. Spearman’s rank was used as the nonparametric test for correlations between continuous variables. A two-tailed P value below 0.05 (P ≤ 0.05) was considered statistically significant.
HTNV 76–118 strain glycoprotein GenBank accession number: P08668.1; Hantaviruses used in this study: HTNV strain 76–118 (NC_005218), HTNV strain A16 (AF288646.1) and HTNV strain 84FLi (AY017064).
|
10.1371/journal.pntd.0005352 | DNA multigene characterization of Fasciola hepatica and Lymnaea neotropica and its fascioliasis transmission capacity in Uruguay, with historical correlation, human report review and infection risk analysis | Fascioliasis is a pathogenic disease transmitted by lymnaeid snails and recently emerging in humans, in part due to effects of climate changes, anthropogenic environment modifications, import/export and movements of livestock. South America is the continent presenting more human fascioliasis hyperendemic areas and the highest prevalences and intensities known. These scenarios appear mainly linked to altitude areas in Andean countries, whereas lowland areas of non-Andean countries, such as Uruguay, only show sporadic human cases or outbreaks. A study including DNA marker sequencing of fasciolids and lymnaeids, an experimental study of the life cycle in Uruguay, and a review of human fascioliasis in Uruguay, are performed.
The characterization of Fasciola hepatica from cattle and horses of Uruguay included the complete sequences of the ribosomal DNA ITS-2 and ITS-1 and mitochondrial DNA cox1 and nad1. ITS-2, ITS-1, partial cox1 and rDNA 16S gene of mtDNA were used for lymnaeids. Results indicated that vectors belong to Lymnaea neotropica instead of to Lymnaea viator, as always reported from Uruguay. The life cycle and transmission features of F. hepatica by L. neotropica of Uruguay were studied under standardized experimental conditions to enable a comparison with the transmission capacity of F. hepatica by Galba truncatula at very high altitude in Bolivia. On this baseline, we reviewed the 95 human fascioliasis cases reported in Uruguay and analyzed the risk of human infection in front of future climate change estimations.
The correlation of fasciolid and lymnaeid haplotypes with historical data on the introduction and spread of livestock into Uruguay allowed to understand the molecular diversity detected. Although Uruguayan L. neotropica is a highly efficient vector, its transmission capacity is markedly lower than that of Bolivian G. truncatula. This allows to understand the transmission and epidemiological differences between Andean highlands and non-Andean lowlands in South America. Despite rainfall increase predictions for Uruguay, nothing suggests a trend towards a worrying human infection scenario as in Andean areas.
| Fascioliasis is a highly pathogenic zoonotic disease emerging in recent decades, in part due to the effects of climate and global changes. South America is the continent presenting more numerous human fascioliasis endemic areas and the highest Fasciola hepatica infection prevalences and intensities known in humans. These serious public health scenarios appear mainly linked to altitude areas in Andean countries, whereas lowland areas of non-Andean countries, such as Uruguay, only show sporadic human cases or outbreaks. To understand this difference, we characterized F. hepatica from cattle and horses and lymnaeids of Uruguay by sequencing of ribosomal DNA ITS-2 and ITS-1 spacers and mitochondrial DNA cox1, nad1 and 16S genes. Results indicate that vectors belong to Lymnaea neotropica instead of to Lymnaea viator, as always reported from Uruguay. Our correlation of fasciolid and lymnaeid haplotypes with historical data on the introduction and spread of livestock species into Uruguay allow to understand the molecular diversity detected. We study the life cycle and transmission features of F. hepatica by L. neotropica of Uruguay under standardized experimental conditions to enable a comparison with the transmission capacity of F. hepatica by Galba truncatula at very high altitude in Bolivia. Results demonstrate that although L. neotropica is a highly efficient vector in the lowlands, its transmission capacity is markedly lower than that of G. truncatula in the highlands. On this baseline, we review the human fascioliasis cases reported in Uruguay and analyze the present and future risk of human infection in front of future climate change estimations.
| The impact of climate change and global change is putting trematodiases in one of the main focuses of infectious disease actuality [1–4]. Among the food-borne trematodiases emphasized in the recent WHO Roadmap for neglected tropical diseases 2020 [5], fascioliasis depicts a specific place due to its worldwide distribution, emergence, and estimated 17 million people infected throughout [6]. The climate change impact on fascioliasis is linked to the high dependence of both fasciolid larval stages and their freshwater lymnaeid snail vectors on climatic and environmental characteristics [2,7,8]. Additionally, fascioliasis emergence appears also related to global change aspects, such as import/export and management of livestock [6], anthropogenic modifications of the environment [9], travelling [10] and changing human diet traditions [11].
Fascioliasis morbidity in humans has been highlighted by the World Health Organization [12]. The acute and long-term chronic phases of this disease show high pathogenicity and immunosuppressive capacity [13–17]. Aspects adding concern are the clinical complexity and severity of symptoms and syndromes, important sequelae and even death [18], pronounced diagnosis difficulties [19] and treatment problems [20].
South America stands out due to the high human prevalences and intensities reported in Andean countries as Bolivia [21–25], Peru [26–28] and Chile [29,30], and the cases from Ecuador [31], Colombia [32], and Venezuela [33]. However, in the non-Andean, lowland countries, human reports only concern sporadic and isolated cases, such as in Brazil [34] and Uruguay [35].
Uruguay has a wide farming and agricultural sector, with 70% of the export trade corresponding to livestock products and subproducts. Fasciola hepatica, locally known as “saguaypé” [36], is distributed throughout the large flat lowlands of the whole country (Fig 1). Cattle and sheep are the most affected, which is related to mixed grazing [37], with high prevalences [37–41] and great impact and economic losses [42]. Horses, sharing the same pastures with cattle and sheep, are the third affected species [43,44]. The liver fluke also infects wild rodents, including the capybara Hydrochoerus hydrochaeris (Caviidae) [45] and the nutria or river rat Myocastor coypus (Myocastoridae) [46]. Fasciolid eggs have also been found in the wild Pampas deer Ozotoceros bezoarticus (Cervidae) [47].
There is a direct effect of altitude on fascioliasis transmission [28]. High altitude lymnaeid vectors produce a higher number of metacercariae throughout a longer cercarial shedding period [48]. This higher transmission capacity is related to the longer life span and post-infection survival of the vector in such altitudes [48,49]. The high fascioliasis transmission in human fascioliasis endemic areas in Andean highlands appears linked to the Galba/Fossaria group species Galba truncatula, a very efficient vector of European origin introduced 500 years ago with the livestock transported by the Spanish "conquistadores" [6,48].
Recent reports indicate that other Galba/Fossaria species may also be involved in human fascioliasis endemic areas, such as Lymnaea neotropica in Peruvian highlands [50], and in extreme aridity-dryness habitats in Argentina [51] together with L. viator (= L. viatrix—for nomenclature see [30]). The later species is also involved in other places of Argentina [52], whereas it was confused with G. truncatula in Chile [30] due to the difficulties in phenotypically differentiating between species of Galba/Fossaria [53].
In Uruguay, only two lymnaeid species have been reported [54: L. viator [55,56] and Pseudosuccinea columella [57]. Pseudosuccinea columella has been detected in 14 of the 19 departments of the country [36], even naturally infected [58]. Although known as an important lymnaeid for the disease transmission to livestock [32], veterinary responsibles have assigned only secondary importance to this species in Uruguay. This is because of its sporadic natural infection linked to its different ecology and pronouncedly lower cercarial shedding when compared to L. viator (a maximum of 10 cercariae/snail in P. columella; between 100 and 200 cercariae/snail in L. viator). Therefore, all efforts have always been conducted to ascertain the epidemiological role of L. viator, both in nature and experimentally in the laboratory [36,59–66].
Three recent findings suggest the need to review the situation in Uruguay: (i) the involvement of Galba/Fossaria species as L. viator and L. neotropica in human fascioliasis endemic areas [50,51]; (ii) their experimentally verified high transmission capacity [67]; and (iii) the potential impact of climate change on fascioliasis and Galba/Fossaria species [1,2,9], given the predictions on the climate change impact on Uruguay, including a rainfall increase [68]. Galba/Fossaria species, including both L. viator and L. neotropica, are amphibious snails which markedly depend on environmental and climatic factors such as temperature, water availability and evapotranspiration [2,7,9]. Thus, the predicted increase of rainfall [68] may a priori offer more possibilities for lymnaeid population growth and consequently the higher number and spread of vectors allow for an increased fascioliasis transmission.
The main aim of the present study is to assess the fascioliasis situation in Uruguay by means of the following aspects:
The aforementioned results furnish the baseline needed for the understanding of the reasons underlying the difference between the high human prevalences and intensities in Andean highlands and the only sporadic human cases or small outbreaks in non-Andean lowlands. This purpose is achieved by means of the appropriate analysis of the following objectives:
The latter objective becomes crucial for a country as Uruguay where cattle raising is the most important activity of the primary sector; cattle are kept on more than 83% of farms; on more than half of them beef cattle are the main source of income. The most important beef breed is Hereford, with 76.0% of the herd. There are more than 6,500 farms specialising in dairying, with more than 750,000 animals, more than 90% of which are Holsteins [69]. Moreover, Hereford and Holstein breeds appear to be the most affected by fascioliasis in Uruguay, with prevalences of 56% (95% confidence interval: 51–61%) and 68% (95% ID: 64–73%), respectively, pronouncedly higher than the prevalences in other breeds [41].
Fasciolid materials were obtained from naturally infected animal hosts from Uruguay. A total of 46 fasciolid adult worms were obtained from livers of two Hereford, two Holando and two Aberdeen Angus breed cattle from three zones in the Salto department. Additionally, fasciolid eggs were obtained from a biliary filtrate of three horses from a slaughterhouse in Montevideo department (Fig 1). According to available facilities in obtaining materials, the collecting strategy was to sample materials from the western part of the country through which border the first fasciolids should have been introduced into Uruguay for the first time in the past, taking into account the original spread of livestock with the early Spanish conquerors. In Uruguay, fasciolids infect both cattle and sheep in the same places, because these two species are kept mixed in the grazing pastures and areas. Taking into account that F. hepatica infects cattle and sheep similarly (i.e., these two species do not select different fasciolid strains), the fasciolid material was obtained from cattle because this is the livestock regularly killed in the slaughterhouses, given that cattle raising is the most important activity of the primary sector in Uruguay. The material from horses was, however, not obtained in the same western areas, because horses also share the same pastures and should logically be infected by the same fasciolid haplotypes. Therefore, horses were selected from the Montevideo department, where these animals are managed in a somehow different way because of the neighbourhood of the big city.
Fasciolids from the aforementioned cattle and horses were used for DNA marker sequencing and a F. hepatica isolate obtained in a 6-year-old Hereford cattle female from Salto was used for the experimental infections of lymnaeid snails. Uruguayan fasciolid materials have been deposited in the Museu Valencià d’Història Natural, Alginet, Spain, under the code MVHN-241016MD01.
Lymnaeid snail materials originated from three different populations in Uruguay, corresponding to the departments of Montevideo (6 specimens), Paysandú (4 specimens) and Canelones (10 specimens) (Fig 1) and were used for their molecular characterization by DNA marker sequencing. Given the very low fasciolid larval stage prevalences in lymnaeid vectors in nature, a broader snail survey was not within the goals of this assessment of F. hepatica in Uruguay.
Moreover, lymnaeids collected in Canelones and shortly maintained in the laboratory of the DILAVE "Miguel C. Rubino", were finally transported and cultured in the Laboratory of the Valencia centre for a standardized experimental study. A total of 50 laboratory-borne specimens were used for the infection experiments with the aforementioned Uruguayan F. hepatica isolate. All lymnaeid specimens collected and used were preliminarily classified by shell morphology as Lymnaea viator, following the literature on lymnaeids in Uruguay (Table 1). Uruguayan lymnaeid materials have been deposited in the Museu Valencià d’Història Natural, Alginet, Spain, under the code MVHN-241016MD02.
The living standard of Uruguay is closely related to earnings from pastoral and agricultural exports of beef and wool. Extensive cattle and sheep rearing is the main activity of Uruguay, where half the grassland is in estancias (usually large, simple buildings with thick walls, of a typical Spanish colonial style, with a lot of wrought iron) exceeding 2,000 acres. More than 13,500,000 ha are under permanent pasture, almost 83% of the agricultural area [69]. Millions of sheep and cattle are raised in the country. The preference for pasture over cropland is due to the excellence of the grasslands and the variable rainfall that makes grain production unreliable. The ratio between sheep and cattle production shifts with demand [70]. The predominant sheep breeds in Uruguay are Corriedale, Merino and Polwarth, which represent 60%, 20% and 10% of the national sheep flock, respectively. These breeds generate income from the sale of wool and sheep meat (surplus offspring and cast for age animals). Traditionally, wool has been the main product of the system. However, in recent years, the importance of sheep meat (lambs and mutton) has increased significantly [71].
Living Galba/Fossaria lymnaeids preliminarily classified as L. viator, collected in Canelones and shortly maintained in the DILAVE laboratory, were transported under isothermal conditions to the laboratory of Valencia. Transfer to Valencia was needed to allow for a standardized laboratory adaptation and subsequent experimental follow-up of the life cycle and transmission of Uruguayan flukes by Uruguayan lymnaeids under abiotic conditions enabling significant comparisons with endemic areas of other countries. The possible natural infection by fasciolids was always individually verified prior to the launch of laboratory cultures. This was performed by keeping each lymnaeid specimen isolated in a Petri dish containing a small amount of natural water. After 24 h, the presence or absence of motionless metacercarial cysts or moving cercariae was verified in each Petri dish.
Afterwards, non-infected lymnaeids were arranged in standard breeding containers containing 2000 ml fresh water, to assure pure specific cultures. Finally, snails were adapted to and maintained under experimentally controlled conditions of 20°C, 90% relative humidity and a 12 h/12 h light/darkness photoperiod in precision climatic chambers (Heraeus-Vötsch VB-0714 and HPS-500). The water was changed weekly and lettuce added ad libitum.
Eggs of F. hepatica from a 6-year-old bovine female from Salto were maintained in fresh water under complete darkness at 4°C until starting the embryogenic process. Embryogenesis was followed at 20°C at intervals of four days by counting eggs presenting an incipient morula, eggs including an advanced morula, eggs with outlined miracidium, and fully embryonated eggs containing a developed miracidium. Developed miracidia were forced to hatch by putting fully embryonated eggs under light and used for the experimental infection of snails [48].
Snails collected in the Uruguayan department of Canelones, shortly maintained in the laboratory of the DILAVE "Miguel C. Rubino", and finally transported and kept in the Laboratory of the Valencia centre were used for the experiments. Only laboratory-borne specimens were used. Snails of different size within the length range of 4.7–7.6 mm (mean 5.74 mm) were used to assess infection susceptibility. A total of 50 lymnaeid specimens were infected monomiracidially by exposing each snail to 1 miracidium for 4 hours in a small Petri dish containing 2 ml of fresh water. The disappearance of the miracidium was taken as verification of its successful penetration into the snail.
Snails were afterwards returned to the same standard conditions in the climatic chamber (2000 ml containers, 20°C, 90% relative humidity (r.h.), 12 h/12 h light/darkness, dry lettuce ad libitum) until day 30 post-infection, in which they were again isolated in Petri dishes to allow daily monitoring of cercarial shedding by individual snails. Lettuce was provided ad libitum to each snail in a Petri dish during both shedding and post-shedding periods until death of the snail. The chronobiology of the cercarial shedding was followed by daily counting of metacercariae in each Petri dish [48]. Furthermore, the strains of both F. hepatica and the lymnaeid used for the experimental assay were characterized by the sequencing of the aforementioned DNA markers. For that purpose, fasciolid metacercariae experimentally obtained and snails fixed after verification of the end of the cercarial shedding period were used (Table 1).
Life cycle aspects analyzed and respective methods used are in agreement with the standards applied for such studies in Fasciola. Following this standardized way allows for significant comparisons of the transmission characteristics in different endemic areas [48].
Ethical approval for the animal work was provided by the Ethics Committee for Animal Experimentation and Welfare of the University of Valencia, Spain (A1263915389140). Additionally, the División de Laboratorios Veterinarios (DILAVE), Montevideo, belongs to the corresponding national ministry (Ministerio de Ganadería, Agricultura y Pesca—MGAP) counting on its own ethics committee (Comité de Etica para Uso de Animales de Experimentación—CEUA) and its animal work is authorized by the Comisión Nacional de Experimentación Animal of Uruguay. Animal ethic guidelines regarding animal care strictly followed the institution’s guidelines based on Directive 2010/63/EU. Informed written consent was received from all animal owners (farm: El Solar, Salto; owners: Sucesores de Alfredo Sanchis; official registry No. at the MGAP Ministry: 150 606 365).
A total of 9 different marker sequences were obtained from the fasciolids. Nucleotide length of the sequences, their GC/AT content and reference codes are noted in Table 1.
Two haplotypes of the complete intergenic rDNA ITS1-5.8S-ITS2 region were detected in the fasciolids infecting cattle and also in horses in Uruguay. ITS-1 proved to have the same sequence in all specimens studied, corresponding to the haplotype A of this spacer (Table 1). ITS-2 showed two sequences in the Uruguayan fasciolids: haplotypes 1 and 2 GC (Table 1). Only one mutation in position 287 of the ITS-2 alignment allows the differentiation between both haplotypes: “C” in haplotype FhITS-2 H1 and “T” in FhITS-2 H2.
The mtDNA cox1 provided three different sequences with the same length. Their alignment showed 6 variable positions (all of them singleton sites). These sequences proved to enter in the intraspecific variability of the 69 cox1 haplotypes so far known in F. hepatica, corresponding to the haplotypes Fhcox1-5, Fhcox1-16 and Fhcox1-42 (Fig 2). The three haplotypes were found in cattle, whereas only Fhcox1-42 was found in horses (Table 1). The COX1 protein was 510 aa long, with start/stop codons of ATG/TAG, identical in all specimens analyzed, and corresponding to the haplotype FhCOX1-1 (Fig 2).
The mtDNA nad1 sequence provided three different haplotypes with the same length. Their alignment showed 3 variable positions (all of them singleton sites). These sequences also proved to enter in the intraspecific variability of the 51 nad1 haplotypes so far known in F. hepatica, corresponding to the haplotypes Fhnad1-2, Fhnad1-12 and Fhnad1-14 (Fig 3). The three haplotypes were found in cattle, whereas only Fhnad1-12 was found in horses (Table 1). The NAD1 protein showed only one 300-aa-long haplotype with start/stop codons of GTG/TAG in all specimens analyzed, corresponding to the haplotype FhNAD1-1 (Fig 3).
A total of 5 different marker sequences were obtained from the lymnaeids. Nucleotide length of the sequences, their GC/AT content and reference codes are noted in Table 1.
The ITS-2 sequences of lymnaeids from the three localities in Uruguay were identical. This unique sequence showed no one nucleotide difference with the ITS-2 haplotype H1 of the Galba/Fossaria species L. neotropica (Table 1). Similarly, the ITS-1 sequences were also identical in the three localities and without any difference when compared to the ITS-1 haplotype HA of L. neotropica (Table 1), which differs by two insertions in the “poli-A” region at the 3' end from haplotype HB (positions 512 to 529 including 16 or 18 consecutive “A” in L.neo-HA and L.neo-HB, respectively).
The 16S rRNA gene of the mtDNA provided only one haplotype in the three localities. This partial sequence presented a biased AT content, and proved to be identical to the provisional haplotype L.neo-16S HA (Table 1).
The partial sequence of the mtDNA cox1 gene of lymnaeids from the three localities in Uruguay showed two haplotypes. The one found in Canelones proved to be identical to the haplotype L.neo-cox1 Ha from the type locality of L. neotropica. The second haplotype, present in the localities of Montevideo and Paysandu, proved to be identical to the provisional haplotype L.neo-cox1 He (Table 1). When comparing these two cox1 sequences from Uruguay with the five cox1 haplotypes of L. neotropica known so far, the resulting 672 bp-long alignment showed 76 variable positions, including two parsimony informative and 74 singleton sites. Nucleotide and amino acid differences are listed in Fig 4.
Experimental life cycle studies were undertaken with the Uruguayan F. hepatica combined haplotype FhITS2-1, FhITS1-A, Fhcox1-5, Fhnad1-2 found in Hereford cattle from Salto, and the L. neotropica combined haplotype L.neo-ITS2-1, L.neo-ITS1-A, L.neo-16S-A, L.neo-cox1-a collected in the Canelones department (Fig 5).
Results of embryogenesis inside the egg, lymnaeid snail infection, intramolluscan parasite larval development and influences of the latter on snail survival are noted in Table 2. The use of identical experimental procedures and standardized abiotic factors allow for a significant comparison with the same data experimentally obtained with F. hepatica and G. truncatula from the high altitude pattern of the life cycle and disease transmission of cattle and sheep isolates of the liver fluke in the Northern Bolivian Altiplano, the human hyperendemic area with the highest human prevalences and intensities known (Table 2).
The Uruguayan liver fluke isolate proved to follow a pronouncedly faster embryogenesis. The first developed miracidium appears already in day 15, with the maximum percentage of eggs including fully developed miracidia in day 18, whereas 46 and 58 days were needed by cattle and sheep isolates from the Bolivian Altiplano, respectively. Such a development speed is three times faster in the lowlands of Uruguay, even in a surprisingly high percentage of eggs of 88.2% (whereas only in 24.9% and 16.4% for the two Bolivian isolates, respectively) (Table 2).
The high snail infectivity rate (74.5%) of the Uruguayan isolate is worth mentioning, although similar to the Bolivian sheep isolate. The prepatent period (days elapsed from infection day up to the first day of cercarial shedding) in the Uruguayan isolate is markedly similar to that of the two Bolivian isolates. However, pronounced differences appear in the shedding period (total number of days in which the snail was shedding cercariae), as well as in the total number of cercariae (and subsequent metacercariae) produced by each infected lymnaeid. The shedding period in the Uruguayan isolate proved to be very short, of only 1–19 days (mean 9.6 days), despite of which the number of cercariae per snail was relatively high, of 4–1186 cercariae/snail (mean 269.2). These features are, however, far from the ones characterizing the Altiplanic liver fluke isolates, in which the shedding period is very long (averages higher than 70 days in both isolates) and the number of cercariae per snail is very high (averages higher than 445 cercariae/snail in both isolates) (Table 2).
The geographical isolate did not seem to influence lymnaeid survival during the prepatent period, results obtained with the Uruguayan fluke being very similar to those in the two Bolivian isolates. Nevertheless, (i) the snail survival after the end of the shedding period, (ii) the postinfection longevity in shedding snails, and (iii) the longevity in non-infected snails, were all three pronouncedly shorter when dealing with the Uruguayan isolate than with the two Bolivian isolates (Table 2).
Despite the fast and short intramolluscan larval development, the Uruguayan liver fluke isolate proved to be able to reach a marked extent of redial infection and massive presence of rediae in the local Uruguayan lymnaeids (Fig 6A and 6B).
The chronobiological pattern of cercarial emergence in the cattle isolate of F. hepatica is shown in (Fig 7A and 7C).
When the shedding period is analyzed from the day of the emergence of the first cercaria by each snail (Fig 7A), the shedding process appears as an irregular succession of waves. After four days of an initial shedding of a reduced daily number of cercariae, a slow decrease of that number is observed. The higher acrophases take place at the end of the first week and during the second week.
When the shedding period is analyzed from the day of the miracidial infection (Fig 7C), most of the cercariae are shed between days 52 and 64 post-infection (p.i.). The days 64 and 73 p.i., in which all snails failed to shed any cercaria, suggest an intramolluscan larval development including up to a maximum of three redial generations. It is mainly for the first generation to produce and shed most of the cercariae.
When comparing this chronobiological pattern of cercarial emergence in F. hepatica/lymnaeid snail from Uruguay (Fig 7A and 7C) with the chronobiological pattern in F. hepatica/G. truncatula from the Northern Bolivian Altiplano, performed under identical procedures and standardized experimental abiotic factors (Fig 7B and 7D), four main differences should be highlighted:
Fasciolids from Uruguay molecularly prove to belong to widespread strains of F. hepatica, fitting within the intraspecific variability of this fasciolid species in Europe and Latin America. Indeed, F. hepatica was introduced into South America throughout a process which started 500 years ago at the time of the first Spanish colonizers, who were transporting livestock in their ships [6]. From the evolutionary point of view, such a period is very short for a parasite. Manter's parasitophylogenetic rule, about the slower evolution of parasites when compared to that of the hosts, should be considered here [80]. Moreover, the livestock host species in Latin America at present are the same than in its original spreading area in Europe 500 years ago, which means that the microhabitat of F. hepatica has not changed despite its hosts having been moved from one continent to another. Additionally, the evolutionary rates of the four DNA markers used are too low [81] as to expect mutations appearing by isolation in the Americas and differentiating American fasciolids from those of the Old World [6].
Short information may be inferred from the only one ITS-1 and two ITS-2 haplotypes of these evolutionarily conserved rDNA spacers. In fact, the worldwide spread of F. hepatica occurred only during the last 12,000–10,000 years, from the moment of the domestication of livestock in the Fertile Crescent in the Near East and in Old Egypt, when humans began to expand livestock species throughout. Similarly occurred with F. gigantica, although restricted to Africa and Asia where its specific Radix lymnaeid species are present. The absence of Radix in the New World (only isolated populations of R. auricularia introduced into North America from Europe) explain why only F. hepatica is present in the Americas [6]. A period of 12,000–10,000 years is too short for ITSs to give rise to mutations, given their evolutionary rate [81]. Thus, in regions where only F. hepatica is present, and consequently there is no possibility for hybridization with F. gigantica, the same only FhITS-1 HA is known [6]. Regarding ITS-2, FhITS-2 H1 and FhITS-2 H2 found in Uruguay also correspond to the two haplotypes known in areas presenting genetically pure F. hepatica [6]. FhITS-2 H1 is the most widely distributed, whereas FhITS-2 H2 has interestingly been described from Spain and Andorra [6] and was already previously reported from Uruguay [82].
More information can be inferred from the three cox1 and three nad1 haplotypes of the faster evolving mtDNA [81]. Among cox1, Fhcox1-5 and Fhcox1-16 are widely dispersed in South America, but Fhcox1-42 has only been found in Bolivia [6]. A similar picture is provided by nad1. Fhnad1-2 and Fhnad1-14 are distributed throughout North and South America, whereas Fhnad1-12 has only been detected in Peru, Bolivia and Argentina [6]. The only two other complete mtDNA gene sequences of F. hepatica from Salt Lake City, Utah, USA [83] and the Geelon strain in Australia [84] are different from the F. hepatica haplotype group of the Iberian Peninsula and Latin America, at the level of both cox1 (Fig 2) and nad1 (Fig 3).
The single COX1 and NAD1 protein haplotypes found in Uruguay are the most abundant, both widely distributed in different countries and different host species [6].
The detection of Fhcox1-42 and Fhnad1-12 in both cattle and horses indicate infection from same sources, e.g. in Uruguay horses may become infected when grazing in pastures used for cattle and sheep [44].
The sequences of the four DNA markers used unambiguously demonstrate that the Galba/Fossaria lymnaeids collected in the three Uruguayan localities belong to the species L. neotropica.
The ITS-2 and ITS-1 found in Uruguay are identical to L.neo-ITS-2 H1 and L.neo-ITS-1 HA in Perú and Argentina (74,50,51,75), the latter differing by only two "A" insertions from L.neo-ITS-1 HB from Argentina [51]. Similarly, the 16S haplotype L.neo-16S HA found in Uruguay has also been reported from Perú and Argentina [50,51]. The first cox1 haplotype L.neo-cox1 Ha found in Canelones was already known from the type locality of L. neotropica and another area in Peru [50,74]. The second L.neo-cox1 He found in Montevideo and Paysandu was known only from Argentina [51].
So far, the only Galba/Fossaria species reported in Uruguay was L. viator [36,55,56,59–66]. Consequently, L. neotropica becomes a new species for the Uruguayan fauna. Its finding in areas located far away one another, indicate that this species should be widely distributed throughout the country. The question posed now is whether both L. neotropica and L. viator coexist in Uruguay, or there is simply only L. neotropica which has always been confused with L. viator.
Indeed, DNA sequencing of lymnaeids started at the end of the last century, including markers of mtDNA [85] and nuclear rDNA [86]. Their progressive use highlighted the problems in specimen classification and species differentiation by traditional malacological approaches [48,72,73]. Interspecific differentiation in Galba/Fossaria became the main focus, due to their importance in the transmission of F. hepatica. The description of the new species L. neotropica and its molecular differentiation from L. viator and G. truncatula was an important step forward [74]. The molecular demonstration that the hitherto overlooked species L. schirazensis, without fascioliasis transmission capacity, had always been confused with G. truncatula and other Galba/Fossaria vector species, illustrated up to which level there was a chaotic systematic situation [53], as verified in Venezuela [33]. Similarly, DNA sequencing proved the presence of G. truncatula, L. neotropica and L. schirazensis in the human fascioliasis hyperendemic area of Cajamarca, in Peru, where only L. viator had been involved [50].
All these major advances were posterior to all studies on lymnaeids in Uruguay. This suggests that we are only dealing with a classification confusion between L. neotropica on one side and L. viator and G. truncatula on the other side, similarly as happened in Peru, Venezuela, Argentina, and also Chile [30]. So, L. neotropica should probably be the only Galba/Fossaria species distributed throughout Uruguay. However, it should not be overlooked that L. neotropica and L. viator may coexist in the same area, as in Mendoza [87,88] and Catamarca [51], both in Argentina. Studies in other areas of Uruguay are needed. For this purpose, it shall be considered that ITS-1 and 16S showed the highest and lowest resolution for interspecific differentiation, respectively, whereas cox1 was the best marker and ITS-1 the worst for intraspecific analyses [51].
Regarding genus ascription of L. neotropica and L. viator, their last molecular comparison, both one another and inside the Galba/Fossaria group, and maximum support values obtained for the internal branching nodes in the phylogenetic analysis of the species of Galba/Fossaria, demonstrated that these Neotropical species do not belong to the genus Galba defined by its Palaearctic type species G. truncatula [51]. Until sequence data from the very large number of Galba/Fossaria species known from the Nearctic region [89,90] is obtained, caution recommends to taxonomically keep L. neotropica and L. viator in the genus Lymnaea sensu lato.
The sequences of L. neotropica from Uruguay and those from Peru [50,74], Venezuela [33] and Argentina [51,75,88], suggest that the spread of this lymnaeid throughout the Neotropical region should have occurred very recently, passively transported with livestock exchange, in a similar way as other Galba/Fossaria species spread throughout even different continents, as G. truncatula [6] and L. schirazensis [53].
Finally, it should be highlighted that all the haplotypes of the four DNA markers found in L. neotropica from Uruguay have also been found in two human fascioliasis endemic areas, such as Cajamarca in Peru [50] and Catamarca in Argentina [51].
Uruguayan F. hepatica and L. neotropica used for the experiments were selected to assess the disease transmission capacity of the most common and widely dispersed strains of both fluke and vector species.
The embryonation time found in the Uruguayan couple proves to be very short. It fits within the fastest inside the range known when tested at 20°C [111–113]. In F. hepatica, the development of the miracidium inside egg is arrested below 9°C and above 37°C and has a duration between 9 and 161 days depending upon the temperature, the range 20–25°C offering the optimum for the hatching of a higher number of miracidia [114–117].
The detected infection percentages and prepatent period in monomiracidial infections may be considered as normal at 20°C when compared to similar studies carried out with F. hepatica isolates and G. truncatula specimens from European areas [111,114,118–122]. The prepatent period found in the Uruguayan couple also fits in the known range (43.1 ± 58.2 days) for European F. hepatica/G. truncatula in the nature [123].
The shedding period in Uruguayan F. hepatica/L. neotropica is very short. In European F. hepatica experimentally infecting G. truncatula snails of the same size as ours under the same constant conditions of 20°C and 12 h/12 h photoperiod, the patent period lasted only 46 ± 27.6 days [124]. Similarly, results obtained in nature show that the patent period in Europe ranges between 5.0 and 9.3 days in the winter generation and 18.3–40.3 days in the summer generation [123]. The pronounced differences of the very short shedding period in the couple from the Uruguayan lowlands when compared to the very long one in the F. hepatica/G. truncatula couple from the highlands of the Bolivian Altiplano (Table 2) [48] should be highlighted.
The mean number of cercariae shed per individual lymnaeid in the Uruguayan couple is close to the mean of 238.5 cercariae/snail found in the European F. hepatica/G. truncatula model under the same experimental conditions [124]. A lower number of 114.9 ± 80.3 cercariae per monomiracidially infected snail were obtained with the same European couple. These experimental assays showed that the duration of the shedding and the number of cercariae were independent of the number of miracidia used for the infection of each individual lymnaeid. However, single-miracidium infections were most effective because of the much higher snail survival rate, despite the mean number of cercariae shed being the same as in multimiracidial infections [125]. However, the most important is the marked difference when compared to the pronouncedly higher mean cercarial numbers in the F. hepatica/G. truncatula couple from the Bolivian Altiplano (Table 2) [48].
Differences in survival of different geographical strains of the same Galba/Fossaria species to F. hepatica infection have already been described [126]. The postinfection longevity in shedding L. neotropica from Uruguay is only slightly shorter than the 70 days p.i. usually observed in European G. truncatula [127–131], with a maximum of 16 weeks described once [128,129], and far from that known in other American lymnaeids such as 119 days p.i. for L. viator [132] and 113.4 days p.i. for L. bulimoides [133]. A fast development and extensive massive infection of the larval stages in L. neotropica (Fig 6) may be the responsible for a quick snail mortality, similarly as in European G. truncatula. It was found that of 102 snails shedding on the first day, the number drastically reduced to only 56 on the second day and subsequently decreased on day 76 to four snails [124]. Regarding this aspect, the pronounced difference when compared with the capacity of G. truncatula from high altitude endemic areas to survive up to more than 4 months after the end of the shedding period (Table 2) should be highlighted [48].
The cercarial shedding pattern detected in the Uruguayan couple does not disagree with the patterns observed by other authors on the F. hepatica/G. truncatula model [124,134–136]. When considering the shorter shedding period in the Uruguayan couple, the three shedding acrophases it shows (Fig 7A and 7C) fit well in the 4–5 shedding waves showed by the majority of G. truncatula in an experiment under constant conditions [124]. The delayed acrophases in F. hepatica/L. neotropica from Uruguay also agree with the pattern found in the European model. Here again, the pronounced differences of the couple from the Uruguayan lowlands with the F. hepatica/G. truncatula couple from the highlands of Bolivia [48] should be highlighted: regarding (i) the daily number of cercariae/snail, (ii) length of the shedding period, (iii) daily number of cercariae/snail, and (iv) cercarial production by the different redial generations. The longer post-infection longevity of G. truncatula under high altitude conditions [48,49] and the higher pathogenicity induced by the fast development and massive infection by F. hepatica in L. neotropica from Uruguayan lowlands underlie the aforementioned differences.
Summing up, our experimental results demonstrate that the F. hepatica/L. neotropica couple from Uruguayan lowlands is markedly less efficient for the disease transmission than the F. hepatica/G. truncatula couple from the Andean highlands, although somewhat more efficient than the F. hepatica/G. truncatula couple from European lowlands. The latter result agrees with other experimental data indicating that L. neotropica and L. viator from Argentina are better hosts than European (French) G. truncatula in both allopatric and sympatric infections by Argentinian and French isolates of F. hepatica [67]. The efficiency results of both, our present study of the Uruguayan couple and the one of the Argentinian and French mixed couples [67], may be interpreted taking into account that allopatric combinations of F. hepatica and lymnaeid species were proved to be more efficient than sympatric ones [137]. This capacity may be considered a useful strategy of the liver fluke for the colonization of new areas [6]. Indeed, the aforementioned historical analysis suggests that L. neotropica did not colonize Uruguay until at a maximum the first part of the 17th century, and consequently, around only 400 years ago. This is a very short period for the parasite from the evolutionary point of view. The aforementioned (i) Manter's rule [80], (ii) similarity of ancestral European livestock and present Uruguayan hosts, and (iii) low evolutionary rates of DNA markers used [81], should again be considered in this assumption. So, the term "allopatric" should be applied with caution here.
Only a total of 95 human fascioliasis cases have been reported in Uruguay. The first report of a human subject infected by F. hepatica in Uruguay was in 1909 and concerned a 49-year-old women suffering from right hypocondrium pain and in whom a fluke was unexpectedly found near the main biliary duct during surgery [138]. Twenty years later, three liver fluke specimens were detected during a gall bladder surgical intervention of another patient [139]. Between 1935 and 1950, isolated human cases were reported after egg detection in stool samples and/or duodenal exploration [140–142].
A familiar outbreak involving 11 members, probably linked to contaminated watercress consumption, was one year later diagnosed in the hospital of Paysandu. An exhaustive clinical and epidemiological study was performed [143].
The flooding events during the 1954/55 and 1958/59 periods were suggested to have spread fascioliasis and therefore related to three subsequently reported human epidemics [35]. Thus, a total of 31 cases were reported from the Florida department in 1960 [144], 20 additional cases were compiled from seven different departments from the country inland, the majority from Florida, Canelones and San José in 1978 [145], and finally another 16 cases diagnosed in an hospital during the 1953–1977 period were anatomo-pathologically described in 1979 [146].
Interestingly, among a review of patients infected by the HIV virus during the 1983–1988 period, F. hepatica eggs were found in the stools of a patient affected by AIDS and dying after 35 days hospitalization [147]. Only two additional cases were detected among a total of 951 samples during a wide serological and coprological survey performed in several localities of the departments of Artigas, Rivera, Florida and Salto throughout 1991 and 1993 (Lopez Lemes et al., 1992 and 1993, in [35]). Another infected patient with fever and eosinophilia was noted to be diagnosed both coprologically and immunologically in 1990 by E. Zanetta, in the same article [35].
Another familiar outbreak involving only three subjects and linked to the consumption of wild watercress, was reported (Lopez Lemes et al., 1994, in [35]). Finally, the last report was in 2003 about two female and one male clinical cases presenting with right hypocondrium pain, eosinophilia and history of watercress consumption [148].
The aforementioned review suggest a sporadic and isolated human infection risk throughout a wide hyperendemic animal fascioliasis zone in Uruguay, according to the epidemiological classification of WHO [149]. Several aspects merit, however, to be considered.
Despite the distribution of the liver fluke covering the whole country [37], the human infection risk does not appear to be homogeneous, i.e. it seems to be higher in given departments, as in Florida department [144,145]. Moreover, the unexpected finding of human infection in surgical interventions [138,139], HIV-infected patient survey [147], and wide surveys [35], suggest underestimation of the real occurrence of human infection, similarly as in Argentina [150]. The familiar outbreaks also remember the human fascioliasis situation in the lowlands of Argentina. However, whereas in the physiographically highly heterogeneous Argentina available data suggested human endemic local areas which have been finally described [51], the physiographic uniformity of Uruguay does not indicate such a scenario.
Nevertheless, the three increases of patient numbers after the flooding events of the 1954/55 and 1958/59 periods [144–146] should be considered by the public health responsibles, given IPCC (Intergovernmental Panel on Climate Change) predictions of a rainfall increase within the future climate change impact affecting Uruguay [68]. Anyway, there is nothing indicating a trend towards a worrying human infection scenario such as in Andean areas. Neither the ecological characteristics and preferences of the main vector L. neotropica [64], nor its transmission capacity verified in the present study, suggest such a future possibility. This does not mean, however, that the very fast larval development of F. hepatica and short shedding of high numbers of cercariae furnished by L. neotropica may take advantage of occasional, more or less prolonged flooding events to increase offspring, population densities and subsequently spread, thus enabling for an increase of familiar outbreaks or short transient epidemic situations.
The comparison of the transmission characteristics and capacities in the F. hepatica/G. truncatula couple from Bolivia with the F. hepatica/L. neotropica couple from Uruguay allow for the understanding of the high transmission patterns and endemicity characteristics of human fascioliasis in Andean highlands, opposite to the rare/sporadic/low human infection in animal endemic areas in Uruguayan lowlands. Consequently, it may be concluded that L. neotropica may be responsible for a human endemic area only under special circumstances, as in isolated foci in aridity/dryness areas described in Argentina [51].
The transmission characteristics and capacities of the Uruguayan F. hepatica/L. neotropica couple are a priori better for a seasonal transmission of the disease, depending on local climatic features. Uruguay has a subtropical to temperate climate with very marked seasonal fluctuations [69]. The climate is sub-humid, because potential evapotranspiration in summer is greater than precipitation. Although rainfall is distributed throughout the year, great variations occur between years. The highest precipitation occurs, in general, in summer and autumn. In the first season, precipitation is very irregular, with summers lacking precipitation and others with more than 600 mm of rain. In the second season, precipitation shows minor variability. Although precipitation has a somewhat smaller volume in winter than in other seasons, there is no marked rainy season. The great rainfall variation, both in regularity and intensity, should be highlighted because it leads to droughts and floods in different seasons of the year. Mean temperatures of the coldest month (July) are 10.8°C and 13.0°C, and the warmest month averages (January) are 22.6°C and 25.1°C for the Southern and Northern regions, respectively [69].
In Uruguay, field studies indicated that the fluke life cycle is maintained throughout the whole year, although it considerably slows down in winter [35]. Lymnaeids naturally infected in autumn-winter, with mean maximum temperatures lower than 20°C and mean minimum ones below 10°C, showed a 4–8 month cercarial shedding, whereas this was reduced to only 37 days in summer [63]. In spring, shedding periods gradually shorten, which together with an increase of lymnaeid population densities at the end of spring, gives rise to an increase of the number of infected animals at the end of spring and summer [38]. In summer, temperatures are ideal for F. hepatica development, but the insufficient rainfall and high evapotranspiration resulting in a humidity shortage become important limiting factors for the lymnaeids [40]. The long survival and infectivity of metacercariae [151] also add to understand the human infection risk the year long, despite it being higher in spring.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.